[
  {
    "state": "California",
    "school": "California Western School of Law",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "None — independent law school, no parent institution",
    "specs": "",
    "notes": "Re-verified Sept 2026, first sourcing for this row. Standalone is CONFIRMED on three independent footings: the school describes itself as an independent, nonprofit law school and repeats that independence as a selling point ('we operate independently, so we're able to innovate and adapt our curriculum more quickly'); its regional accreditor WSCUC lists its Sponsorship as 'No related entity'; and it holds its own separate WSCUC accreditation (candidacy 2019, accredited 2023) rather than sitting under a parent's. Founded 1924 as Balboa Law College, ABA-accredited since 1962, AALS member since 1967, roughly 670-830 JD students, San Diego's oldest law school. Historical note that can mislead: California Western is one of two successor organisations to the former California Western University, the other being Alliant International University — but that is a shared ancestor, not a current parent. ONE LEAD INVESTIGATED AND RULED OUT: the WSCUC accreditation record shows two joint degrees run WITH San Diego State University — a JD/MBA (2002) and a JD/MSW (1997) — and SDSU does have real research computing (Computational Science Research Center clusters including a dual P100 GPU server, an NSF-funded node in the NRP Kubernetes cluster, and the NSF TIDE project, a $991,749 CSU-wide GPU initiative SDSU leads with the San Diego Supercomputer Center). That looks like the Albany Law situation but is not. The Albany affiliation explicitly funds joint FACULTY research through a Collaborative Venture Fund; these SDSU arrangements are STUDENT degree programs, and nothing found extends them to California Western faculty or to compute. Recording it here so the next person doesn't have to re-investigate the same lead. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.cwsl.edu/about/index.html",
      "https://www.wscuc.org/institutions/california-western-school-of-law/",
      "https://www.wscuc.org/institutions/california-western-school-of-law/?print=pdf",
      "https://www.cwsl.edu/academics/juris_doctor_jd_program/index.html",
      "https://en.wikipedia.org/wiki/California_Western_School_of_Law"
    ]
  },
  {
    "state": "California",
    "school": "Southwestern Law School",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "None — independent law school, no parent institution",
    "specs": "",
    "notes": "Re-verified Sept 2026, first sourcing for this row. Standalone CONFIRMED: Southwestern is described by LSAC and by itself as an independent, nonprofit, nonsectarian institution, founded 1911, roughly 950-1,000 students, Los Angeles' second-oldest law school, campus centred on the Bullocks Wilshire Art Deco landmark. It briefly held a university charter in 1913 when it expanded into other disciplines, but that is historical and it operates today as a law school alone. Not to be confused with Southwestern University, the liberal arts college in Georgetown, Texas. LEADS INVESTIGATED AND RULED OUT: Southwestern has an unusually long list of institutional partnerships — 3+3 BA/JD programs with California State University Northridge and CSU Dominguez Hills, a JD/MA in Negotiation with CSUDH, JD/MBA programs with the Drucker Graduate School of Management, a public policy certificate with the Pardee RAND Graduate School, and a bioscience certificate with the Keck Graduate Institute. Every one of these is a STUDENT degree or certificate pathway. None is a faculty research affiliation, and none was found to extend compute to Southwestern faculty. This is the same pattern as California Western's San Diego State ties, and the opposite of Albany Law's UAlbany affiliation, which explicitly funds joint faculty research. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.lsac.org/choosing-law-school/find-law-school/jd-programs/southwestern",
      "https://en.wikipedia.org/wiki/Southwestern_Law_School",
      "https://www.swlaw.edu/"
    ]
  },
  {
    "state": "Georgia",
    "school": "Atlanta's John Marshall Law School",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "",
    "specs": "",
    "notes": "CORRECTED Sept 2026 — the previous entry recorded this school as 'closed 2020', which is wrong. AJMLS is open and ABA-accredited (approved 2005), with roughly 379 JD students. It converted from for-profit to 501(c)(3) nonprofit status (operated by Atlanta Law Center Inc.), sold its Midtown building in 2020 and relocated to One Marquis Tower in Peachtree Center in August 2021, and launched the Criminal and Civil Justice Institute in 2025. The 'closed' note most likely conflated this school with two different institutions: Chicago's John Marshall Law School (acquired by UIC in 2019, renamed UIC School of Law in 2021) and AJMLS's own Savannah Law School branch campus (ceased operations 2021). Independent law school; no parent research university, so there is no campus HPC to check. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://en.wikipedia.org/wiki/Atlanta's_John_Marshall_Law_School",
      "https://www.justia.com/law-schools/georgia/atlantas-john-marshall-law-school/",
      "https://www.americanbar.org/groups/legal_education/accreditation/approved-law-schools/archives/",
      "https://en.wikipedia.org/wiki/Savannah_Law_School"
    ]
  },
  {
    "state": "Massachusetts",
    "school": "New England Law Boston",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "None — independent law school, no parent institution",
    "specs": "",
    "notes": "Re-verified Sept 2026, first sourcing for this row. Standalone CONFIRMED: New England Law | Boston is an independent, ABA-accredited law school with no parent university. It was founded in 1908 as Portia Law School — the first law school in the nation established exclusively for women — was ABA-accredited in 1969, and shortened its name from New England School of Law to New England Law | Boston during its centennial in September 2008. MASSACHUSETTS CONTEXT WORTH RECORDING: this is one of eight Massachusetts law schools in this dataset, five of which sit on parents with serious compute (Harvard, BU, BC, Northeastern, UMass). New England Law has no such parent, and the Massachusetts Green High Performance Computing Center — the obvious regional route — is a closed membership consortium of its founding universities, reached through a member institution's own research computing group rather than open to any Massachusetts school. Same finding as Suffolk. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.justia.com/law-schools/massachusetts/new-england-law-boston/",
      "https://lawschool.uslegal.com/law-school-accreditaion/massachusetts-accredited-law-schools/",
      "https://www.abajournal.com/web/article/independent-law-schools-seek-backup-accreditor-as-aba-council-under-review"
    ]
  },
  {
    "state": "Minnesota",
    "school": "Mitchell Hamline School of Law",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "",
    "specs": "",
    "notes": "Re-verified Sept 2026 — standalone confirmed despite the shared name. Mitchell Hamline was formed in December 2015 by the combination of Hamline University School of Law and William Mitchell College of Law, and it retains an affiliation with Hamline University, but both institutions state plainly that they are separate entities operated independently and that Mitchell Hamline is an autonomous nonprofit governed by its own board of trustees. The affiliation delivers dual degrees, a 3+3 pathway, and student access to Hamline's library, athletic facilities and cultural programming — not research infrastructure. Hamline is itself a ~4,000-student institution offering 11 graduate degrees with no research-computing centre, so even a generous reading of the affiliation would not produce a compute pathway. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://mitchellhamline.edu/about/affiliation-with-hamline-university/",
      "https://www.hamline.edu/academics/college-schools/mitchell-hamline-school-law",
      "https://en.wikipedia.org/wiki/Mitchell_Hamline_School_of_Law"
    ]
  },
  {
    "state": "New York",
    "school": "Albany Law School",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "",
    "specs": "",
    "notes": "Re-verified Sept 2026. Albany Law School remains legally independent — it is the nation's oldest independent law school and has no parent university — so the standalone classification stands. However, the flat 'no parent research university' note undersold the situation: since September 2015 Albany Law has had a formal affiliation with the University at Albany (SUNY) that explicitly covers joint faculty research and grant-seeking, including a dedicated Collaborative Venture Fund that has funded cross-institution faculty teams, alongside cross-registration and joint degrees. UAlbany itself is well resourced: its AI Supercomputer is 24 NVIDIA DGX A100 systems (8 A100 GPUs each, 192 A100s total), it runs a separate on-premises HPC cluster with its own ITS access-request path, and UAlbany faculty can request allocations on New York State's Empire AI (Alpha at UB, plus Beta on NVIDIA DGX GB200 SuperPOD). See the caveat for what this does and does not mean for a law faculty member. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.albanylaw.edu/about/faq-the-university-albany-affiliation",
      "https://albany.atlassian.net/wiki/spaces/askit/pages/305168385/AI+Supercomputer",
      "https://tdx.albany.edu/TDClient/184/ITS/Requests/ServiceDet?ID=623",
      "https://www.albany.edu/news-center/news/2024-ualbany-unveils-powerful-new-ai-supercomputer",
      "https://tdx.albany.edu/TDClient/184/ITS/Requests/ServiceOffering/621/Empire-AI-Access-Request"
    ],
    "caveat": "Affiliation, not ownership — access unconfirmed. UAlbany's DGX A100 access page is written for UAlbany faculty, staff, postdocs and students, and its SUNY-wide extension is described for other SUNY campuses at a price. Albany Law is neither. The realistic route is a co-investigator relationship with a UAlbany faculty member under the affiliation, which the affiliation is explicitly designed to support but which no public page confirms as an account-granting path. Worth a direct ask to UAlbany ITS."
  },
  {
    "state": "New York",
    "school": "Brooklyn Law School",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "None — independent law school, no parent institution",
    "specs": "",
    "notes": "Re-verified Sept 2026, first sourcing for this row. Standalone is CONFIRMED in the school's own words: Brooklyn Law School states it is 'an independent institution, unaffiliated with any university or college, and the only law school in Brooklyn.' Founded 1901. The independence is deliberate rather than incidental — St. Lawrence University operated the school and conferred its degrees until 1968, when it decided to shut the school down; alumni organised and negotiated a repurchase of the school's assets specifically to keep it operating as an independent institution, and it moved to 250 Joralemon Street in 1969. So there is no parent university, and no prospect of one. ONE LEAD RULED OUT: New York State's Empire AI would be the obvious route for a New York law school, and it is a large one — Alpha at the University at Buffalo has supported 130+ projects, and Beta is a $40M NVIDIA system. But Empire AI is a closed consortium of exactly ten members: Columbia, Cornell, CUNY, the Icahn School of Medicine at Mount Sinai, NYU, Rochester Institute of Technology, RPI, the University of Rochester, SUNY and the Flatiron Institute. Brooklyn Law School is not among them, and the 2026 expansion of access was to SUNY campuses, not to unaffiliated private institutions. Unlike Albany Law, Brooklyn Law also has no affiliation agreement with a member institution to work through. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.brooklaw.edu/about-us/a-legacy-of-excellence/",
      "https://en.wikipedia.org/wiki/Brooklyn_Law_School",
      "https://www.empireai.edu/about/",
      "https://www.governor.ny.gov/news/governor-hochul-announces-empire-ai-suny-campus-partnerships-expand-access"
    ]
  },
  {
    "state": "New York",
    "school": "New York Law School",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "None — independent law school, no parent institution",
    "specs": "",
    "notes": "Re-verified Sept 2026, first sourcing for this row. Standalone CONFIRMED in the school's own words: NYLS describes itself as 'an independent law school' and 'an independently operated institution', founded 1891, Tribeca campus, marking its 135th anniversary in 2026. IMPORTANT DISAMBIGUATION: New York Law School is NOT New York University School of Law. They are entirely separate institutions with similar names in the same city — exactly the kind of confusion that produced this dataset's 'John Marshall closed 2020' error, so it is worth stating explicitly. MATERIAL DEVELOPMENT: NYLS has received institutional accreditation from the Middle States Commission on Higher Education, which its dean called a landmark moment in the school's 134-year history. It is now dual-accredited by both MSCHE and the ABA, and the school states this gives it 'the flexibility to create robust interdisciplinary programs and degrees typically available at law schools within larger universities.' That matters for this directory: an independent school building interdisciplinary and technology-adjacent programs is more likely to need compute over time. Worth re-checking. NYLS Dean and President Anthony W. Crowell is also founding president of the National Association of Standalone Graduate Schools. Empire AI ruled out — a closed ten-member consortium that does not include NYLS. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.nyls.edu/?p=617",
      "https://news.nyls.edu/new-york-law-school-receives-accreditation-from-middle-states-commission-on-higher-education/",
      "https://nyls.edu",
      "https://www.empireai.edu/about/"
    ]
  },
  {
    "state": "South Carolina",
    "school": "Charleston School of Law",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "",
    "specs": "",
    "notes": "Re-verified Sept 2026 — standalone confirmed, and the obvious merger hypothesis checked and closed. In October 2013 South Carolina's Commission on Higher Education waived a licence condition to permit exploration of a merger with the College of Charleston or another public school; that merger never happened. The school instead remained independent, survived a contested near-sale to InfiLaw, and converted from for-profit to nonprofit status in December 2024 following ABA acquiescence. It is ABA-accredited, has roughly 700 JD students, and lists no parent institution. Nothing to check for campus HPC — there is no parent university. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://en.wikipedia.org/wiki/Charleston_School_of_Law",
      "https://che.sc.gov/sites/che/files/Documents/Meetings/2024/CAAL/July%2018/5.a.Charleston%20School%20of%20Law_Final.pdf",
      "https://www.justia.com/law-schools/south-carolina/charleston-school-of-law/"
    ]
  },
  {
    "state": "Texas",
    "school": "South Texas College of Law Houston",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "None — independent law school, no parent institution",
    "specs": "",
    "notes": "Re-verified Sept 2026, first sourcing for this row. Standalone CONFIRMED: STCL Houston describes itself, and is described by LSAC and AALS, as a private, nonprofit, independent institution. Founded 1923, Houston's first and oldest law school, the only one in downtown Houston, roughly 350 first-year JD students, occupying a full city block. THE TEXAS A&M QUESTION, RESOLVED: this row needed care because STCL signed an affiliation agreement with Texas A&M University in 1998 and attempted a private/public merger under which it would have been branded the Texas A&M Law Center and awarded degrees under the A&M seal. That deal FELL THROUGH. Texas A&M instead acquired Texas Wesleyan's law school in 2013 to create today's Texas A&M School of Law, which is a separate institution from this one. What survives of the A&M relationship is student-facing: a joint JD/MBA with A&M's Mays Business School (and another with the University of St. Thomas Cameron School of Business), plus a 3+3 bachelor's/JD. No faculty research affiliation and no compute pathway. Also noted: MSCHE was approved as a secondary institutional accreditor in May 2025 and granted accreditation in October 2025 — the same backstop-accreditation move NYLS made. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.stcl.edu/about/",
      "https://www.lsac.org/choosing-law-school/find-law-school/jd-programs/south-texas",
      "https://en.wikipedia.org/wiki/South_Texas_College_of_Law_Houston",
      "https://www.msche.org/institution/13052/",
      "https://www.tshaonline.org/handbook/entries/south-texas-college-of-law"
    ]
  },
  {
    "state": "Vermont",
    "school": "Vermont Law and Graduate School",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "None — independent law school, no parent institution",
    "specs": "",
    "notes": "Re-verified Sept 2026, first sourcing for this row. Standalone CONFIRMED: VLGS is a private, independent law and graduate institution in South Royalton, founded 1972, the only ABA-accredited law school in Vermont, with roughly 720 students. It has been institutionally accredited by the New England Association of Schools and Colleges since 1980, so unlike NYLS and South Texas it has long had the accreditation backstop independent schools rely on for Title IV eligibility. NAME CHANGE EXPLAINED: the row's name is current. Vermont Law School rebranded to Vermont Law and Graduate School in 2022, funded in part by an anonymous $8 million gift — its largest ever — adding master's programs in environmental policy and animal law policy and putting equal weight on non-JD offerings. Anyone searching the older name will still find it. GOVERNANCE NOTE FOR RE-CHECKING: the school is between presidents. Rod Smolla led it from 2022 and stepped down in 2025 to return to teaching; Law Dean Beth McCormack and Graduate Dean Dan Bromberg are jointly managing day-to-day operations while a national search, launched April 2026, runs. Strategy can shift with a new president, so this row has a natural re-check trigger. No compute resource of its own was found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.vermontlaw.edu/news-and-events/vermont-law-and-graduate-school-launches-search-for-next-president",
      "https://en.wikipedia.org/wiki/Vermont_Law_and_Graduate_School",
      "https://vtdigger.org/2022/06/21/vermont-law-school-to-rebrand-add-masters-degree-programs/",
      "https://www.lsac.org/choosing-law-school/find-law-school/llm-and-other-law-programs-us-canada/vermont-law-and-graduate"
    ]
  },
  {
    "state": "Virginia",
    "school": "Appalachian School of Law",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "",
    "specs": "",
    "notes": "Re-verified Sept 2026 — standalone confirmed and the school is operating. ASL is a small private, independent, ABA-accredited law school in Grundy, Virginia, founded in 1994 with first students in 1997, currently enrolling roughly 180 students. It has no parent university and none of the acquisition or teach-out activity that affected other small independents. Under Virginia Code 23.1-219(D) it is exempt from a further SCHEV certificate because it has maintained a main campus under the same ownership for 20 years while remaining accredited. It does offer a Cybersecurity certificate among its five certificate programs, but that is curricular, not a compute resource. No parent university, so nothing to check. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://en.wikipedia.org/wiki/Appalachian_School_of_Law",
      "https://www.lsac.org/choosing-law-school/find-law-school/jd-programs/appalachian",
      "https://www.asl.edu/wp-content/uploads/2024/08/202425-Catalog-Student-Handbook-082024.pdf"
    ]
  },
  {
    "state": "California",
    "school": "Western State College of Law at Westcliff University",
    "university": "Westcliff University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "RECLASSIFIED Sept 2026 — schema fix plus a status check. The old row was internally inconsistent: tier 'standalone' and standalone=true, while also naming Westcliff University as the parent. Westcliff is a real parent institution, so standalone is now false and the tier is 'none' (parent exists, no research computing found) rather than 'standalone' (no parent at all). On status: the school survived. After parent Argosy University went into federal receivership in 2019 the ABA approved teach-out plans that would have closed it in December 2022, but Westcliff University acquired it instead, an acquisition approved by a federal court in Ohio; it hired five new full-time faculty in 2022 and continues to operate as an ABA-approved law school, now in Tustin, California. Westcliff is a career-focused private university offering bachelor's, master's and doctorate degrees with no research-computing center, HPC cluster or scheduler found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.westcliff.edu/academics/college-of-law/",
      "https://www.lsac.org/choosing-law-school/find-law-school/jd-programs/western-state",
      "https://nationaljurist.com/western-state-lives/",
      "https://www.businesswire.com/news/home/20220622005329/en/Western-State-College-of-Law-Adds-Five-New-Full-time-Faculty"
    ]
  },
  {
    "state": "Alabama",
    "school": "Thomas Goode Jones School of Law",
    "university": "Faulkner University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026 — Faulkner is a small, teaching-focused private university (Christian liberal arts, ~2,300 students) with no research-computing mission; its own 2026 IT update announcement covers only classroom/lab PC refreshes, not HPC. No evidence of any shared research-computing cluster found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.faulkner.edu/summer-it-updates/",
      "https://www.faulkner.edu/academics/computer-sciences/"
    ]
  },
  {
    "state": "Alabama",
    "school": "Cumberland School of Law",
    "university": "Samford University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026 — Samford's Technology Services department is entirely enterprise IT (classroom tech, wifi, email, printing, helpdesk); no mention of research computing, HPC, or GPU resources anywhere on the university's IT pages or its CS department page. Consistent with a teaching-focused university without a dedicated research-computing mission. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.samford.edu/departments/technology-services/about",
      "https://www.samford.edu/departments/technology-services/"
    ]
  },
  {
    "state": "Alabama",
    "school": "University of Alabama School of Law",
    "university": "University of Alabama (Tuscaloosa)",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "NVIDIA A100 tensor core GPUs on CHPC, exposed through a dedicated 'gpu' QoS (limit 3 jobs / 3 GPUs / 192 cores, up to 2 days)",
    "slurm": "Yes — Slurm QoS tiers (main, long, highmem, gpu) documented on CHPC",
    "hpcName": "UAHPC and CHPC clusters (UA Office for Research & Economic Development)",
    "specs": "Two clusters, UAHPC and CHPC, with published CPU and GPU specification tables. CHPC QoS limits: main 5 days / 1,000 jobs or cores; long 14 days / 100 jobs or cores; highmem 1 week with up to 6,000 GB RAM per user. A separate $96M, 40,000 sq ft High Performance Computing and Data Center at 709 Johnny Stallings Drive was completed in 2026 and is slated to open in early 2027 — a 5 MW facility scalable to 14.6 MW, funded by a $44.5M NIST disbursement plus $46M in state appropriation, intended to be petascale and the only Alabama system meeting Top500 benchmarks.",
    "notes": "RECLASSIFIED Sept 2026 — this row was scored 'construction', which was wrong in a way worth explaining. The new UA High Performance Computing and Data Center genuinely is still opening (early 2027), but 'construction' implies the university has no operational resource, and it does: UA already supervises two advanced computing clusters, UAHPC and CHPC, and CHPC's own technical specifications page documents an NVIDIA A100 'gpu' QoS available to users today. So the correct reading is 'modern hardware now, frontier-class facility pending'. The new centre is also explicitly designed to support research across all campuses of the University of Alabama System and statewide collaborations, so it may become a network-tagged resource once operational. Re-check this row in 2027 — it is the most predictable upgrade in the dataset.",
    "networks": [],
    "sources": [
      "https://hpc.ua.edu/current-services/technical-specifications/chpc/",
      "https://hpc.ua.edu/current-services/technical-specifications/",
      "https://hpc.ua.edu/",
      "https://news.ua.edu/2024/11/stage-iv-approval-for-high-performance-computing-center/",
      "https://hpc.ua.edu/about/"
    ],
    "caveat": "The 2027 facility is not the current resource. Do not read the petascale/Top500 language in press coverage as describing what a researcher can use today; that describes the building opening in early 2027. Today's confirmed hardware is the CHPC A100 QoS. Primary users named for the new centre are science and engineering institutes (Alabama Life Research Institute, Alabama Materials Institute, Alabama Water Institute, Alabama Transportation Institute), so law-faculty access is unconfirmed."
  },
  {
    "state": "Arkansas",
    "school": "University of Arkansas School of Law (Fayetteville)",
    "university": "University of Arkansas",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Pinnacle II (2022): 71 AMD-based nodes with 74 NVIDIA Ampere-class GPUs (A100-generation). Pinnacle I (2019): roughly 100-106 mostly Intel-based nodes with 20-26 NVIDIA Volta (V100) GPUs across its GPU nodes. Combined floating-point capacity exceeds 1 PetaFLOPS, contributed mostly by the GPUs. No Hopper-class hardware was found, which is why this is modern rather than frontier.",
    "slurm": "Yes — Slurm across Pinnacle and Karpinski",
    "hpcName": "Arkansas High Performance Computing Center (AHPCC)",
    "specs": "Pinnacle I also carries 8 big-memory nodes at 768GB RAM each. InfiniBand fabric at 100 Gb/s. Slurm scheduling across Pinnacle and the newer Karpinski system. Pinnacle offers the Open OnDemand 'Pinnacle Portal' web interface giving GUI access to applications such as MATLAB without any command-line work, alongside virtual machines, containers and remote desktop. Older Trestles and Razor systems remain listed. AHPCC is a University of Arkansas Core Facility directed by Rick McMullen.",
    "notes": "Re-verified Sept 2026 — modern CONFIRMED, and the statewide access claim in this row is GENUINE, which is worth stating because several comparable claims in this dataset failed when tested. AHPCC states repeatedly and in its own words that it is available to faculty, staff and students at ALL OF THE ARKANSAS PUBLIC UNIVERSITIES, and to their collaborators inside and outside the state, with most services provided free of charge to eligible researchers and priority condo access available on top. Accounts are requested through an internal form. The eligibility line is institutional and carries no disciplinary limit, so a law faculty member qualifies on its face. Contrast with Mississippi's MCSR, which is also public-institution-only and therefore excluded private Mississippi College — the difference here is simply that both Arkansas law schools sit at public universities. Useful secondary benefit: AHPCC participates in the NSF Campus Champions programme and explicitly assists researchers in acquiring and using resources at national supercomputer centres, making it a practical on-ramp to the national ACCESS allocations described in this site's header. Fayetteville is AHPCC's home campus — the centre is a University of Arkansas core facility located there, so this is an on-campus resource and the strongest position of the two Arkansas rows.",
    "networks": [
      "AHPCC (Arkansas statewide)"
    ],
    "sources": [
      "https://hpc.uark.edu/",
      "https://aralliance.org/core-facilities/ua-arkansas-high-performance-computing-center/",
      "https://hpc.uark.edu/hpc-resources/",
      "https://hpc.uark.edu/hpc-about/",
      "https://hpcwiki.uark.edu/"
    ],
    "caveat": "Public institutions only, and Karpinski is unverified here. AHPCC's eligibility is scoped to Arkansas PUBLIC universities and their collaborators — fine for both Arkansas law schools, but it would not extend to a private institution in the state. Separately, AHPCC's documentation references a newer system named Karpinski alongside Pinnacle in its Slurm scheduler pages; its specifications were not established in this pass, so the GPU inventory here may understate current capacity. Worth re-checking."
  },
  {
    "state": "Arkansas",
    "school": "William H. Bowen School of Law",
    "university": "University of Arkansas at Little Rock",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Pinnacle II (2022): 71 AMD-based nodes with 74 NVIDIA Ampere-class GPUs (A100-generation). Pinnacle I (2019): roughly 100-106 mostly Intel-based nodes with 20-26 NVIDIA Volta (V100) GPUs across its GPU nodes. Combined floating-point capacity exceeds 1 PetaFLOPS, contributed mostly by the GPUs. No Hopper-class hardware was found, which is why this is modern rather than frontier.",
    "slurm": "Yes — Slurm across Pinnacle and Karpinski",
    "hpcName": "Arkansas High Performance Computing Center (AHPCC) - statewide access",
    "specs": "Pinnacle I also carries 8 big-memory nodes at 768GB RAM each. InfiniBand fabric at 100 Gb/s. Slurm scheduling across Pinnacle and the newer Karpinski system. Pinnacle offers the Open OnDemand 'Pinnacle Portal' web interface giving GUI access to applications such as MATLAB without any command-line work, alongside virtual machines, containers and remote desktop. Older Trestles and Razor systems remain listed. AHPCC is a University of Arkansas Core Facility directed by Rick McMullen.",
    "notes": "Re-verified Sept 2026 — modern CONFIRMED, and the statewide access claim in this row is GENUINE, which is worth stating because several comparable claims in this dataset failed when tested. AHPCC states repeatedly and in its own words that it is available to faculty, staff and students at ALL OF THE ARKANSAS PUBLIC UNIVERSITIES, and to their collaborators inside and outside the state, with most services provided free of charge to eligible researchers and priority condo access available on top. Accounts are requested through an internal form. The eligibility line is institutional and carries no disciplinary limit, so a law faculty member qualifies on its face. Contrast with Mississippi's MCSR, which is also public-institution-only and therefore excluded private Mississippi College — the difference here is simply that both Arkansas law schools sit at public universities. Useful secondary benefit: AHPCC participates in the NSF Campus Champions programme and explicitly assists researchers in acquiring and using resources at national supercomputer centres, making it a practical on-ramp to the national ACCESS allocations described in this site's header. Bowen is at UA Little Rock, a different Arkansas public university from AHPCC's Fayetteville home. That is exactly the situation the statewide eligibility rule is written to cover, so this row rests on the published policy rather than on campus co-location — a stronger footing than the cross-campus assumptions that failed for Missouri and the University of Baltimore, because here the policy is explicit.",
    "networks": [
      "AHPCC (Arkansas statewide)"
    ],
    "sources": [
      "https://hpc.uark.edu/",
      "https://aralliance.org/core-facilities/ua-arkansas-high-performance-computing-center/",
      "https://hpc.uark.edu/hpc-resources/",
      "https://hpc.uark.edu/hpc-about/",
      "https://hpcwiki.uark.edu/"
    ],
    "caveat": "Public institutions only, and Karpinski is unverified here. AHPCC's eligibility is scoped to Arkansas PUBLIC universities and their collaborators — fine for both Arkansas law schools, but it would not extend to a private institution in the state. Separately, AHPCC's documentation references a newer system named Karpinski alongside Pinnacle in its Slurm scheduler pages; its specifications were not established in this pass, so the GPU inventory here may understate current capacity. Worth re-checking."
  },
  {
    "state": "Arizona",
    "school": "Sandra Day O'Connor College of Law",
    "university": "Arizona State University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Sol: over 290 GPUs — 80GB A100s (224 at launch across 56 nodes of four), 24GB A30s (15 across four nodes of three), and a smaller number of 80GB H100s. Phoenix: 360 GPUs (A100, V100, GTX 1080, RTX 2080). Agave's GPUs are gone with the cluster.",
    "slurm": "Yes — Slurm with a job submit plugin, high-availability slurmctld with failover, MariaDB backend, and depth-oblivious Fairshare priority",
    "hpcName": "ASU Research Computing — Sol (flagship) and Phoenix; Agave decommissioned",
    "specs": "Sol: Dell-built, 21,000+ AMD EPYC (Zen3/7713) cores, most nodes 512GB RAM, seven high-memory nodes at 2TB, 200GB/s HDR InfiniBand, 4PB Dell BeeGFS scratch plus 2PB Dell PowerScale, observed GPU-only HPL of 2.272 PFLOP/s. Housed OFF-CAMPUS in a Tier 3+ Iron Mountain commercial datacenter in Phoenix providing FISMA High security, in twenty APC 42U racks with CoolIT liquid-chilled doors. Phoenix: 17,000+ heterogeneous Intel cores, 128–384GB per node, three high-memory nodes at 1.5TB.",
    "notes": "Re-verified Sept 2026 and the headline figure corrected. The old '580+ GPU accelerators' was an aggregate that no longer describes the estate: it appears to have summed Sol against the older Agave cluster, and ASU Research Computing's own documentation now states plainly that AGAVE HAS BEEN DECOMMISSIONED, with users migrated to the Iron Mountain datacenter. The current picture is Sol (over 290 GPUs) plus Phoenix (360 GPUs, public access listed as coming soon). Frontier is CONFIRMED but on a narrow basis — ASU's own wiki describes Sol as carrying A30s, 80GB A100s 'and a few H100s', so the Hopper-class presence is real but small, and the bulk of the fleet is Ampere. Access is excellent for a law faculty member: Sol is free for researchers, ASU describes AI applications as 'freely available for faculty use', and the launch framing was explicitly that Sol gives 'researchers across all disciplines' additional resources. GPU time is metered in core-hour equivalents (an A100 hour counts as 25 CHE, an A30 hour 20, a MIG slice 3) rather than billed in dollars.",
    "networks": [],
    "sources": [
      "https://docs.rc.asu.edu/intro",
      "https://asurc.atlassian.net/wiki/spaces/RC/pages/1852637228/The+Sol+supercomputer",
      "https://dl.acm.org/doi/fullHtml/10.1145/3569951.3597573",
      "https://cores.research.asu.edu/computing-and-data-services/research-computing/capabilities/",
      "https://www.hpcwire.com/2022/07/13/asu-launching-sol-supercomputer/"
    ],
    "caveat": "Frontier rests on 'a few H100s'. Sol's GPU fleet is overwhelmingly A100/A30; the H100 count is not published and ASU's capability page describes Sol as 'over 290 NVIDIA A100 GPUs' without mentioning Hopper at all. If you need H100-class hardware specifically rather than GPU access generally, confirm availability before relying on this row. Phoenix is also listed as 'public access coming soon', so its 360 GPUs may not yet be generally requestable."
  },
  {
    "state": "Arizona",
    "school": "James E. Rogers College of Law",
    "university": "University of Arizona",
    "standalone": false,
    "tier": "older",
    "tierLabel": "Older-gen GPUs confirmed (V100/T4/P100/K-series/etc.)",
    "gpu": "Currently operational: NVIDIA V100/V100S 32GB on Puma — 6 Altus XE2214GT GPU nodes with four V100S each. The H200s referenced in the previous version of this row belong to LYNX, which is not yet in service: two 8-GPU H200 nodes (16 H200 total) alongside 40 CPU-compute nodes and one 3TB high-memory node. Ocelote's 46 P100 nodes and El Gato's K20X GPUs are gone — both clusters have been decommissioned.",
    "slurm": "Yes — Slurm across UArizona HPC clusters since July 2021",
    "hpcName": "UArizona HPC — Puma (current); Lynx arriving late Fall 2026",
    "specs": "Puma (Penguin Computing, 2020): 48 Altus XE2242 CPU nodes at 96 cores and 512GB RAM each, 6 GPU nodes, 2 high-memory nodes at 3TB, originally 19,200 AMD EPYC 7642 Rome cores and grown by researcher buy-in to roughly 28,950 cores, Rocky Linux 9. Housed in the Research Data Center. In FY2025 UArizona HPC supported over 300 million compute hours for 80+ departments, 500 PIs and 1,500 users; Puma passed its 10 millionth job. Lynx: assembly and installation scheduled September 2026, testing and integration to follow, public rollout expected late in the Fall semester.",
    "notes": "RECLASSIFIED Sept 2026, frontier -> older. This is the mirror image of the Alabama error found earlier in this project. Alabama was scored 'construction' while A100s were already running; Arizona was scored 'frontier' on hardware that has not arrived. UArizona's own HPC documentation is explicit on both halves: the H200 nodes belong to Lynx, which as of this check is 'scheduled for assembly and installation in September 2026, followed by testing and integration', with public rollout expected 'sometime in late Fall semester' and no firm release date announced; meanwhile 'both El Gato and Ocelote have been decommissioned'. What a researcher can actually use today is Puma, whose GPU nodes carry V100S — which is precisely the 'older' tier definition. Access itself is very good and unchanged: HPC services are at no cost, every faculty member gets a free standard allocation of CPU time and storage, systems are open to all university faculty, staff and students, and windfall hours are available on idle capacity. THIS ROW SHOULD FLIP TO FRONTIER once Lynx is live — it is the most predictable upgrade in the dataset after Alabama's, and worth re-checking in early 2027.",
    "networks": [],
    "sources": [
      "https://hpc.arizona.edu/",
      "https://hpcdocs.hpc.arizona.edu",
      "https://hpcdocs.static.arizona.edu/",
      "https://uarizona.atlassian.net/wiki/spaces/UAHPC/pages/75990208/Compute+Resources",
      "https://hpcdocs.hpc.arizona.edu/support_and_training/grants/files/University_Arizona_Cyberinfrastructure_Report.pdf"
    ],
    "caveat": "Announced is not operational. Sixteen H200 GPUs are coming on Lynx, but no firm release date had been published at the time of this check, and new clusters routinely slip. Do not cite Arizona as an H200 site yet. Equally, do not rely on Ocelote or El Gato documentation found elsewhere — both machines are decommissioned and their P100/K20X GPUs are no longer available."
  },
  {
    "state": "California",
    "school": "Stanford Law School",
    "university": "Stanford University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Sherlock's free gpu partition, open to every Sherlock user: 16 NVIDIA H100 SXM5 (two 4-GPU nodes and one 8-GPU node) and 32 L40S (8 nodes x 4), plus older RTX 3090, RTX 2080 Ti, V100/V100S, TITAN Xp and P100 nodes (49 GPU nodes in all). The dev partition adds A30 MIG slices. Owner-only: PI groups can buy 8x NVIDIA H200 nodes (SH4_G8TF64.1, sold since Aug 2025), which run in the owners partition rather than the free pool. Separately, Marlowe has 248 NVIDIA H100 GPUs across 31 nodes.",
    "slurm": "Yes",
    "hpcName": "Stanford Research Computing — Sherlock cluster; plus Marlowe (Stanford Data Science's NVIDIA H100 system)",
    "specs": "Sherlock (as of October 2026): 2,056 compute nodes; 75,972 CPU cores; 1,256 GPUs; 18.78 PFLOPs (FP64); 14.6 PB Lustre scratch with over 600 GB/s of I/O; direct access to the 93.9 PB Oak research storage; 9,152 user accounts and 1,347 PI groups from all seven Stanford schools. Marlowe: 31 H100 nodes (248 GPUs), scheduled with Slurm.",
    "notes": "Open to all Stanford faculty incl. Law School (free base tier); condo/owner model for dedicated priority nodes; also HIPAA/PHI-compliant Carina environment. CORRECTED Sept 2026: initial research understated this — Sherlock is one of the most powerful academic clusters in the US. UPDATED Oct 2026: Stanford also runs Marlowe, a 248-GPU NVIDIA H100 system open to all Stanford faculty PIs; the PI must apply personally. New PIs get 5,000 free introductory GPU-hours (Basic Access, preemptible, valid one year). Larger non-preemptible projects pay a subsidized recharge of $0.30 per GPU-hour from Sep 1, 2026 ($0.25 preemptible), and jobs run through Slurm. On Sherlock itself, the free pool every user can reach is the gpu partition, which includes 16 H100s. The 8x H200 nodes sold since Aug 2025 are owner-purchased and sit in the owner-only partition. Stanford has postponed indefinitely the planned retirement of its oldest Sherlock 2.0 nodes, so the legacy GPUs remain in service. Carina, the PHI/high-risk-data platform, is still offered to PI-led teams.",
    "networks": [],
    "sources": [
      "https://www.sherlock.stanford.edu/docs/tech/",
      "https://news.sherlock.stanford.edu/publications/sherlock-4-0-a-new-cluster-generation",
      "https://news.sherlock.stanford.edu/labels/blog,data,hardware,new",
      "https://www.sherlock.stanford.edu/docs/tech/facts/",
      "https://www.sherlock.stanford.edu/docs/user-guide/gpu/",
      "https://www.sherlock.stanford.edu/docs/",
      "https://news.sherlock.stanford.edu/publications/introducing-sh4_g8tf64-1-now-with-8x-h200-gpus",
      "https://news.sherlock.stanford.edu/publications/an-update-about-our-plans-to-retire-sherlock-2-0",
      "https://marlowe-research.stanford.edu/access",
      "https://marlowe-research.stanford.edu/",
      "https://srcc.stanford.edu/systems"
    ]
  },
  {
    "state": "California",
    "school": "UC Berkeley School of Law",
    "university": "University of California-Berkeley",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Savio regular-priority FCA pool: 136 RTX A5000 (savio4_gpu), 16 A40, 2 V100 and 28 GTX 2080Ti (savio3_gpu). L40 nodes (3 x 8 L40, savio4_gpu) and TITAN RTX nodes are reachable by FCA users only through the preemptible low-priority queue (savio_lowprio). GTX 1080Ti nodes were retired in summer 2026. No H100/H200 on Savio. Separately, the Statistics department's own SCF cluster (not part of Savio) has research-group-owned H200 and A100 servers.",
    "slurm": "Yes",
    "hpcName": "Berkeley Research Computing - Savio cluster",
    "specs": "600 compute nodes and 15,300+ cores (BRC's April 2020 figure; the current hardware tables list roughly 785 nodes); 12 PB Lustre global scratch plus VAST home storage",
    "notes": "Free access to all faculty via Faculty Computing Allowance (FCA); condo option for dedicated priority nodes. CORRECTED Sept 2026 with specific hardware config table. CORRECTED Oct 2026: Savio's GTX 1080Ti nodes were retired in summer 2026; its L40 and TITAN RTX nodes are open to Faculty Computing Allowance users only through the preemptible low-priority queue; and the campus H200s are on the Statistics department's separate SCF cluster, not Savio condos. Tier unchanged (modern).",
    "networks": [],
    "sources": [
      "https://docs-research-it.berkeley.edu/services/high-performance-computing/user-guide/hardware-config",
      "https://docs-research-it.berkeley.edu/services/high-performance-computing/user-guide/running-your-jobs/specific-resources",
      "https://docs-research-it.berkeley.edu/services/high-performance-computing/overview/system-overview/",
      "https://docs-research-it.berkeley.edu/services/high-performance-computing/getting-account/faculty-computing-allowance/",
      "https://computing.stat.berkeley.edu/servers/gpu-servers/"
    ],
    "caveat": "Savio itself has no Hopper-class GPUs. Its regular-priority Faculty Computing Allowance pool tops out at A40 and RTX A5000 (16 A40, 136 A5000, 2 V100, 28 GTX 2080Ti); the L40 and TITAN RTX nodes are open to FCA users only through the preemptible low-priority queue. The campus H200s (and many A100s) are on the Statistics department's own Statistical Computing Facility (SCF) cluster, which is separate from Savio; they are owned by research groups (e.g. Steinhardt, Song Mei, Berkeley NLP) and are available to other SCF users only at lower, preemptible priority. This row is therefore scored modern on what a law faculty member could actually request through the FCA, by the same reasoning applied to Hawai'i and New Hampshire. Reaching the H200s would mean an SCF account or a collaboration with an owning group."
  },
  {
    "state": "California",
    "school": "UCLA School of Law",
    "university": "University of California-Los Angeles",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Hoffman2 production (Altair Grid Engine) GPU table: NVIDIA H200 (141GB, 4 per node), H100 (96GB, 1 per node), L40S (48GB), A6000 (48GB, 2 per node), A100 (80GB or 40GB, 4 per node), plus older RTX 2080Ti, V100 (32GB), GTX 1080Ti, Tesla P4 and Tesla K40. Each type is requested explicitly (e.g. -l gpu,H200,cuda=1).",
    "slurm": "Transitioning",
    "hpcName": "UCLA OARC - Hoffman2 Cluster",
    "specs": "800+ nodes; 26,000+ cores; 179TB+ aggregate memory; 3 PB NetApp plus 6 PB flash-based VAST storage (home, project, scratch)",
    "notes": "Free to entire UCLA campus incl. Law; historically Altair Grid Engine (qsub/qrsh), migrating to SLURM (Slurm preview partition live since ~2024). CORRECTED Sept 2026 — A100 80GB confirmed present. CORRECTED Oct 2026: Hoffman2's GPU table now lists H200 and H100 cards alongside L40S, A6000 and A100; the tier stays modern until campus-user access to the Hopper nodes is documented (see caveat). The Slurm preview cluster (60 new dedicated nodes; 24-hour job limit unless the PI's nodes have migrated) was announced in 2026, not around 2024. Production still runs Altair Grid Engine.",
    "networks": [],
    "sources": [
      "https://www.hoffman2.idre.ucla.edu/Using-H2/Computing/Computing.html",
      "https://www.hoffman2.idre.ucla.edu/About/System-overview.html",
      "https://www.hoffman2.idre.ucla.edu/Using-H2/Computing/Computing-Altair-production.html",
      "https://www.hoffman2.idre.ucla.edu/Policies/Job-scheduling-policy.html",
      "https://www.hoffman2.idre.ucla.edu/Using-H2/Computing/Computing-Slurm-preview.html",
      "https://www.hoffman2.idre.ucla.edu/News-archive/News.html"
    ],
    "caveat": "Hopper access for campus users isn't documented, which is why this row stays modern. OARC's GPU table lists H200 (4 per node) and H100 cards on the production Grid Engine cluster, but users whose PI hasn't bought nodes ('campus users') run only on OARC's base pool, with 24-hour jobs, and the docs don't say whether that pool includes the Hopper nodes or whether they are group-owned. A law faculty member should confirm with OARC before planning on H100/H200 time."
  },
  {
    "state": "California",
    "school": "UC Davis School of Law (King Hall)",
    "university": "University of California-Davis",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Hive general use: NVIDIA A6000 (5 for general use; the free publicgrp share in the high partition is 4 A6000s combined, max 1 GPU per job). PI-owned on Hive and reachable by any user only via the preemptible low partition: 4x H100, 8x A100, 4x A6000. Franklin (College of Biological Sciences, restricted): 72 RTX A4000/A5000/RTX 6000 Ada/RTX 2080 Ti.",
    "slurm": "Yes",
    "hpcName": "HPC@UCD (Farm, Franklin, Hive clusters)",
    "specs": "Hive: 20 core nodes x 128 CPU cores / 2 TB RAM; 6,720 CPU cores and 61 TiB RAM total (as of Sept 22, 2025); Quobyte parallel file system. Across all HPC@UCD-managed clusters (Farm, Franklin, Hive, HPC2, many restricted): 48,000+ CPU cores, 200+ GPUs, dozens of PB storage.",
    "notes": "Free access to all UC Davis staff, faculty, grad students on Hive; SLURM confirmed. CORRECTED Sept 2026 — H100/A100 confirmed present on Hive. CORRECTED Oct 2026: tier frontier → modern. Hive's H100s/A100s are PI-owned and reachable by free users only in the preemptible low partition; the free share is A6000s. Same treatment as Hawai'i.",
    "networks": [],
    "sources": [
      "http://docs.hpc.ucdavis.edu/hive/",
      "https://hpc.ucdavis.edu/clusters",
      "https://docs.hpc.ucdavis.edu/scheduler/free-access/",
      "https://docs.hpc.ucdavis.edu/franklin/"
    ],
    "caveat": "Hopper hardware exists on Hive but is PI-owned: 4 H100s (plus 8 A100s and 4 A6000s) sit in the low partition and can be used by any UC Davis user only when the owning groups are not using them; low jobs are capped at 7 days and are killed and requeued when owners need the nodes. The guaranteed free share for all users (publicgrp) is 128 CPUs, 2,000 GB RAM and 4 A6000 GPUs combined, one GPU per job. Scored modern on that general-use pool, consistent with Hawai'i."
  },
  {
    "state": "California",
    "school": "UC Irvine School of Law",
    "university": "University of California-Irvine",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "HPC3 (as of May 2026): 14 nodes x 4 NVIDIA V100 (16GB); 18 nodes x 4 A30 (24GB); 4 nodes x 2 A100 (80GB); 3 nodes x 4 L40S (48GB); 7 nodes x 4 RTX PRO 6000 Blackwell (96GB). V100/A30/A100 run in the gpu partition and L40S/RTX 6000 in gpu32; both need a GPU lab account requested by a faculty PI. Anyone can use idle GPUs without a special account through the preemptible free-gpu/free-gpu32 partitions. No H100/H200.",
    "slurm": "Yes",
    "hpcName": "Research Cyberinfrastructure Center (RCIC) - HPC3 cluster",
    "specs": "261+ batch-accessible nodes; DFSx/CRSP parallel storage",
    "notes": "Open to all UCI researchers, no school/dept affiliation required; SLURM confirmed; allocation/banking model (no owner-dedicated queues). CORRECTED Sept 2026 with exact node counts. UPDATED Oct 2026: HPC3 added L40S and RTX PRO 6000 Blackwell nodes (gpu32 partition); still no Hopper, so modern stands. Any UCInetID holder gets a free account. Faculty get a no-cost baseline of 100,000 CPU core-hours and 1,000 GPU-hours every 6 months. The allocated GPU partitions need a PI-requested GPU lab account; free-gpu is open to anyone but preemptible.",
    "networks": [],
    "sources": [
      "https://rcic.uci.edu/hpc3/specs.html",
      "https://rcic.uci.edu/slurm/slurm.html",
      "https://rcic.uci.edu/about/allocations.html",
      "https://rcic.uci.edu/account/getting-account.html",
      "https://rcic.uci.edu/hpc3/hpc3.html"
    ]
  },
  {
    "state": "California",
    "school": "USC Gould School of Law",
    "university": "University of Southern California",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Discovery (general use, Slurm): over 180 NVIDIA GPUs across 200+ GPU nodes. Types: L40S (3 per node), A100 40GB and 80GB (2 per node, selected with --constraint=a100-40gb/a100-80gb), A40 (2), V100 (2) and P100 (2), requested as e.g. --gpus-per-task=a40:1; max 36 GPUs per user. No Hopper on Discovery. Topanga AI Computing (launched Sept 9, 2026; pay-as-you-go container sessions on Backend.AI instead of Slurm): NVIDIA H200 NVL 141GB, fractional GPUs up to 4 per session; max 4 GPUs per user. Endeavour condo lease options: V100, A40 and A100 80GB nodes.",
    "slurm": "Yes — Slurm with named partitions (gpu, epyc-64, largemem, main) and GPU type selection",
    "hpcName": "CARC — Discovery (general use), Endeavour (condo), and Topanga AI Computing (pay-as-you-go H200)",
    "specs": "Discovery: 2 shared login nodes, roughly 20,000 CPU cores across about 500 compute nodes, typically dual 8-to-16-core processors. Node types include AMD EPYC 9534 (64-core, 748GB, 3x L40S), EPYC 7513 and 7313 (248GB, some with 2x A100 or A40), EPYC 7282 (32-core, 248GB, 2x A40), and largemem EPYC 9354 nodes with about 1.5 TB. Launched August 2020. A user may hold at most 36 GPUs at one time across all their jobs. Default Discovery allocations are free (Small: 12,000,000 SUs; up to 120,000,000 free). CARC supports over 100 research groups across a range of academic disciplines and runs an annual project and allocation review cycle plus summer training workshops.",
    "notes": "Re-verified Sept 2026 — modern CONFIRMED and the '180+ GPUs' figure in this row matches CARC's own services page. Discovery is the general-use cluster open to USC researchers, with Endeavour available as a condo option for groups that buy hardware. CARC describes its user base as spanning a variety of academic disciplines rather than being science-limited, so a Gould faculty member is in scope. CARC is also building out a regional role through an NSF CC* Regional Computing award aimed at forging a research computing alliance across Southern California, which is worth watching as a possible future network tag for other California schools in this dataset. UPDATED Oct 2026: CARC launched Topanga AI Computing on Sept 9, 2026: pay-as-you-go H200 NVL sessions open to all CARC users with an active project and a USC billing worktag. The tier therefore moves to frontier, though Discovery itself still tops out at L40S/A100. The NSF CC* regional award (#2232917) has produced the Laguna cluster (16 CPU nodes plus 8 GPU nodes with 2 L40S each). USC CARC and Los Nettos run it for under-resourced Southern California institutions; the participating institutions are not listed publicly, so no network tag yet.",
    "networks": [],
    "sources": [
      "https://www.carc.usc.edu/about/services",
      "https://www.carc.usc.edu/user-guides/advanced-hpc-programming/gpu-programming",
      "https://www.carc.usc.edu/user-guides/hpc-systems/discovery/resource-overview-discovery",
      "https://www.carc.usc.edu/",
      "https://www.carc.usc.edu/latest-news/2026-09-09-topanga-ai-computing-launch",
      "https://www.carc.usc.edu/user-guides/ai-computing/topanga-ai-computing/topanga-ai-computing-overview",
      "https://www.carc.usc.edu/user-guides/project-and-allocation-management/allocation-types-and-requests",
      "https://www.carc.usc.edu/research/awards/laguna-cluster-and-socal-research-computing-alliance",
      "https://uschpc.github.io/regional-computing-website/user-guides/get-started-laguna/resource-overview.html"
    ],
    "caveat": "No sensitive data, which matters more for law than for most fields. CARC states plainly that its systems DO NOT currently support the use or storage of sensitive data, explicitly including HIPAA-, FERPA- and CUI-regulated material, and that a secure computing environment is still being built. A law researcher working with confidential client files, sealed records, protected student data or any regulated dataset cannot use Discovery for it and should contact CARC before starting. This is the sharpest data-handling restriction found on any row in this dataset. Hopper access is paid: the H200s are on Topanga, which bills each session-hour to a project's USC worktag (e.g. $3.00/hour for one full GPU), caps users at 4 GPUs. Discovery's free general-use GPUs top out at L40S/A100."
  },
  {
    "state": "California",
    "school": "Santa Clara University School of Law",
    "university": "Santa Clara University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "General-use nodes open to all WAVE users: 1 hub node with 8x NVIDIA RTX PRO 6000, 1 AMD node with 8x AMD Instinct MI100, 4 GPU nodes with 2x NVIDIA Tesla V100 32GB each, and 2 login nodes with 1x NVIDIA A16 each. Lab-owned condo nodes that general users can use when idle, subject to preemption: 2 nodes with 1x NVIDIA GH200 Grace Hopper each, 1 node with 8x A100, 1 node with 4x L40S, 3 nodes with A16 (8/4/8), 1 node with 2x V100, plus older GeForce GTX 1080/1080 Ti and GTX Titan X nodes. 69 GPUs in total, per SCU's About WAVE page.",
    "slurm": "Yes",
    "hpcName": "WAVE HPC (Wiegand Advanced Visualization Environment)",
    "specs": "2,076 CPU cores, 69 GPUs and 400TB of VDURA parallel storage (About WAVE). Main compute is 12 heterogeneous physical nodes of five types: 4 CPU nodes, 4 V100 GPU nodes, 2 high-memory nodes with 2TB RAM each, 1 AMD MI100 node and 1 RTX PRO 6000 hub node. Add 2 login nodes and lab-owned condo nodes. 100G InfiniBand (Mellanox QM8700). NSF-funded storage: a 92TB all-flash array and a 456TB HDD array. Access only through Slurm login nodes or Open OnDemand, behind the SCU VPN or campus network.",
    "notes": "All SCU faculty and staff eligible (students need faculty sponsor); SLURM confirmed. CORRECTED Sept 2026 — A16 (Ampere-gen) confirmed, total 69 GPUs. CORRECTED Oct 2026: WAVE's technical-specifications page moved and now itemizes every GPU, so the A16-only description was out of date. General-use nodes carry RTX PRO 6000, MI100, V100 and A16 GPUs. Lab-owned condo nodes add GH200, A100, L40S and older GeForce cards (see caveat). Eligibility re-confirmed: all SCU faculty and staff can request an account through a WAVE support ticket, using SCU login with DUO from the campus network or VPN. Students need a faculty sponsor.",
    "networks": [],
    "sources": [
      "https://www.scu.edu/wave/wave-hpc-technical-specifications/",
      "https://www.scu.edu/wave/about-wave/",
      "https://www.scu.edu/wave/getting-started/"
    ],
    "caveat": "Rated on the general-use RTX PRO 6000 and MI100 nodes. The frontier-class GH200s (two nodes, one GPU each), the 8x A100 node and the 4x L40S node are lab-owned condo hardware. General users can run on them only when idle, and SCU says those jobs can be cancelled and rerun when the owning lab needs the node. Per-user limits are also tight: the gpu partition allows 4 GPUs for 2 days, the RTX PRO 6000 hub partition 1 GPU for 1 day, and the condo partition 2 days."
  },
  {
    "state": "California",
    "school": "Pepperdine Caruso School of Law",
    "university": "Pepperdine University",
    "standalone": false,
    "tier": "cpuonly",
    "tierLabel": "No GPU confirmed (CPU/RAM-only)",
    "gpu": "None — the published system description lists CPU blades only, with no GPU of any generation",
    "slurm": "Not confirmed — no scheduler named in the public documentation",
    "hpcName": "Pepperdine IT research computing environment (no named cluster)",
    "specs": "Two Cisco blades, each with 1TB of RAM and Intel Xeon Gold 6238 CPUs providing up to 44 cores per blade, plus 30TB of storage — described by Pepperdine IT as roughly eight times the processing power of a standard workstation. Housed in an energy-efficient data centre designed for high-density research computing clusters and storage.",
    "notes": "Re-verified Sept 2026 — cpuonly CONFIRMED, and now with exact hardware rather than an assumption. Pepperdine IT, in partnership with the offices of the Vice Provost and of Research, Grants and Foundation Relations, offers a high-performance computing environment for qualified university research projects, and works directly with faculty to design and manage projects at scale. The full published system description is two Cisco blades with Xeon Gold 6238 CPUs — no GPU appears anywhere in it, which is what confirms the tier rather than merely failing to contradict it. Worth noting that Pepperdine is a Carnegie R2 institution with real AI activity: the Keck Data Science Institute is directed by Fabien Scalzo, whose work is in AI for medical imaging, and an undergraduate mentored there published 18 peer-reviewed articles and won an NSF GRFP. That AI work is evidently not running on these two blades, so there is likely lab-level or cloud GPU capacity somewhere at Pepperdine that public IT pages do not describe. The Caruso School of Law is one of Pepperdine's four graduate schools. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://community.pepperdine.edu/it/services/research-computing.htm",
      "https://www.pepperdine.edu/academics/research/",
      "https://en.wikipedia.org/wiki/Pepperdine_University"
    ],
    "caveat": "Project-qualified, not open-access, and no scheduler is named. Pepperdine frames this as a service for 'qualified university research projects' designed and deployed with IT on a per-project basis, rather than a cluster with a standing account-request path. No scheduler, queue structure or allocation policy is published. A law faculty member with a GPU workload would need to look outside Pepperdine — the national NSF ACCESS route tagged on this row is the realistic option."
  },
  {
    "state": "California",
    "school": "Dale E. Fowler School of Law",
    "university": "Chapman University",
    "standalone": false,
    "tier": "older",
    "tierLabel": "Older-gen GPUs confirmed (V100/T4/P100/K-series/etc.)",
    "gpu": "24 NVIDIA Tesla GPUs, which Chapman's cluster summary describes as 'A100/V100'. The itemized server list on the same page names only V100s: one 8x V100 SXM2 node, two 5x V100 nodes and two AMD nodes with 3 V100s each.",
    "slurm": "Likely — the Keck cluster page links a SLURM Scheduler Quickstart Guide and SLURM video tutorials, though it never states the scheduler in prose",
    "hpcName": "Keck Computational Research Cluster — IS&T Research Computing; plus Chapman SRE",
    "specs": "Over 2,560 Intel Xeon and AMD EPYC compute cores, 12.8TB of RAM, and direct access to all-flash and disk-storage SAN arrays. Located in the 140,000 sq ft Keck Center for Science & Engineering, opened 2018. Chapman can also instantiate up to 1,000 additional on-demand cloud nodes in AWS, Azure or Google, and can reserve dedicated physical, virtual or cloud instances for individual researchers or projects.",
    "notes": "Re-verified Sept 2026 — modern CONFIRMED and the '24 Tesla A100/V100' figure in this row is EXACT, matching Chapman's own Keck Computational Research Cluster page word for word. Access and management run through the University's IS&T Research Computing team, which describes itself as mainly supporting faculty members and graduate students needing HPC for applications specific to their academic disciplines — a phrasing that neither excludes nor specifically invites law. NOTABLE FOR A LAW AUDIENCE: Chapman operates a Secure Research Environment on Microsoft Azure supporting NIST 800-171, CMMC, HIPAA and Controlled Unclassified Information, tailored per project with security logging, MFA and encryption, and explicitly supporting HPC, machine learning and AI use cases. Along with Emory's HIPAA-compliant HyPER C3 and Michigan's Armis2, that puts Chapman in the small group of institutions in this dataset where a researcher can compute on regulated or confidential material — directly relevant to legal work involving protected records. DISAMBIGUATION: Chapman has two units named for Dale E. Fowler. This row is the Dale E. Fowler School of LAW. The Keck Center houses Schmid College of Science and Technology and the Dale E. and Sarah Ann Fowler School of ENGINEERING — a different school, same donor. CORRECTED Oct 2026: tier modern → older. Chapman's summary sentence ('24 Tesla A100/V100') is contradicted by the itemized server list on the same page, which accounts for all 24 GPUs as V100s (see caveat). The cluster page also links SLURM guides. The Secure Research Environment is billed per project: Chapman's budget examples run from $2,500 (small) to $12,000+ (large).",
    "networks": [],
    "sources": [
      "https://www.chapman.edu/campus-services/information-systems/research-technology-support/computational-resources/keck-cluster.aspx",
      "https://www.chapman.edu/campus-services/information-systems/research-technology-support/computational-resources/index.aspx",
      "https://blogs.chapman.edu/information-systems/2024/05/01/chapman-universitys-research-computing-services-and-resources/",
      "https://en.wikipedia.org/wiki/Chapman_University",
      "https://web.archive.org/web/20240404195038/https://www.chapman.edu/campus-services/information-systems/research-technology-support/computational-resources/keck-cluster.aspx"
    ],
    "caveat": "Rated on the itemized hardware. Chapman's one-sentence cluster summary mentions 'A100/V100', but the itemized server list on the same page accounts for all 24 GPUs as V100s (8 + 2x5 + 2x3), and before 2025 the page listed just 16 V100s. If A100s have been added, this row belongs in modern; a law researcher who needs A100-class GPUs should confirm with Chapman's Research Computing group first."
  },
  {
    "state": "California",
    "school": "Loyola Law School Los Angeles",
    "university": "Loyola Marymount University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "None on-prem; cloud-based only",
    "slurm": "No",
    "hpcName": "No dedicated on-prem SLURM/GPU HPC cluster found",
    "specs": "",
    "notes": "Re-verified Sept 2026 — LMU has a \"Research Computing Services\" team, but its own page directs researchers to commercial cloud providers (Google Cloud, AWS via a discounted Internet2 account, CloudBank for NSF grants) rather than describing any on-premise GPU cluster. This is consulting/brokering support, not a dedicated HPC resource, so \"no HPC resource identified\" still applies — but it is a genuinely different situation from a school with no research-computing support at all. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://its.lmu.edu/whatwedo/researchcomputing/",
      "https://its.lmu.edu/whatwedo/researchcomputing/systemmanagementcloudcomputing/"
    ]
  },
  {
    "state": "California",
    "school": "University of San Diego School of Law",
    "university": "University of San Diego",
    "standalone": false,
    "tier": "cpuonly",
    "tierLabel": "No GPU confirmed (CPU/RAM-only)",
    "gpu": "None confirmed — SABER1 (USD's only documented central HPC cluster) is CPU-only: Intel Xeon E5-2630 v3, 18 nodes, 288 cores total, no GPU hardware listed anywhere in the technical specification",
    "slurm": "No (TORQUE/MOAB)",
    "hpcName": "SABER1",
    "specs": "SABER1: 18 nodes (17 compute + 1 master), 288 cores, 1,088GB cluster memory, TORQUE/MOAB scheduler (not SLURM)",
    "notes": "Re-verified Sept 2026 by following USD's own site navigation to its dedicated \"Specifications\" page (not just the FAQ) — same stale 2015/2016 SABER1 CPU-only cluster confirmed both places. Moved from \"undetailed\" to \"no GPU confirmed\" since the spec sheet is actually complete, it just doesn't include a GPU. CORRECTED Oct 2026: USD's spec sheet names the scheduler as TORQUE/MOAB, so SLURM is ruled out rather than unconfirmed. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://sites.sandiego.edu/hpc/faq/",
      "https://sites.sandiego.edu/hpc/technical-specifications/"
    ],
    "caveat": "USD's HPC website appears effectively abandoned — every dated page found (FAQ, technical specs) was last substantively updated in 2016-2017, and both list the exact same 2015-era SABER1 hardware. This could mean USD genuinely has no newer central GPU resource, or it could mean they have one that was never documented publicly. Distinct from a normal 'undetailed' case: this is a fully detailed spec sheet that explicitly shows no GPU, not an absence of detail."
  },
  {
    "state": "California",
    "school": "University of San Francisco School of Law",
    "university": "University of San Francisco",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "No dedicated institution-wide HPC/SLURM resource found",
    "specs": "",
    "notes": "Re-verified Sept 2026 with a search explicitly excluding UCSF (the much larger, unrelated UC San Francisco medical campus, which otherwise dominates search results for this name) — USF is a private Jesuit university, master's-focused in CS (no PhD program), and no shared campus HPC/GPU cluster was found on its CS faculty pages or elsewhere. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.usfca.edu/arts-sciences/programs/graduate/computer-science/faculty",
      "https://en.wikipedia.org/wiki/University_of_San_Francisco"
    ]
  },
  {
    "state": "California",
    "school": "McGeorge School of Law",
    "university": "University of the Pacific",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "No dedicated public HPC/SLURM resource found",
    "specs": "",
    "notes": "Re-verified Sept 2026 — \"Pacific Technology\" (the university's IT department) explicitly describes itself as focused on wireless, classroom tech, and student computer labs (16 PCs, 4 Macs in the library); no shared research-computing or HPC resource found even on the School of Engineering and Computer Science pages. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.pacific.edu/about-pacific/administrative-offices/pacific-technology",
      "https://www.pacific.edu/engineering"
    ]
  },
  {
    "state": "California",
    "school": "Golden Gate University School of Law",
    "university": "Golden Gate University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026 — GGU is a small, professional-graduate-focused institution (business, law, taxation, psychology; founded 1901) with no STEM PhD programs. Its \"Applied and Agentic AI\" doctorate is a practice-oriented professional program for working technologists, not a research lab; no shared HPC/GPU infrastructure found anywhere. UPDATED Oct 2026: the ABA lists Golden Gate as on a teach-out plan. Its approved plan (Feb 2024) ends the J.D. program: no new students, no credit-bearing courses after summer 2024, and ABA approval continuing only until July 1, 2027 so remaining students can finish elsewhere and receive the degree. Kept in the directory because the ABA still lists it. UPDATED Oct 2026: GGU has not left the J.D. business for good. In October 2025 it asked the State Bar of California's Committee of Bar Examiners to let the law school restart as a California-accredited, non-ABA school; this check did not find the outcome. The ABA teach-out itself still runs only until July 1, 2027. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://catalog.ggu.edu/preview_program.php?catoid=9&poid=886",
      "https://ggu.edu/programs/applied-and-agentic-ai-dtech/",
      "https://www.americanbar.org/groups/legal_education/accreditation/approved-law-schools/additional-approvals/",
      "https://dev.americanbar.org/content/dam/aba/administrative/legal_education_and_admissions_to_the_bar/PublicNoticeAnnouncements/2024/golden-gate-u-notice-teach-out-plan-february-2024.pdf",
      "https://www.abajournal.com/web/article/golden-gate-law-school-aims-to-reopen-with-california-accreditation"
    ]
  },
  {
    "state": "California",
    "school": "UC College of the Law San Francisco (fka UC Hastings)",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "",
    "specs": "",
    "notes": "RECLASSIFIED Sept 2026 — despite carrying the \"University of California\" name, UC Law SF (formerly Hastings) is not directly controlled by the UC Board of Regents and does not share a campus, governance, or infrastructure with any other UC campus; it was chartered in 1878 as an unusually autonomous entity within the UC system. This is structurally closer to this directory's standalone schools than to a law school with a true shared-resource parent university, so it moves there rather than staying in \"no resource identified,\" which implied a parent university that simply hadn't been checked thoroughly. UC Law SF does partner with UCSF, Berkeley, Davis, and UC Santa Cruz on specific joint-degree programs, but no evidence found that this extends to shared HPC/GPU infrastructure. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.magnifiedmedia.net/uc-hastings-college-of-the-law/",
      "https://www.uclawsf.edu/",
      "https://www.uclawsf.edu/our-story/board-of-directors/"
    ]
  },
  {
    "state": "Colorado",
    "school": "University of Colorado Law School",
    "university": "University of Colorado Boulder",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Alpine's Slurm resource types now include a100, l40, mi100, H200 and RTX Pro 6000, with dedicated partitions --partition=ah200, --partition=artxpro6000 and --partition=gh200 alongside aa100, al40 and ami100. CU Boulder's facilities statement records 23 compute nodes carrying multiple GPUs each (NVIDIA L40, NVIDIA A100, AMD MI100) plus 2 NVIDIA GRACE-HOPPER nodes. Both H200 and Grace Hopper are frontier-tier parts under this project's table.",
    "slurm": "Yes — Slurm with gres directives (--gres=gpu:<type>:N) and QoS tiers extending walltime to 7 days",
    "hpcName": "CU Boulder Research Computing — Alpine (+ Blanca condo, The Preserve secure enclave)",
    "specs": "406 standard CPU compute nodes, 24 high-memory nodes, 23 multi-GPU nodes and 2 Grace-Hopper nodes. Mellanox InfiniBand at 200 Gb/s; 3PB high-performance IBM GPFS. RedHat Enterprise Linux 8. Alpine is funded jointly by CU Boulder, Colorado State University, CU Anschutz and the NSF, and CU Anschutz's contribution alone adds 16 AMD Milan compute nodes, 2 nodes at 1TB RAM and 4 GPU nodes of 3x A100. Research Computing also runs Blanca, a 229-node condo cluster, and The Preserve, a secure computing environment.",
    "notes": "RECLASSIFIED Sept 2026, modern -> frontier. Like Minnesota and Indiana, this row was an understatement rather than an overclaim: it recorded Alpine's A100, L40 and MI100 hardware accurately, but Alpine has since grown H200 and Grace-Hopper capacity, which CU Boulder's own quick-start documentation now exposes as first-class Slurm feature tags and partitions (ah200, gh200, artxpro6000). Two access points are notably good. First, CU Boulder states that ALL NODES ARE AVAILABLE TO ALL USERS, including hardware contributed by partner campuses, governed by a published FairShare policy rather than by who paid. Second, Alpine is explicitly hosted for students, staff and researchers at CU Boulder AND PARTNER INSTITUTIONS, the partners being Colorado State and CU Anschutz. Research Computing also provides training workshops and consultation as part of its educational mission, and offers compile/test partitions (acompile, atesting) that give near-immediate access for trying things out — a practical on-ramp for a researcher outside the usual HPC disciplines.",
    "networks": [],
    "sources": [
      "https://curc.readthedocs.io/en/latest/clusters/alpine/alpine-hardware.html",
      "https://curc.readthedocs.io/en/latest/clusters/alpine/quick-start.html",
      "https://www.colorado.edu/rc/media/110",
      "https://colorado.edu/rc/sites/default/files/attached-files/facilities_statement.pdf",
      "https://research-computing-user-tutorials.readthedocs.io/en/stable/clusters/alpine/quick-start.html"
    ]
  },
  {
    "state": "Colorado",
    "school": "University of Denver Sturm College of Law",
    "university": "University of Denver",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "2 nodes w/ 1x V100 each; 4 nodes w/ 4x A100 80GB each; 2 nodes w/ 8x RTX PRO 6000 96GB each",
    "slurm": "Yes",
    "hpcName": "Research Data Analysis Cluster (RDAC)",
    "specs": "24 compute nodes; 1TB user storage quota (Rocky Linux 9)",
    "notes": "Re-verified Sept 2026 — modern CONFIRMED. This row matters more than most because DU's Research Data Analysis Cluster is the worked example used in this site's own footer and in CONTRIBUTING.md's field schema, so an error here would propagate into every contributor's mental model. It holds up. DU IT states that the University of Denver hosts a High Performance Computing cluster available to faculty members, researchers and students engaged in research activities, with the stated goal of providing the computational resources needed to construct, analyse and interpret complex data — no disciplinary restriction, which is what makes the Sturm College of Law example in the footer legitimate rather than aspirational. On the hardware: the A100 80GB nodes place this firmly in the modern tier, and the RTX PRO 6000 96GB nodes stay modern too rather than promoting the row, because CONTRIBUTING.md's tier table lists RTX PRO 6000 under modern — it is a large-memory workstation-class part rather than a datacentre training GPU. Same classification logic applied to Michigan's Armis2.",
    "networks": [],
    "sources": [
      "https://www.du.edu/it/services/research-services/research-computing"
    ]
  },
  {
    "state": "Connecticut",
    "school": "Yale Law School",
    "university": "Yale University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "80 NVIDIA H200 GPUs on Bouchet plus 48 A5000 Ada, with a b200 Slurm constraint also documented on the Bouchet gpu partition; 32 H200 on Hopper (for NIST 800-171 / HIPAA / sensitive data); 12 H100 on Milgram. A100 80GB, L40S, RTX 5000 Ada and A5000 cards appear across partitions. All YCRC GPU nodes carry four GPUs except the H200 nodes, which carry eight.",
    "slurm": "Yes — Slurm across all YCRC clusters, with a Priority Tier 'fast-lane' since Dec 2024",
    "hpcName": "Yale Center for Research Computing (YCRC) — Bouchet (primary), Hopper, Milgram, Misha",
    "specs": "Bouchet: ~10,000 direct-liquid-cooled cores at MGHPCC, growing. Multiple PB of shared storage. Open OnDemand web portal.",
    "notes": "UPDATED Sept 2026 — the cluster lineup has changed materially and the old row was going stale. Grace and McCleary are being decommissioned in 2026 and have been replaced by Bouchet, YCRC's first installation at the MGHPCC datacenter in Holyoke; new Grace accounts are now only approved for groups with dedicated nodes. Bouchet opened to all Yale researchers in July 2025 and is the general-purpose cluster for low-risk data, which is the relevant one for a law faculty member. The 80 H200 GPUs came from the Provost's AI Initiative following an AI Task Force recommendation that the university invest in high-end GPUs, so further expansion is likely. Anyone re-checking this row should look at Bouchet, not Grace.",
    "networks": [
      "MGHPCC (Massachusetts consortium)"
    ],
    "sources": [
      "https://docs.ycrc.yale.edu/news",
      "https://docs.ycrc.yale.edu/clusters/bouchet/",
      "https://docs.ycrc.yale.edu/ai/resources/",
      "https://research.computing.yale.edu/ycrc-computing-systems",
      "https://docs.ycrc.yale.edu/clusters/grace/"
    ]
  },
  {
    "state": "Connecticut",
    "school": "University of Connecticut School of Law",
    "university": "University of Connecticut",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "NVIDIA A100 confirmed (general-gpu partition, epyc64+a100 nodes); separate gpu_v100, gpu_gtx, gpu_rtx partitions for older cards; 135 GPUs total",
    "slurm": "Yes",
    "hpcName": "UConn Storrs HPC (ITS)",
    "specs": "20,000+ cores across 400 nodes; 220TB scratch + 3.65PB persistent storage (WekaIO)",
    "notes": "Serves all UConn researchers/schools & colleges; Condo model for priority access; SLURM confirmed. CORRECTED Sept 2026 — A100 confirmed present.",
    "networks": [],
    "sources": [
      "https://kb.uconn.edu/space/SH/26032963685/SLURM+Guide",
      "https://hpc.uconn.edu/storrs/cluster-resources/"
    ]
  },
  {
    "state": "Connecticut",
    "school": "Quinnipiac University School of Law",
    "university": "Quinnipiac University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026 — Quinnipiac has an active School of Computing & Engineering with AI/CS research faculty, but its ITS pages describe only standard enterprise services (helpdesk, wifi, print) with no research-computing or HPC/GPU cluster page found across two search angles. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.qu.edu/academics/about-our-programs/areas-of-interest/computing-and-technology/",
      "https://www.qu.edu/schools/engineering/faculty-and-staff/"
    ]
  },
  {
    "state": "District of Columbia",
    "school": "Georgetown University Law Center",
    "university": "Georgetown University",
    "standalone": false,
    "tier": "unspecified",
    "tierLabel": "GPUs present, specific model not named in source",
    "gpu": "GPU allocations available via GCP (types vary by allocation)",
    "slurm": "Yes",
    "hpcName": "HPC@GU (Google Cloud Platform-based)",
    "specs": "Cloud-based elastic compute/storage",
    "notes": "Re-verified Sept 2026 — 'unspecified' is CORRECT, and the reason is worth recording rather than leaving as a shrug. Georgetown runs HPC@GU on Google Cloud Platform, and the access position is unusually good: HPC@GU states that ALL faculty or staff doing research at Georgetown University are eligible for access as Principal Investigators, with other users authorised through an associated PI. That phrasing is university-wide and not discipline-limited, so a Law Center faculty member qualifies on its face. The GPU model cannot be confirmed because the cluster specification pages are themselves access-restricted: HPC@GU states its contents are restricted to Georgetown University members and require a NetID sign-in, so the Current Clusters page returns nothing to an outside reader. This is therefore a documented 'cannot be determined publicly', not an unchecked row — confirming the hardware would need a Georgetown login or a direct question to UIS Research Technologies. Separately, Georgetown's Massive Data Institute provides Google and Microsoft cloud infrastructure specifically for faculty research related to public policy, including Spark clusters, Linux servers, private servers and BigQuery — a route that is unusually well aligned with empirical legal scholarship.",
    "networks": [],
    "sources": [
      "https://hpc.georgetown.edu/",
      "https://hpc.georgetown.edu/current-clusters",
      "https://mdi.georgetown.edu/resources-and-training/enabling-research/compute/",
      "https://uis.georgetown.edu/research/"
    ],
    "caveat": "Cloud-based and quota-metered, with hardware unverifiable from outside. HPC@GU runs on GCP rather than an on-premises cluster, and the site points to Current Allocation and Overage Rates, so costs can be incurred above an allocation. Because the specification pages are behind Georgetown authentication, no public source establishes the GPU generation available — treat any assumption about A100/H100-class hardware here as unverified."
  },
  {
    "state": "District of Columbia",
    "school": "George Washington University Law School",
    "university": "George Washington University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Pegasus now incorporates NVIDIA Grace Hopper 'superchip' nodes (2 in the cluster) and L40S GPUs, alongside 2 eight-GPU nodes. GW describes the Grace Hopper systems as delivering 10x higher performance than the previous generation by combining a Grace CPU with a Hopper GPU. The V100 inventory previously recorded in this row describes the earlier Pegasus build.",
    "slurm": "Yes — Slurm across Pegasus; Raptor uses HTCondor instead (Open Science Grid based)",
    "hpcName": "Pegasus HPC Cluster",
    "specs": "Pegasus: 200+ nodes (about 210 compute), Dell R740 and C4140, reachable through 4 high-availability login nodes with 2 HA head nodes, 2.14 PFLOPS single precision. Node breakdown includes 54 medium-memory, 2 high-memory, 6 high-throughput, 2 eight-GPU and 2 Grace Hopper nodes. Lenovo DSS scratch with 2PB. GW also runs Cerberus (a teaching cluster mirroring Pegasus technologies at smaller scale) and Raptor (HTCondor high-throughput computing on the Open Science Grid). Networking via CAAREN, GW's Capital Area Advanced Research and Education Network, connecting the DC region to Internet2.",
    "notes": "UPDATED Sept 2026. This row was internally inconsistent — scored 'frontier' while its hardware field listed only NVIDIA Tesla V100, which is the 'older' tier by this project's own table. Checking resolved it in favour of the TIER, not the hardware string: GW IT states that the Pegasus expansion implements the NVIDIA Grace Hopper superchip and L40S GPUs, and Grace Hopper is explicitly a frontier-tier part. The V100 text was simply describing the pre-expansion cluster and has been replaced. Access is among the more clearly stated in this dataset: GW IT says HPC resources are available to all GW researchers, faculty and students, and describes open access to the university community for research, academic classroom use and student projects, with no discipline restriction. GW reports over $350M in annual research expenditures. Note for anyone re-checking: GW has three or four clusters and the documentation is spread across it.gwu.edu, hpc.gwu.edu and the legacy colonialone.gwu.edu, with OS versions reported inconsistently (CentOS 7.4 in older material, CentOS 8 in newer).",
    "networks": [],
    "sources": [
      "https://it.gwu.edu/hpc-pegasus",
      "https://it.gwu.edu/hpc-expansion",
      "https://it.gwu.edu/hpc",
      "https://colonialone.gwu.edu/",
      "https://hpc.gwu.edu/about-hpc-at-gw/"
    ],
    "caveat": "Law is not among the contributing schools. GW states Pegasus is professionally managed by GW Information Technology 'with additional support from Columbian College of Arts and Sciences, School of Engineering and Applied Science, School of Public Health, and School of Medicine and Health Sciences' — the Law School is not in that list of school-level contributors. Unlike Harvard's FASRC, GW's access language is explicitly university-wide, so this is a note about where the embedded support staff sit rather than an access barrier; a law researcher may simply have less local expertise to draw on. Also note Pegasus storage is explicitly not for long-term use and backups are the user's responsibility."
  },
  {
    "state": "District of Columbia",
    "school": "American University Washington College of Law",
    "university": "American University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "One node with dual NVIDIA H100 GPUs (80GB each) and six nodes with quad NVIDIA L40 GPUs (40GB) — 2 H100 and 24 L40 in total, matching AU's published specification exactly.",
    "slurm": "Uncertain (LSF-like scheduler mentioned; not confirmed SLURM)",
    "hpcName": "Lovelace HPC System",
    "specs": "Red Hat Enterprise Linux. One 2TB big-memory node. Every compute node carries 512GB RAM and dual-socket Intel Xeon Platinum 8462Y+ CPUs at 32 cores per socket. Storage: 500GB per user home directory, with faculty able to request up to a 1TB project directory for collaborative work. Software includes OpenMPI, OpenBLAS, R, Anaconda/Miniconda Python, Mathematica, Stata MP, and a broad MATLAB licence whose toolboxes include Econometrics, Statistics and Machine Learning, and an unlimited 32-worker MDCS licence.",
    "notes": "Re-verified Sept 2026 — frontier CONFIRMED and the figures in this row are EXACT, matching AU's own system specifications page verbatim. Access is one of the cleanest statements in this entire dataset: AU says plainly that 'Lovelace supports research projects in ALL ACADEMIC UNITS', and the Washington College of Law is listed among AU's schools, so a law faculty member is unambiguously in scope. Accounts are requested through a single ServiceNow form rather than a proposal process. Worth flagging for empirical legal scholarship specifically: Lovelace's licensed software stack includes Stata MP and MATLAB's Econometrics and Statistics & Machine Learning toolboxes, which is a better fit for law-and- economics or quantitative legal work than the molecular-dynamics-heavy stacks typical of larger clusters. Historical note for anyone searching older material: Lovelace succeeded an earlier AU system called Zorro, whose name still appears in page metadata and in the HPC team site URL.",
    "networks": [],
    "sources": [
      "https://www.american.edu/cas/hpc/software.cfm",
      "https://www.american.edu/cas/hpc/",
      "https://www.american.edu/cas/hpc/procedures.cfm"
    ],
    "caveat": "Small and administratively housed in Arts & Sciences. This is a modest system — two H100s, not a partition of them — so a large training run is out of scope even though the tier is frontier. The HPC service also sits under the College of Arts & Sciences (its pages live at american.edu/cas/hpc), so while the 'all academic units' language is explicit, day-to-day support is CAS-based rather than centrally run."
  },
  {
    "state": "District of Columbia",
    "school": "Howard University School of Law",
    "university": "Howard University",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "No general-access campus GPU cluster found across three separate searches; Howard researchers instead use NSF ACCESS-allocated national resources (SDSC, Pittsburgh Supercomputing Center) for HPC-scale work",
    "slurm": "Unconfirmed",
    "hpcName": "No named general-access campus cluster identified; RITA (UARC) has specialized HPC",
    "specs": "A dedicated HPC resource exists for the Research Institute for Tactical Autonomy (RITA), a University Affiliated Research Center — but this is defense-research infrastructure tied to a specific institute, not a general campus cluster a law researcher could access",
    "notes": "Re-verified Sept 2026 across three distinct search angles (general HPC/GPU, HBCU-specific coverage, Howard IT/research-computing pages) — found no publicly documented, general-access campus GPU cluster with its own specs page. What is documented: (1) Howard researchers publish work run on national ACCESS-allocated systems like SDSC and PSC rather than a campus-owned cluster, and (2) a specialized HPC role exists for RITA, a defense-focused University Affiliated Research Center, which is restricted infrastructure rather than a general campus resource. Distinguishing a school's own dedicated cluster from generic ACCESS/XSEDE eligibility (which is available to nearly any US research institution) matters here — this entry does not credit the latter as campus infrastructure.",
    "networks": [],
    "sources": [
      "https://access-ci.org/howard-university-oak-ridge-national-lab-use-access-resources-to-study-crystal-structures-of-solids/",
      "https://howard.wd1.myworkdayjobs.com/en-US/HU/job/High-Performance-Computing-Engineer_JR104450-1"
    ]
  },
  {
    "state": "District of Columbia",
    "school": "Columbus School of Law",
    "university": "Catholic University of America",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026 — CUA's central Technology Services is thoroughly documented but is enterprise/administrative IT (network, virtualization for instructional use, helpdesk); its own infrastructure page explicitly says it \"provides guidance to schools and academic departments needing to run servers... for research purposes\" rather than describing a central shared HPC/GPU cluster — meaning any research computing at CUA is decentralized/departmental, not a shared resource this directory can point to. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://technology.catholic.edu/services/infrastructure.html",
      "https://technology.catholic.edu/about/whatwedo.html"
    ]
  },
  {
    "state": "District of Columbia",
    "school": "David A. Clarke School of Law",
    "university": "University of the District of Columbia",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026 across two angles (central IT services, and the School of Engineering & Applied Sciences' PhD in Computer Science and Engineering page) — UDC does have real doctoral CS/engineering research, but no shared campus HPC or GPU cluster is described anywhere; IT services are enterprise-only (helpdesk, network, Banner). || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.udc.edu/about/administration/it/about-our-services/",
      "https://www.udc.edu/seas/phd-in-computer-science-and-engineering/"
    ]
  },
  {
    "state": "Florida",
    "school": "Fredric G. Levin College of Law",
    "university": "University of Florida",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "504 NVIDIA B200 'Blackwell' GPUs — 63 NVIDIA DGX B200 nodes at 8 GPUs each, forming the first DGX B200 SuperPOD at a US university. Also 600 NVIDIA L4 GPUs and A100/RTX6000 cards elsewhere in HiPerGator.",
    "slurm": "Yes",
    "hpcName": "HiPerGator (UFIT Research Computing)",
    "specs": "~60,000 CPU cores (HPG4); 11PB all-flash parallel filesystem",
    "notes": "Re-verified Sept 2026 and the headline number checks out exactly: 63 DGX B200 nodes x 8 GPUs = 504 Blackwell GPUs, a $33M upgrade unveiled at UF East Campus on 14 October 2025 after an early-access phase from June 2025. Note one nuance for anyone updating this row: the B200 SuperPOD REPLACED the 2020-era NVIDIA DGX A100 'Ampere' SuperPOD rather than adding to it, so A100 references should be read as applying to other parts of HiPerGator, not the SuperPOD. UF reports 8,000+ users and 33 million research requests in a year, with more than 60% of university research projects relying on HiPerGator. Access is open to UF faculty across every academic discipline, free for teaching with paid tiers for research.",
    "networks": [],
    "sources": [
      "https://www.hpcwire.com/off-the-wire/university-of-florida-expands-ai-leadership-with-new-hipergator-blackwell-supercomputer/",
      "https://news.ufl.edu/2025/05/the-fourth-generation-model-of-ufs-hipergator/",
      "https://news.ufl.edu/2025/01/fastest-academic-supercomputer-arrives-at-uf/"
    ]
  },
  {
    "state": "Florida",
    "school": "Florida State University College of Law",
    "university": "Florida State University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "NVIDIA H100 Tensor Core GPU (4x, 2/node) — limited/pilot-program access, request required; RTX A4500 Ada Generation (64x total across two pools); RTX A4000 (16x); GeForce GTX 1080 Ti (28x, legacy)",
    "slurm": "Yes",
    "hpcName": "FSU Research Computing Center (RCC) HPC cluster",
    "specs": ">18,000 cores, >700 nodes; InfiniBand interconnect; SLURM confirmed (sinfo/sbatch documented); GPU nodes cost $6,000/5yr to buy in, or free via backfill2/gpu_q general-access queues (4-hr limit unless approved)",
    "notes": "Available to faculty and departments (paid buy-in, or free general-access queues with time limits); RCC documentation includes exact `sinfo` output naming every GPU model and node. CORRECTED Sept 2026 — confirmed by following the \"Compute GPU\" service page to its underlying RCC technical docs (docs.rcc.fsu.edu/hpc/gpus), which the ITS overview page alone did not show.",
    "networks": [],
    "sources": [
      "https://docs.rcc.fsu.edu/hpc/gpus/",
      "https://its.fsu.edu/services/compute-gpu",
      "https://its.fsu.edu/services/high-performance-compute-cluster"
    ],
    "caveat": "The H100s are explicitly gated as \"Limited access for pilot programs; Request access\" per FSU's own GPU documentation — unlike this directory's other Frontier-tier schools, H100 access here is not yet general. The RTX A4500/A4000 pools are the realistic day-one access point for most researchers via the backfill2 queue."
  },
  {
    "state": "Florida",
    "school": "FIU College of Law",
    "university": "Florida International University",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "\"GPU nodes\" confirmed present and current (live 2026-copyrighted page) but no specific model named anywhere on IRCC's public site. Historical record: 2013 news posts describe 2, then 6 GPU nodes added with NVIDIA Tesla K20s; a 2020 facilities PDF describes a 12-node GPU cluster (model unnamed) — that hardware is almost certainly long since refreshed, so it is not used here as the current answer.",
    "slurm": "Yes",
    "hpcName": "IRCC Panther Cluster",
    "specs": "Panther Cluster: 3,000+ Intel cores, up to 1.5TB RAM/node, 500TB+ parallel storage, 56Gbps InfiniBand, SLURM confirmed",
    "notes": "Re-verified Sept 2026 across three angles (general search, SLURM/partition-specific search, and a full fetch of the live HPC page itself to rule out a truncated snippet) — the current page genuinely just says \"GPU nodes\" with no model, confirming this is a real information gap on FIU's end rather than a search miss. Free to FIU faculty/students via hpcadmin@fiu.edu request.",
    "networks": [],
    "sources": [
      "https://ircc.fiu.edu/hpc/",
      "https://ircc.fiu.edu/gpu-expansion/",
      "https://ircc.fiu.edu/wp-content/uploads/2020/10/IRCCFacilities.pdf"
    ]
  },
  {
    "state": "Florida",
    "school": "University of Miami School of Law",
    "university": "University of Miami",
    "standalone": false,
    "tier": "older",
    "tierLabel": "Older-gen GPUs confirmed (V100/T4/P100/K-series/etc.)",
    "gpu": "NVIDIA Tesla V100 (Volta), 2 per node, across 96 IBM Power Systems AC922 nodes on Triton",
    "slurm": "Not confirmed",
    "hpcName": "Frost Institute for Data Science & Computing (IDSC) - Pegasus + Triton clusters",
    "specs": "Triton: 96 AC922 nodes, each with two 20-core POWER9 processors and two V100 GPUs; Red Hat Enterprise Linux 8.6. Pegasus: 340 iDataPlex nodes (Intel Xeon E5-2670), later augmented with 1,024 cores and 6TB memory plus 1PB IBM ESS storage — CPU-only. 10+ PB high-performance storage; 500+ users. Hosted at the NAP of the Americas datacenter in downtown Miami.",
    "notes": "Re-verified Sept 2026 — 'older' is CORRECT and was worth confirming rather than assuming decay. IDSC's current cluster documentation still lists Triton as IBM Power9 with NVIDIA Tesla V100 GPUs at 2 per node, and Pegasus remains a CPU-only Intel Xeon system. Triton was rated a top-5 US academic institution supercomputer at purchase in 2019 and was UM's first GPU-accelerated HPC system, but it has not been refreshed to Ampere or Hopper class. IDSC references a 'planned investment' to enhance its infrastructure, so this row is a good candidate for re-checking. Access is by IDSC account with per-service-unit charging; the scheduler is still not named as Slurm in public docs.",
    "networks": [],
    "sources": [
      "https://acs-docs.readthedocs.io/clusters/general-usage/1-intro.html",
      "https://idsc.miami.edu/magazine/enhancing-the-universitys-computing-capabilities/",
      "https://idsc.miami.edu/triton/",
      "https://it-resources.research.miami.edu/services/high-performance-computing/index.html"
    ]
  },
  {
    "state": "Florida",
    "school": "Nova Southeastern University Shepard Broad College of Law",
    "university": "Nova Southeastern University",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "No GPU model named anywhere across the main HPC page and the dedicated Sphyrna (research) sub-page — both are thorough on policy/access/SLURM but publish no hardware specs at all",
    "slurm": "Not confirmed",
    "hpcName": "Selachii (academic) + Sphyrna (research)",
    "specs": "Two named SLURM clusters: Selachii (academic/teaching) and Sphyrna (research); Red Hat Enterprise Linux 9; SCP/SFTP data transfer",
    "notes": "Re-verified Sept 2026 — followed the site's own navigation into the dedicated Sphyrna (research cluster) sub-page specifically looking for a hardware/specs section; found detailed OS/access/policy info but no GPU (or even CPU) hardware specs published anywhere. NSU Faculty can sponsor accounts (students permitted on Selachii, not on Sphyrna).",
    "networks": [],
    "sources": [
      "https://www.nova.edu/oiit/high-performance-computing/index.html",
      "https://www.nova.edu/oiit/high-performance-computing/sphyrna/index.html"
    ]
  },
  {
    "state": "Florida",
    "school": "Barry University School of Law",
    "university": "Barry University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026 — Barry's CS program is undergraduate-focused with a teaching lab described explicitly as \"virtual machine-based computer security simulation, 3D printing, Linux and Windows servers\" — no GPU/HPC. Law school's own technology page describes only a Windows computer lab and library printing/copying. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.barry.edu/en/academics/arts-and-sciences/break-through-in-computer-science/",
      "https://www.barry.edu/en/academics/law/technology/"
    ]
  },
  {
    "state": "Florida",
    "school": "St. Thomas University School of Law",
    "university": "St. Thomas University (FL)",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026 — STU's AI initiatives (AiREN, Digital Education Council membership) are explicitly ethics/policy focused, not technical infrastructure; small (~6,000 student) teaching-focused Catholic university with no evidence of research-computing infrastructure anywhere. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.stu.edu/stuai/",
      "https://www.stu.edu/science/research/"
    ]
  },
  {
    "state": "Florida",
    "school": "Stetson University College of Law",
    "university": "Stetson University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026 — Stetson's IT department operates only general-access teaching computer labs (Windows 10, standard software); Computer Science is undergraduate-only with a career-prep focus rather than a PhD research program. No HPC/GPU infrastructure found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.stetson.edu/administration/information-technology/computer-labs/",
      "https://www.stetson.edu/other/academics/undergraduate/computer-science.php"
    ]
  },
  {
    "state": "Florida",
    "school": "Florida A&M University College of Law",
    "university": "Florida A&M University",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "No GPU model published. FAMU's Center for Plasma Science and Technology (CePaST) operates \"a computer cluster\" for laser-plasma simulation, quantum control/machine learning, and molecular physics — a real, current, active research center, though department-specific rather than a campus-wide general-access resource. FAMU is also a SSERCA (Sunshine State Education & Research Computing Alliance) member and its researchers separately use national ACCESS allocations (Stampede2 at TACC, Expanse at SDSC) for published computational chemistry work.",
    "slurm": "No",
    "hpcName": "Center for Plasma Science and Technology (CePaST)",
    "specs": "CePaST: 32,000 sq ft facility incl. laser plasma lab; historical (2015) CePaST cluster was a 128-node/256-processor system, department-restricted (3 faculty + a few postdocs/grad students)",
    "notes": "RECLASSIFIED Sept 2026 from \"no resource identified\" to \"undetailed\" — a 2015 planning presentation showing this cluster's history read initially like something defunct, but CePaST's current department page (2026) confirms the center and its compute infrastructure are still active.",
    "networks": [],
    "sources": [
      "https://cst.famu.edu/departments-and-centers/center-for-plasma-science-and-technology/index.php",
      "https://www.slideshare.net/mjgrav2001/famuresearchcomputing",
      "https://www.hpcwire.com/off-the-wire/florida-am-research-team-advances-renewable-energy-studies-using-access-supercomputers/"
    ],
    "caveat": "CePaST's cluster is a department/center-level resource (physics/plasma science), not a general campus-wide facility a law researcher would have standing access to — distinct from schools where the whole university shares one HPC pool."
  },
  {
    "state": "Georgia",
    "school": "Emory University School of Law",
    "university": "Emory University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "A mix of nodes featuring NVIDIA A100 and A10 GPUs; HyPER C3's start guide also documents an L40S partition alongside A100. Resources are upgraded on a regular cadence because the cluster is cloud-provisioned rather than fixed hardware.",
    "slurm": "Yes — Slurm, with a head/logon node and several partitions",
    "hpcName": "HyPER C3 — community cloud HPC cluster on Emory's AWS environment",
    "specs": "Hosted on AWS with FSx for Lustre storage. /users home directories at 50GB, backed up nightly; /scratch on FSx for Lustre at 1TB per user, purged after 60 days and not backed up. Usage governed by community fair-share policies. Emory also runs HySci, a dedicated Science DMZ with data transfer nodes and Globus for large dataset movement, plus separate core facilities (the EICC cluster on RHEL 8 with Slurm, and CSIC nodes carrying P100, A40 and L40S GPUs).",
    "notes": "Re-verified Sept 2026 — modern CONFIRMED. HyPER C3 stands for Hybrid High-Performance Computing Platform for Education and Research Community Cloud Cluster, and it is unusual in this dataset for being genuinely cloud-native rather than an on-premises cluster. Access is excellent: Emory states shared base-level usage is offered AT NO COST to active Emory faculty, academic staff and students, for research and teaching, with a team of HPC engineers handling onboarding and training. A PI requests accounts through an application form. TWO POINTS THAT MATTER PARTICULARLY FOR A LAW AUDIENCE. First, HyPER C3 IS HIPAA COMPLIANT and designed to protect patient health information — one of only a handful of rows in this dataset where regulated data is permitted rather than forbidden, and the direct opposite of USC's position. Second, the cluster underpins Emory's AI.Humanity initiative, which is explicitly framed around AI across disciplines rather than AI in engineering, so the institutional posture is welcoming to non-STEM computational work.",
    "networks": [],
    "sources": [
      "https://it.emory.edu/catalog/cloud-services/hyper-c3.html",
      "https://aihumanity.emory.edu/resources/computing-infrastructure.html",
      "https://it.emory.edu/research-services/index.html",
      "https://yanglab-emory.github.io/assets/ComputationSlides/HyPER_C3_StartGuide.html",
      "https://aws.amazon.com/blogs/publicsector/emory-university-supports-ai-humanity-initiative-with-high-performance-computing-on-aws"
    ],
    "caveat": "Free tier is capped, and cloud means metering. The maximum is 500 GPU hours per user on the A100 and L40S partitions, enforced for fair access. /scratch is purged after 60 days and cannot be administratively restored. Accounts inactivate if unused for 60 days, with a 30-day grace period. Note that while AWS can supply H100-class hardware and Emory's cloud materials mention it, HyPER C3's documented partitions are A100, A10 and L40S — so do not read general AWS capability as an available Emory partition."
  },
  {
    "state": "Georgia",
    "school": "University of Georgia School of Law",
    "university": "University of Georgia",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "215 GPUs of various models across Sapelo2 (July 2025 figure). A $2.4M 2024 investment added 26 GPU compute nodes: 12 quad-H100, 2 quad-A100 and 12 quad-L4 (48 H100, 8 A100, 48 L4), plus three 3TB high-memory nodes, operational from the fall 2024 semester. Older V100 nodes remain in the mix. The separate teaching cluster carries 2 quad-A30 nodes.",
    "slurm": "Yes — Slurm (sbatch, squeue, sacct) on Rocky Linux 9.5, with Open OnDemand and Globus",
    "hpcName": "Georgia Advanced Computing Resource Center (GACRC) - Sapelo2 cluster",
    "specs": "Sapelo2: roughly 46,000–51,060 cores and 268.1TB total memory, spanning general-purpose nodes (128–512GB RAM) and high-memory nodes (1TB, 2TB, 3TB), EDR InfiniBand in a 2:1 fat tree, over 1,400 software packages. Storage: Aeon hybrid Lustre appliance (2.4PB flash + 9.6PB spinning) providing /scratch (no quota, 30-day purge) and /work (500GB per group); /home 200GB per user with backup; /project 1TB per group with backup; /lscratch local NVMe up to 3.5TB on GPU nodes. Housed at the Boyd Data Center on campus. Over 120 nodes come from the Faculty Buy-In program.",
    "notes": "Re-verified Sept 2026 — frontier CONFIRMED and the hardware list in this row is accurate. GACRC is a partnership between the Offices of the VP for Information Technology and for Research, and its policy states its resources are for use in direct support of research programs at the University of Georgia, with support also provided for classes teaching computational methods — no disciplinary limit, so a law faculty member is in scope. The access mechanism is a two-step process worth knowing before starting: a UGA faculty member registers a lab as PI through the TeamDynamix service portal, then requests user accounts for group members. GACRC runs a full-time staff of systems administrators and scientific computing consultants and publishes grant-proposal boilerplate, which is a practical convenience for anyone writing UGA computing resources into a funding application.",
    "networks": [],
    "sources": [
      "https://gacrc.uga.edu/systems/index.php",
      "https://gacrc.uga.edu/about/proposal_info.php",
      "https://wiki.gacrc.uga.edu/wiki/Policies",
      "https://news.uga.edu/new-investments-in-high-performance-computing-to-advance-ai-research-efforts-at-uga",
      "https://wiki.gacrc.uga.edu/images/9/99/GACRC_overview_20251117_Anthropology.pdf"
    ]
  },
  {
    "state": "Georgia",
    "school": "Georgia State University College of Law",
    "university": "Georgia State University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "NVIDIA A100 (8x per node on DGX-class node \"arctrddgxa001\"), NVIDIA V100 (2x/node, 13 nodes on gpu-v100/qGPU partitions), NVIDIA P100, NVIDIA A2000, plus RTX-class cards — confirmed via `sinfo`/`arc.hardware` output on three separate ARCTIC wiki job-submission pages",
    "slurm": "Not confirmed",
    "hpcName": "ARCTIC (Advanced Research Computing Technology, Innovation & Collaboration)",
    "specs": "75 million CPU-hours/year + 1.2 million GPU-hours/year available cluster-wide; condominium (buy-in) model; SLURM confirmed",
    "notes": "Free startup allocation for every GSU researcher (capped at 1,000 GPU service-units), condo buy-in for more. CORRECTED Sept 2026 — A100 and V100 confirmed via SLURM command-output pages after the main hardware-specs page proved unfetchable.",
    "networks": [],
    "sources": [
      "https://arctic.gsu.edu/systems-and-services/",
      "https://arcwiki.rs.gsu.edu/en/home/slurm/get-list-of-resources-in-node",
      "https://arcwiki.rs.gsu.edu/en/gpu/gdn/running-gdn-model",
      "https://arcwiki.rs.gsu.edu/en/policy/resource-allocations"
    ],
    "caveat": "ARCTIC's own dedicated 'Hardware Specifications' wiki page exists (arcwiki.rs.gsu.edu/en/home/acids/specifications) but blocks automated fetching (robots.txt); the GPU models above are confirmed instead via three separate job-submission documentation pages on the same wiki that show real sinfo/arc.hardware command output naming A100 and V100 directly."
  },
  {
    "state": "Georgia",
    "school": "Mercer University Law School",
    "university": "Mercer University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026 across two search angles — Mercer IT is entirely enterprise/administrative (helpdesk, MyMercer, Workday, teaching computer labs); the \"Computational Science\" academic track name suggested a possible research center, but no dedicated HPC/GPU infrastructure was found associated with it. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://it.mercer.edu/",
      "https://liberalarts.mercer.edu/academic-programs/majors-and-minors/information-science-and-technology/"
    ]
  },
  {
    "state": "Hawaii",
    "school": "William S. Richardson School of Law",
    "university": "University of Hawaii",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Roughly 147-148 GPUs across 22 GPU nodes, spanning NVIDIA GeForce RTX 2070 at the low end up to NVIDIA L40 at the high end — an Ada Lovelace ceiling, not a Hopper one. Two H100 GPUs do exist on Koa, but they arrived as part of a five-node researcher condo purchase rather than as community hardware. Note published GPU counts drift with condo additions: 139 (late 2023), 148 (post-CC* upgrade), 147 (current cluster wiki).",
    "slurm": "Yes — Slurm Workload Manager with a fair-share algorithm; Open OnDemand available alongside ssh",
    "hpcName": "Koa HPC Cluster (formerly Mana) - ITS Cyberinfrastructure",
    "specs": "275-295 nodes, 7,683-9,312 cores spanning Intel Ivy Bridge through Sapphire Rapids and AMD EPYC Milan, 53-62TB aggregate RAM, 6PB+ storage. Fully online 8 December 2023, succeeding the Mana cluster deployed in late 2014, whose components were folded in. Funded by the University, an NSF MRI award (#1920304) and two NSF CC* awards (#2201428, #2232862) totalling $900,000. Storage: 50GB per user home, 500GB free per lab, Cray ClusterStor E1000 flash scratch, KoaStore Lustre. Heterogeneous networking — all nodes 1GbE, 7 at 25GbE, the newest 76 on HDR100 InfiniBand. Managed with open-source xCAT. UH is also a regional Jetstream 2 host site.",
    "notes": "RECLASSIFIED Sept 2026, frontier -> modern. The row's '148 GPUs' headline is CORRECT and verified in three places including a peer-reviewed ACM paper by the Koa architects — but the tier drawn from it was not. The same paper states the fleet runs 'from Nvidia GeForce RTX 2070 up to Nvidia Tesla L40 GPUs', and L40 is Ada Lovelace, which this project's own tier table places in 'modern', not 'frontier'. The only Hopper hardware found is two H100s included in a five-node condo purchase by a researcher. That is real but it is not a community H100 resource, so scoring the whole row frontier overstated what a law faculty member would find. Access itself is excellent and unchanged: Koa is free to all active UH faculty, staff and students across all ten campuses, supports over 90 disciplines, and needs only a self-paced onboarding to get an account. Community users get uninterrupted access to the 215 University-owned machines and can also run on the 130 PI-owned condo nodes when idle, subject to preemption — which is the only route to those two H100s. Worth noting UH ITS Research Cyberinfrastructure explicitly offers to help researchers apply for national ACCESS allocations, making it one of the few universities in this dataset that actively signposts the national floor.",
    "networks": [
      "University of Hawaii System (Koa)"
    ],
    "sources": [
      "https://dl.acm.org/doi/fullHtml/10.1145/3626203.3670611",
      "https://datascience.hawaii.edu/facilities/",
      "https://uhawaii.atlassian.net/wiki/spaces/HPC/pages/429752517",
      "https://datascience.hawaii.edu/about-koa/",
      "https://uhawaii.atlassian.net/wiki/spaces/HPC/pages/1073840152/"
    ],
    "caveat": "Condo nodes are preemptible, and GPUs are scarce. The 130 PI-owned nodes — including the two H100s — are available to community users only when idle and jobs there can be preempted. UH states directly that 'due to the limited supply of GPUs in Koa, we do not currently lease GPU nodes', so buying guaranteed GPU capacity is not currently an option either. Also note sensitive or regulated data cannot be stored on Koa's standard storage — relevant for law faculty working with confidential case or client material."
  },
  {
    "state": "Idaho",
    "school": "University of Idaho College of Law",
    "university": "University of Idaho",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "IBEST CRC (on-campus): 'several GPU nodes' added 2017-2019, no model named on their own hardware page (dated hardware). Separately, as of April 2025 U of Idaho researchers get access to INL's Lemhi system — but Lemhi's own technical docs describe it as a CPU-only Dell C6420 cluster (dual Xeon Gold 6148), no GPUs mentioned, despite press coverage calling it a powerful 'supercomputer.' INL separately operates Sawtooth (V100 GPUs) and Hoodoo (44x A100), but broad university access to those specific systems (vs. just Lemhi) isn't confirmed in public sources.",
    "slurm": "Not confirmed",
    "hpcName": "IBEST Computational Resources Core (CRC) + INL Lemhi partnership",
    "specs": "IBEST CRC: 1,500+ cores, 8+TB RAM, InfiniBand; Lemhi (INL, shared w/ Boise State + Idaho State): 504 nodes, 20,160 cores, 94.5TB RAM",
    "notes": "Re-verified Sept 2026 across four searches covering the on-campus IBEST CRC, the 2025 Lemhi handoff, and INL's broader HPC portfolio — real GPU nodes exist on campus but are undated/unnamed, and the newly-touted Lemhi partnership appears to be CPU-only compute despite being framed as a major upgrade.",
    "networks": [
      "Idaho INL / Lemhi (statewide)"
    ],
    "sources": [
      "https://researchcomputing.uidaho.edu/high-performance-computing/",
      "https://crc.ibest.uidaho.edu/general/Hardware/",
      "https://docs.c3plus3.org/docs/general/Overviews/Lemhi.html",
      "https://www.uidaho.edu/newsroom/lemhi-supercomputer",
      "https://hpc.inl.gov/"
    ],
    "caveat": "News coverage of the April 2025 Lemhi handoff to Idaho's universities emphasizes dramatically increased compute power, but Lemhi's own technical specification (hosted by the C3+3 collaboration that manages it) lists no GPUs at all — this looks like a case where press framing ('supercomputer,' 'punch above its weight') outran the literal hardware spec. Confirmed GPUs at INL (Sawtooth's V100s, Hoodoo's A100s) are documented, but not confirmed as available to University of Idaho researchers specifically, as opposed to Lemhi alone."
  },
  {
    "state": "Illinois",
    "school": "Northwestern University Pritzker School of Law",
    "university": "Northwestern University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "269 NVIDIA GPU cards across 91 GPU nodes (General Access + Priority Access combined); 200 of those cards are available to General Access users. General Access GPU nodes: 24 nodes x 4 H100 SXM (80GB), 18 nodes x 4 A100 SXM (80GB), 16 nodes x 2 A100 PCIe (40GB).",
    "slurm": "Yes",
    "hpcName": "Quest High-Performance Computing Cluster",
    "specs": "1,200+ compute nodes, 80,000+ cores, 7 login nodes, 22 high-memory nodes; Red Hat Enterprise Linux 8.10; InfiniBand HDR; newest nodes are Intel Emerald Rapids Xeon Platinum 8592+. Serves 3,000+ researchers per year.",
    "notes": "Re-verified Sept 2026 and corrected upward. The previous '200 GPU cards' figure was the General Access subset; Quest's own technical specifications page gives 269 GPU cards on 91 GPU nodes in total. Access is favourable for a law faculty member: Northwestern provides dedicated Quest resources for the entire Northwestern research community at no cost, with roughly half the CPU and most of the GPU nodes reserved for General Access, allocated for one year and renewable. The Genomics Compute Cluster on the same infrastructure is separately open to researchers 'from any Northwestern school'. Short jobs under 4 hours can also backfill onto reserved capacity. This is the Quest example named in the site's own footer.",
    "networks": [],
    "sources": [
      "https://www.it.northwestern.edu/departments/it-services-support/research/computing/quest/specs.html",
      "https://www.it.northwestern.edu/departments/it-services-support/research/computing/quest/",
      "https://www.it.northwestern.edu/departments/it-services-support/research/computing/gpu/",
      "https://rcdsdocs.it.northwestern.edu/systems/quest/user-guide/gpu/gpu.html"
    ]
  },
  {
    "state": "Illinois",
    "school": "University of Chicago Law School",
    "university": "University of Chicago",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Midway3: NVIDIA A100, V100, RTX6000 (11 GPU nodes: 5 RTX6000, 5 V100, 1 A100); Beagle3 (44 nodes): NVIDIA A100 and A40",
    "slurm": "Yes",
    "hpcName": "Research Computing Center (RCC) - Midway2/Midway3/Beagle3",
    "specs": "222+ nodes, 10,560+ cores (Midway3 Intel); 2.2PB GPFS storage; Skyway cloud-burst to AWS/GCP",
    "notes": "Free to all University of Chicago researchers across every division incl. professional schools; SLURM confirmed. CORRECTED Sept 2026 — A100/A40 confirmed present on Midway3 and Beagle3.",
    "networks": [],
    "sources": [
      "https://mdw3.rcc.uchicago.edu/?p=1063",
      "https://git.rcc.uchicago.edu/rcc-public/beagle3/-/wikis/Beagle3-User-Guide"
    ]
  },
  {
    "state": "Illinois",
    "school": "University of Illinois College of Law",
    "university": "University of Illinois Urbana-Champaign",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "NCSA Delta: 400x NVIDIA A100 (40GB) + 400x NVIDIA A40 + 40x more A100 (dense nodes) + 8x NVIDIA H200 (141GB) + 8x AMD MI100; DeltaAI: NVIDIA H100 (allocations measured in H100 GPU-hours)",
    "slurm": "Likely (NCSA systems typically SLURM-based)",
    "hpcName": "NCSA Delta / DeltaAI supercomputers",
    "specs": "Delta: 124 CPU nodes + 200+ GPU nodes; 7PB Lustre; 200Gb/s HPE Slingshot; DeltaAI quadruples AI compute capacity",
    "notes": "World-class NSF-funded resource on campus, SLURM confirmed; UIUC researchers get direct allocation share plus national ACCESS access. CORRECTED Sept 2026 — H200/H100 confirmed present, one of the most powerful academic GPU systems in the US.",
    "networks": [],
    "sources": [
      "https://docs.ncsa.illinois.edu/systems/delta/en/latest/user_guide/architecture.html",
      "https://www.ncsa.illinois.edu/resources-and-services/compute-resources/delta/",
      "https://delta.ncsa.illinois.edu/"
    ]
  },
  {
    "state": "Illinois",
    "school": "UIC John Marshall Law School",
    "university": "University of Illinois Chicago",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Lakeshore cluster: 4x NVIDIA A100 (80GB) per GPU node (2 free \"Chicago Computes\" GPU nodes; more via HPC Partnership/Condo buy-in)",
    "slurm": "Not confirmed",
    "hpcName": "ACER (Advanced Cyberinfrastructure for Education & Research) - Lakeshore cluster + AWS Cloud HPC",
    "specs": "10,304 cores; 57.2TB memory; 3PB storage; 141 compute nodes",
    "notes": "Chicago Computes program gives free A100 GPU-hours to ALL UIC faculty (up to 50 GPU-hrs/yr, renewable), shareable with lab members; serves 350+ researchers across 10 colleges. CORRECTED Sept 2026 — A100 80GB confirmed present and free-tier accessible.",
    "networks": [],
    "sources": [
      "https://acer.uic.edu/services/hpc/lakeshore/",
      "https://acer.uic.edu/chicago-computes"
    ]
  },
  {
    "state": "Illinois",
    "school": "DePaul University College of Law",
    "university": "DePaul University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Data-center clusters with NVIDIA H100 NVL and A100 GPUs",
    "slurm": "Not confirmed",
    "hpcName": "School of Computing HPCC (Jarvis College of Computing and Digital Media)",
    "specs": "Positioned around fine-tuning and running open-weight large language models, deep learning, CUDA programming and coding assistants. Services include hosted containers and VMs, DevOps infrastructure for classes and research, Kubernetes, and Metal-as-a-Service via Chameleon. Based at the Jarvis College of Computing and Digital Media, 243 South Wabash Avenue.",
    "notes": "Re-verified Sept 2026. The hardware claim holds — DePaul's HPCC site confirms data-center clusters with NVIDIA H100 NVL and A100 GPUs, so the frontier tier is correct on hardware. The question for this directory is access, and that is where the row needs its caveat: the cluster lives at hpcc.cs.depaul.edu, is run by the School of Computing within the Jarvis College of Computing and Digital Media, and its published services are framed around computing coursework and research. No university-wide access policy was found. ONE REAL PATHWAY WORTH RECORDING, and it is unusually concrete: DePaul's Cybersecurity Clinic is described by the university as an interdisciplinary collaboration between the School of Computing, the Driehaus College of Business AND THE COLLEGE OF LAW. That is a standing, documented working relationship between the law school and the unit that owns the H100s — a better-evidenced route than most affiliation leads examined in this project, and the obvious place for a DePaul law faculty member to start.",
    "networks": [],
    "sources": [
      "https://hpcc.cs.depaul.edu/",
      "https://www.cdm.depaul.edu/academics/Pages/School-of-Computing.aspx"
    ],
    "caveat": "Departmental resource, not a campus service. This is a School of Computing cluster with a named School of Computing contact, not a central-IT offering with a public account-request path for all DePaul faculty. Same situation as Chicago-Kent's Ocient Computational Center. The frontier tier describes hardware that exists at DePaul; it does not establish that a College of Law researcher can request time on it. Ask via the Cybersecurity Clinic relationship rather than assuming an account."
  },
  {
    "state": "Illinois",
    "school": "Chicago-Kent College of Law",
    "university": "Illinois Institute of Technology",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "",
    "slurm": "Not confirmed",
    "hpcName": "Ocient Computational Center (CS dept); Active Computational Thinking (ACT) Center",
    "specs": "Not published — no node, core or GPU counts on Illinois Tech's public pages",
    "notes": "RECLASSIFIED Sept 2026 (was 'none'). Illinois Tech's College of Computing does operate real computational infrastructure: the Ocient Computational Center is described as housing the backbone of the computer science department, containing the networks, architecture and servers behind its AI, cloud computing and HPC teaching and research. Separately, the Active Computational Thinking (ACT) Center is described as a campus hub that works to physically connect Illinois Tech to regional and national computing facilities and research networks. HPC is also a named CS research area, reinforced by joint faculty appointments with Argonne National Laboratory. Moved off 'none' because a genuine computational center exists; kept at 'undetailed' rather than a GPU tier because nothing public names GPU models or a scheduler, and see the caveat on who can actually use it.",
    "networks": [],
    "sources": [
      "https://www.iit.edu/computing/research/research-centers",
      "https://www.iit.edu/computer-science/research/research-areas/high-performance-and-parallel-computing"
    ],
    "caveat": "Department-level, not confirmed campus-wide. The Ocient Computational Center is described as CS-department infrastructure, and the Argonne tie is a faculty-appointment relationship rather than a general access route. No public page establishes that a Chicago-Kent law faculty member could request an account. Treat as unconfirmed for law until someone checks with Illinois Tech OTS."
  },
  {
    "state": "Illinois",
    "school": "Loyola University Chicago School of Law",
    "university": "Loyola University Chicago",
    "standalone": false,
    "tier": "unspecified",
    "tierLabel": "GPUs present, specific model not named in source",
    "gpu": "Three Linux-based GPU servers + one Windows-based GPU server (6-20 cores each), positioned for Deep Learning (TensorFlow/Theano); no specific GPU model named on public documentation",
    "slurm": "Not confirmed",
    "hpcName": "Research Computing Services - various clusters (ROCKS, Hadoop, GPU servers)",
    "specs": "40TB (ROCKS); 21TB Hadoop HDFS; 2.5PB overall LUC-HSC storage",
    "notes": "Re-verified Sept 2026 with a second, more targeted search — Loyola's own Research Computing Services page still does not name a specific GPU model or chip, even in its most detailed hardware listing. Genuinely small-scale (departmental GPU servers, not a campus HPC cluster with SLURM); staying at \"unspecified\" is accurate here, not a research gap.",
    "networks": [],
    "sources": [
      "https://www.luc.edu/its/rcs/computationalresources/"
    ]
  },
  {
    "state": "Illinois",
    "school": "Northern Illinois University College of Law",
    "university": "Northern Illinois University",
    "standalone": false,
    "tier": "unspecified",
    "tierLabel": "GPUs present, specific model not named in source",
    "gpu": "Metis cluster: hybrid CPU/GPU, 400+ teraflops peak performance; no specific GPU model named across two rounds of searching NIU's own Center for Research Computing and Data pages",
    "slurm": "Not confirmed",
    "hpcName": "Center for Research Computing and Data (CRCD) - Metis cluster",
    "specs": "Not detailed publicly",
    "notes": "Open to all NIU faculty, staff, students and collaborators. Re-verified Sept 2026 with a second targeted search for the Metis cluster's specific GPU model — still not named in any public NIU documentation found. Staying at \"unspecified\" reflects a genuine information gap, not under-research.",
    "networks": [],
    "sources": [
      "https://www.niu.edu/crcd/index.shtml"
    ]
  },
  {
    "state": "Illinois",
    "school": "Southern Illinois University School of Law",
    "university": "Southern Illinois University",
    "standalone": false,
    "tier": "older",
    "tierLabel": "Older-gen GPUs confirmed (V100/T4/P100/K-series/etc.)",
    "gpu": "BigDawg's documented GPU complement is 2 NVIDIA K40 GPU nodes, from the Cisco UCS build. The NVIDIA A6000 previously cited in this row is not on BigDawg — it is a single Dell server in one School of Computing research laboratory (Engineering Building A0405C, Dr. Daren Che's group, access limited to permitted students). GPU inventory after the Dell/OpenHPC transition is not published.",
    "slurm": "Yes — SLURM, adopted during the BigDog-to-BigDawg transition which replaced Rocks with OpenHPC and SGE with SLURM",
    "hpcName": "BigDawg HPCC — Office of Research Computing and Cyber Infrastructure (RCC)",
    "specs": "BigDawg is rated at 34.7 TeraFLOPS. Its predecessor BigDog went into production in December 2015 with 40 Cisco servers and 800 CPU cores; the Cisco UCS build is documented as 40 nodes of two Intel Xeon E-5 with 64GB RAM, 2 nodes at 768GB RAM, and 2 K40 GPU nodes. The transition to BigDawg replaced some Cisco hardware with Dell for sustainability and security reasons. SIU also runs a separate Dell Linux cluster of 106 nodes (Intel Xeon dual quad-core 2.3GHz, 8GB RAM, 90TB storage). Data centre network capability is 560 Gbps with headroom to double. Housed in the Wham Data Center.",
    "notes": "RECLASSIFIED Sept 2026, modern -> older. This is the same failure mode found at Missouri and Tulane: hardware from a narrower context recorded as though it belonged to the university's shared cluster. The row's modern tier rested on an NVIDIA A6000, but SIU's own laboratory listing places that card in a single Dell server inside one faculty member's research lab, available to his permitted students — not on BigDawg and not requestable by a law faculty member. BigDawg's own documented GPUs are 2 NVIDIA K40 nodes, which is squarely the older tier. THE ACCESS STORY IS GENUINELY GOOD THOUGH, and better than the tier suggests: SIU's RCC states that unlike larger institutions where HPC access is limited and often scheduled, SIU offers BigDawg FREE, and the RCC assists campus researchers with research data storage, high performance computing and scientific data visualization. SIU Carbondale is an R1 institution. So a law faculty member here has easy, free access to a modest older-generation cluster — a different and arguably more useful situation than restricted access to newer hardware.",
    "networks": [],
    "sources": [
      "https://oit.siu.edu/rcc/",
      "https://soc.siu.edu/laboratories.php",
      "https://bigdata.asa.siu.edu/about/index.php",
      "https://www.researchgate.net/publication/334782074_From_BigDog_to_BigDawg_Transitioning_an_HPC_Cluster_for_Sustainability",
      "https://soc.siu.edu/research/research-areas/index.php"
    ],
    "caveat": "GPU inventory is stale and possibly incomplete. The 2 K40 figure comes from the Cisco UCS era, and the BigDog-to-BigDawg migration replaced some Cisco hardware with Dell without publishing a refreshed GPU list. BigDawg may now hold newer cards than this row records, or fewer. Anyone relying on this should ask RCC directly. Do not treat the School of Computing lab A6000 as an available resource."
  },
  {
    "state": "Indiana",
    "school": "Indiana University Maurer School of Law (Bloomington)",
    "university": "Indiana University Bloomington",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Big Red 200: 256 NVIDIA A100 GPUs across 64 GPU nodes of four. Quartz: 22-24 GPU nodes with four NVIDIA V100 each, PLUS a dedicated HOPPER GPU PARTITION of 12 nodes with four NVIDIA H100 each (48 H100) — described by IU as being for intensive work that cannot fit on the V100s or A100s.",
    "slurm": "Yes",
    "hpcName": "Big Red 200 + Quartz (incl. Hopper H100 partition) — UITS Research Technologies",
    "specs": "Big Red 200 (HPE Cray EX/Shasta, AMD Rome CPUs, ~6 PFLOPS): 640 compute nodes with 256GB RAM and two 64-core CPUs each, 4-day job limit; 64 GPU nodes with a 2-day limit. It was the first Cray EX installed at a US university. Quartz: ~90-92 general compute nodes with two 64-core CPUs and 512GB RAM, 4-day limit; GPU nodes carry 768GB RAM with a 2-day limit. Slate provides additional attached storage, with a tape archive for long-term retention. Research Desktop (RED) gives GUI-based interactive access for users who would rather not work at the command line.",
    "notes": "RECLASSIFIED Sept 2026, modern -> frontier. This row is the first UNDERSTATEMENT found in the tier-claim group, and it came from stopping at Big Red 200. That machine's 256 A100s are real and correctly recorded, but IU's supercomputing documentation also describes a HOPPER GPU PARTITION on Quartz — 12 nodes with four NVIDIA H100 each — explicitly positioned for work that cannot fit on the V100s or A100s. H100 presence makes this frontier under this project's own tier table. Access is excellent and explicitly universal: IU describes Quartz and Research Desktop as OPEN TO ALL STUDENTS, FACULTY AND STAFF, and characterises its supercomputing generally as easy to get access to, generally free to use, with unlimited access. That is an unusually low barrier for a law faculty member, and the Research Desktop route means no command- line expertise is strictly required to start. Maurer sits on the Bloomington campus where Big Red 200 and Quartz are operated by UITS Research Technologies, so this is a same-campus resource.",
    "networks": [],
    "sources": [
      "https://scholarworks.iu.edu/iuswrrest/api/core/bitstreams/d16228a0-bf79-4772-a136-9656e6c49ee0/content",
      "https://cl.indiana.edu/Supercomputing_at_IU.pdf",
      "https://uits.iu.edu/services/technology-for-research/high-performance-computing-and-storage/index.html",
      "https://news.iu.edu/it/live/news/29064-big-red-200-is-a-leap-forward-for-indiana"
    ]
  },
  {
    "state": "Indiana",
    "school": "IU Robert H. McKinney School of Law (Indianapolis)",
    "university": "Indiana University Indianapolis",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Big Red 200: 256 NVIDIA A100 GPUs across 64 GPU nodes of four. Quartz: 22-24 GPU nodes with four NVIDIA V100 each, PLUS a dedicated HOPPER GPU PARTITION of 12 nodes with four NVIDIA H100 each (48 H100) — described by IU as being for intensive work that cannot fit on the V100s or A100s.",
    "slurm": "Yes",
    "hpcName": "Big Red 200 + Quartz (incl. Hopper H100 partition) — UITS Research Technologies",
    "specs": "Big Red 200 (HPE Cray EX/Shasta, AMD Rome CPUs, ~6 PFLOPS): 640 compute nodes with 256GB RAM and two 64-core CPUs each, 4-day job limit; 64 GPU nodes with a 2-day limit. It was the first Cray EX installed at a US university. Quartz: ~90-92 general compute nodes with two 64-core CPUs and 512GB RAM, 4-day limit; GPU nodes carry 768GB RAM with a 2-day limit. Slate provides additional attached storage, with a tape archive for long-term retention. Research Desktop (RED) gives GUI-based interactive access for users who would rather not work at the command line.",
    "notes": "RECLASSIFIED Sept 2026, modern -> frontier. This row is the first UNDERSTATEMENT found in the tier-claim group, and it came from stopping at Big Red 200. That machine's 256 A100s are real and correctly recorded, but IU's supercomputing documentation also describes a HOPPER GPU PARTITION on Quartz — 12 nodes with four NVIDIA H100 each — explicitly positioned for work that cannot fit on the V100s or A100s. H100 presence makes this frontier under this project's own tier table. Access is excellent and explicitly universal: IU describes Quartz and Research Desktop as OPEN TO ALL STUDENTS, FACULTY AND STAFF, and characterises its supercomputing generally as easy to get access to, generally free to use, with unlimited access. That is an unusually low barrier for a law faculty member, and the Research Desktop route means no command- line expertise is strictly required to start. McKinney is on the Indianapolis campus rather than Bloomington. UITS systems are described as open to all IU students, faculty and staff rather than being campus-restricted, so this should apply — but a McKinney researcher may wish to confirm with UITS, since cross-campus assumptions have proved wrong elsewhere in this dataset (see Missouri and University of Baltimore).",
    "networks": [],
    "sources": [
      "https://scholarworks.iu.edu/iuswrrest/api/core/bitstreams/d16228a0-bf79-4772-a136-9656e6c49ee0/content",
      "https://cl.indiana.edu/Supercomputing_at_IU.pdf",
      "https://uits.iu.edu/services/technology-for-research/high-performance-computing-and-storage/index.html",
      "https://news.iu.edu/it/live/news/29064-big-red-200-is-a-leap-forward-for-indiana"
    ]
  },
  {
    "state": "Indiana",
    "school": "Notre Dame Law School",
    "university": "University of Notre Dame",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "CRC/CAML (general access, NSF-funded \"Cyberinfrastructure to Accelerate Machine Learning\"): NVIDIA Quadro RTX 6000 confirmed via real job output; ACMS dept has priority access to 8 GPUs (model unspecified) on 800 dedicated cores. Separately, Mendoza College of Business has 2 dedicated machines with 8x NVIDIA H100 SXM GPUs each (640GB total GPU RAM) — confirms H100 exists on the Notre Dame network, though this specific pool is Mendoza-restricted, not general CRC access.",
    "slurm": "Not confirmed (established center,  specs not captured)",
    "hpcName": "Center for Research Computing (CRC) / CAML",
    "specs": "CRC serves the whole Notre Dame research community incl. external/industry partners; CAML allocates 20% of capacity via Open Science Grid",
    "notes": "CRC provides faculty-sponsored access to any Notre Dame community member, plus facilitates access to national XSEDE/ACCESS resources. CORRECTED Sept 2026 — confirmed RTX 6000 via live CAML job logs and H100 SXM via Mendoza's own resources page.",
    "networks": [],
    "sources": [
      "https://docs.crc.nd.edu/resources/caml.html",
      "https://mendozaresearch.nd.edu/resources/computing/",
      "https://acms.nd.edu/about/computational-facilities/",
      "https://docs.crc.nd.edu/resources/gpu.html"
    ],
    "caveat": "The confirmed H100 SXM pool belongs specifically to Mendoza College of Business, not the general-access CRC pool a Law School researcher would default to — that general pool is confirmed only at RTX 6000. Listed as Frontier because H100s demonstrably exist on campus and CRC facilitates cross-department resource requests, but a Law School researcher's default access point (CAML/general CRC) is Modern-tier hardware, not Frontier."
  },
  {
    "state": "Iowa",
    "school": "University of Iowa College of Law",
    "university": "University of Iowa",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "NVIDIA A100 (40GB and 80GB variants), A40, A30, A10 available as purchasable GPU cards; RTX8000/2080Ti/3080/3090Ti also offered; up to 8 GPUs/node",
    "slurm": "Not explicitly confirmed as SLURM (uses qlogin/queue terminology)",
    "hpcName": "Argon HPC (ITS-Research Services)",
    "specs": "~23,000 CPU processors; 350+ GPU accelerators total",
    "notes": "Free to all UI faculty, staff, students conducting research; buy/rent dedicated nodes available. CORRECTED Sept 2026 — A100 confirmed as a current purchasable card option.",
    "networks": [],
    "sources": [
      "https://research.its.uiowa.edu/our-services/computing-services/argon-high-performance-computing-hpc/buy-or-rent-compute-nodes-0"
    ]
  },
  {
    "state": "Iowa",
    "school": "Drake University Law School",
    "university": "Drake University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026 against both the central-IT and department angles — the pattern that caught Creighton's cluster. Neither turned anything up. Drake ITS is enterprise support (passwords, software downloads, classroom response, a service portal and support centre) with no research-computing service, no cluster, no scheduler and no account-request path. The Department of Mathematics and Computer Science is undergraduate-focused; its published research is student-led group work (VR apps, game development, data mining of tweets, robotics in Collier-Scripps Hall) of a kind that runs on workstations. No HPC resource found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.drake.edu/its/facstaff/",
      "https://www.drake.edu/cs/research/",
      "https://www.drake.edu/cs/about/"
    ]
  },
  {
    "state": "Kansas",
    "school": "University of Kansas School of Law",
    "university": "University of Kansas",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "NVIDIA T4, V100, A40, RTX A6000, L40S, A100, GTX 1080 Ti GPUs",
    "slurm": "Yes — Slurm, with named partitions including sixhour and bigjay",
    "hpcName": "KU Community Cluster (condo) — Center for Research Computing; also Bigjay and Hawk",
    "specs": "Housed in the NIH-funded Advanced Computing Facility at Nichols Hall, completed 2013, which added 32 high-density racks and capacity for more than 24,000 processing cores. The ACF also houses Information and Telecommunications Technology Center servers. Two sibling systems: Bigjay, an NSF-funded cluster (MRI Award #2117449) currently restricted to its PIs, Co-PIs and collaborators but contributing to the shared sixhour partition; and Hawk, a small condo cluster built for Research Health Information and research-identifiable data.",
    "notes": "Re-verified Sept 2026 — modern retained, but the significant finding on this row is ACCESS, not hardware. The KU Community Cluster runs a pure condo model, and KU's HPC documentation states the rule bluntly: YOU MUST BE AN OWNER OF HARDWARE IN THE KU COMMUNITY CLUSTER OR BE SPONSORED BY AN OWNER GROUP WITHIN THE CLUSTER TO GAIN ACCESS. There is no free general tier of the kind Nebraska, New Mexico, Indiana or Northeastern offer. A law faculty member would need either to buy nodes or to find a sponsoring owner group — a materially higher barrier than most rows in this dataset, and the thing worth knowing before approaching CRC. What owners and their sponsored users get is good: dedicated use of purchased nodes plus the ability to run larger jobs on other owners' idle hardware, with CRC staff handling administration, training sessions and one-to-one support. The ACF is described as serving an exceptionally diverse range of researchers from all KU campuses.",
    "networks": [],
    "sources": [
      "https://docs.crc.ku.edu/clusters/",
      "https://crc.ku.edu/systems-services/ku-community-cluster",
      "https://crc.ku.edu/systems-services",
      "https://technology.ku.edu/research/advanced-computing-facility",
      "https://docs.crc.ku.edu/"
    ],
    "caveat": "Condo-only access, and the GPU inventory is unverified. Two separate cautions. First, ownership or owner-group sponsorship is a stated precondition of access, so the tier describes hardware a law researcher may not be able to reach without buying in. Second, CRC's public pages describe the cluster's structure and policies but do not publish a GPU model inventory, so the specific cards recorded in this row could not be independently confirmed in this pass and should be treated as provisional."
  },
  {
    "state": "Kansas",
    "school": "Washburn University School of Law",
    "university": "Washburn University",
    "standalone": false,
    "tier": "cpuonly",
    "tierLabel": "No GPU confirmed (CPU/RAM-only)",
    "gpu": "None — the published hardware inventory is entirely CPU; no GPU of any generation is listed",
    "slurm": "Not confirmed — no scheduler named in the public documentation",
    "hpcName": "HiPACE — High-Performance Academic Computing Environment (Natural Sciences Division)",
    "specs": "Three parts: (1) a cluster with a master node and 34 compute nodes, each with two dual-core Opteron processors and 4GB RAM; (2) an SMP system of 8 dual-core Opterons (16 cores) with 64GB RAM for large-memory jobs; (3) a 12TB file system. Paired gigabit Ethernet per machine, one for data and one for control. All nodes run OpenSuSE Linux 10.2.",
    "notes": "RECLASSIFIED Sept 2026, none -> cpuonly. Washburn DOES have a named research-computing resource, which the previous 'none' missed: HiPACE, the High-Performance Academic Computing Environment, operated and administered by faculty and students primarily in Washburn's Natural Sciences Division, with its own site at hipace.washburn.edu. Its stated mission is to enrich science and technology education for Washburn students and enhance the capabilities of faculty, staff and community members. BUT READ THE SPECS BEFORE GETTING EXCITED — and this is exactly why the row is 'cpuonly' rather than 'undetailed'. The hardware is fully enumerated on the public page and contains no GPU at all: dual-core Opterons with 4GB of RAM per node. HiPACE was funded by a Washburn Innovation Grant covering 2006 to 2009, and the nodes are documented as running OpenSuSE Linux 10.2, a release that reached end of life in 2008. So this is a genuine resource on paper but an eighteen-year-old CPU cluster on an unsupported OS. A fast pass that saw the words 'high-performance computing system' and stopped there would have upgraded this row wrongly; reading the actual specification page is what produced the correct tier. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://hipace.washburn.edu/hipace/home.php",
      "https://www.washburn.edu/academics/college-schools/arts-sciences/computer-information-sciences/index.html"
    ],
    "caveat": "Possibly dormant, and CPU-only regardless. The public documentation describes 2006-2009 funding and an operating system that went end-of-life in 2008, with no evidence of subsequent refresh. Before relying on this row, confirm with the HiPACE Steering Committee that the system is still running. Even if it is, there is no GPU here — a law faculty member with an AI or machine-learning workload would need a different resource."
  },
  {
    "state": "Kentucky",
    "school": "University of Kentucky College of Law",
    "university": "University of Kentucky",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "NVIDIA P100/V100/A100 confirmed (34 GPU nodes, 128 GPUs total, mixed generations) per CCS's own internal state-of-the-center presentation; 120 GPU accelerators cluster-wide per CCS overview page (1.9 PFLOPS combined)",
    "slurm": "Not confirmed (traditional batch system; scheduler unspecified in snippets)",
    "hpcName": "Center for Computational Sciences (CCS) — Lipscomb Compute Cluster (LCC)",
    "specs": "LCC (Lipscomb Compute Cluster): ~25,000 processor cores; MCC (Morgan, AMD EPYC, CPU-focused); ECC (EduceLab); 5PB+ GPFS storage; condo model",
    "notes": "Open to UK researchers via ITS-RCI/CCS partnership; condo buy-in for priority access; SLURM confirmed via CCS docs. CORRECTED Sept 2026 — A100 confirmed via a CCS staff conference presentation (\"State of Research Computing and Data at UK\") that names the exact P100/V100/A100 mix, where the public-facing overview pages only said \"GPU accelerators.\"",
    "networks": [],
    "sources": [
      "https://www.ccs.uky.edu/wp-content/uploads/2024/02/20240130-Griffioen-1.pdf",
      "https://docs.ccs.uky.edu/display/HPC/Research+computing+infrastructure",
      "https://www.ccs.uky.edu/?p=3153"
    ]
  },
  {
    "state": "Kentucky",
    "school": "University of Louisville Brandeis School of Law",
    "university": "University of Louisville",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "UofL ITS describes its flagship HPC system as featuring NVIDIA H200 and H100 GPUs alongside AMD CPUs. The most concretely confirmed Hopper hardware is LARCC (Louisville Academic Research Compute Cluster), NSF CC*-funded, delivering twenty NVIDIA H100 GPUs, which ITS states is ready for use by researchers. A further DGX H200 platform is a separate $3.8M investment described as under development.",
    "slurm": "Not confirmed",
    "hpcName": "Cardinal Research Cluster (CRC) + LARCC + CARDS — ITS Research Computing",
    "specs": "High-memory compute nodes at 6TB, 3TB and 1.5TB. CARDS (Cardinal Academic Research Data Storage), also NSF CC*-funded, provides roughly 6PB of shared storage and is ready for use; the in-development DGX H200 platform adds 5PB of parallel distributed storage and up to 6TB RAM per node. ITS has also worked with Google and CDW on assured workloads for secure, compliant research environments.",
    "notes": "Re-verified Sept 2026 — frontier CONFIRMED, with a distinction the old row blurred. Two different Hopper deployments are in play and they are at different stages. LARCC's twenty H100 GPUs are explicitly stated by ITS to be ready for use by researchers, so Hopper-class hardware is genuinely available today and the tier is correct on that basis alone. The DGX H200 platform is a separate $3.8M investment that ITS describes as under development, even though the main Research Computing page already advertises the flagship as featuring H200 and H100. Anyone updating this row should keep those apart rather than reading the marketing line as a single delivered system. Access is unusually clean for a law faculty member: ITS Research Computing states its HPC, storage and consulting are provided at no cost to UofL-affiliated faculty, students and staff, with no disciplinary restriction, plus support for grant applications and regulated-data requirements.",
    "networks": [],
    "sources": [
      "https://louisville.edu/its/services/research-computing/computing-and-research",
      "https://louisville.edu/its/news/major-research-computing-advancements"
    ],
    "caveat": "Two Hopper deployments at different stages — do not conflate them. LARCC's 20 H100 GPUs are stated by UofL ITS to be ready for use by researchers, and that is what carries the frontier tier. The separate $3.8M DGX H200 platform with 5PB of storage is described as UNDER DEVELOPMENT, even though the main Research Computing page already advertises the flagship as featuring H200 and H100. Treat the H200 capacity as pending until confirmed with ITS."
  },
  {
    "state": "Kentucky",
    "school": "Salmon P. Chase College of Law",
    "university": "Northern Kentucky University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026. No campus HPC or research-computing center was found at Northern Kentucky University. NKU's School of Computing and Analytics is an active unit with an NSA Center of Excellence in Cybersecurity designation, but its published student and faculty research uses commercial cloud (Amazon ECS/Fargate container clusters) rather than an on-campus scheduler-managed cluster, and no NKU page advertises HPC accounts. Worth flagging one unresolved statewide thread: the University of Kentucky's Center for Computational Sciences states that its supercomputing facilities are available to UK faculty and to research institutions throughout the Commonwealth of Kentucky. That language is broad enough that an NKU-affiliated researcher might have a route in, but CCS does not publish eligibility terms for non-UK institutions and nothing confirms NKU or Chase specifically, so this is logged as a lead rather than scored as access. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.nku.edu/academics/informatics/schools/sca/beyond-the-classroom/research.html",
      "https://www.ccs.uky.edu/?p=39",
      "https://www.ccs.uky.edu/services-and-support/research-computing"
    ]
  },
  {
    "state": "Louisiana",
    "school": "LSU Paul M. Hebert Law Center",
    "university": "Louisiana State University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "DeepBayou (LSU's own GPU cluster): NVIDIA V100S (2-4/node, NVLink on 2 nodes); plus 144x NVIDIA A100 via statewide LONI QB4 supercomputer, which all LSU HPC users can access",
    "slurm": "Yes (partition-based scheduling,  standard SLURM terminology)",
    "hpcName": "HPC@LSU - SuperMike-III / SuperMIC / DeepBayou (+ LONI network)",
    "specs": "SuperMike-III: 11,712 cores, 1.3 PFLOP, 183 nodes; DeepBayou: 13 GPU nodes; QB4 (LONI): 35,008 cores, 4.3 PFLOPS",
    "notes": "Available to all LSU faculty and their affiliates; also connects to statewide LONI HPC network (QB4). CORRECTED Sept 2026 — V100S confirmed on LSU's own DeepBayou; A100 confirmed via LONI QB4 access.",
    "networks": [
      "LONI (Louisiana)"
    ],
    "sources": [
      "http://www.hpc.lsu.edu/resources/hpc/system.php?system=DeepBayou",
      "https://www.hpc.lsu.edu/resources/hpc/system.php?system=SuperMike-III"
    ]
  },
  {
    "state": "Louisiana",
    "school": "Tulane University School of Law",
    "university": "Tulane University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "CORRECTED: Tulane's own Cypress cluster has NO NVIDIA GPUs — each node carries dual Intel Xeon Phi 7120P COPROCESSORS, not GPUs. GPU access comes through LONI, where Queen Bee 4 provides 10 four-GPU NVIDIA A100 nodes (40 A100) alongside 480 CPU compute nodes and 5 big-memory 2TB nodes, at no cost to Tulane researchers. The '960 NVIDIA Tesla K20x' figure previously attributed to Cypress belongs to LONI's earlier Queen Bee 2 generation, not to any Tulane machine.",
    "slurm": "Yes",
    "hpcName": "LONI (Queen Bee 4) for GPUs; Cypress (Tulane's own cluster, CPU/Xeon Phi only)",
    "specs": "Cypress: 124 nodes, each with dual 10-core 2.8GHz Intel Xeon E5-2680 v2 (Ivy Bridge) CPUs, 64GB RAM and dual Xeon Phi 7120P coprocessors, on 40 Gigabit Ethernet through a single Dell Z9500 fabric switch. 699TB-1.476PB Lustre with Intel Enterprise Edition, 1TB and 1 million inodes allocated per research group, 10GB backed-up home directories. Cypress ranked 271st on the TOP500 in November 2014. Located in Tulane's downtown New Orleans datacenter. Tulane notes Cypress was designed to resemble larger LONI and TACC machines so code ports upward easily.",
    "notes": "CORRECTED Sept 2026. The tier stands at modern, but the hardware attribution in this row was wrong in the same way Missouri's was — GPUs from a shared regional machine were recorded as if they belonged to the university's own cluster. Tulane's Cypress is a 2014-era Ivy Bridge system whose accelerators are Intel Xeon Phi coprocessors; it has no NVIDIA GPUs at all. What justifies 'modern' is LONI: Tulane is a confirmed member institution of the Louisiana Optical Network Infrastructure (the same membership check that ruled OUT Loyola New Orleans elsewhere in this dataset), and Tulane IT lists LONI's Queen Bee 4 — 480 CPU nodes, 10 four-GPU A100 nodes, 5 big-memory nodes — as an advanced computing system available to its researchers at no cost. Access to Cypress requires association with a Tulane faculty PI, and accounts are available to any Tulanian on that basis; Tulane runs periodic HPC workshops and offers consultation through its Center for Research and Scientific Computing. Encouragingly for a law audience, Tulane's own list of projects enabled by its HPC services includes financial market trade data analysis, so non-laboratory quantitative work is already represented.",
    "networks": [
      "LONI (Louisiana)"
    ],
    "sources": [
      "https://wiki.hpc.tulane.edu/trac/wiki/cypress/about",
      "https://it.tulane.edu/resources/research-resources",
      "https://wiki.hpc.tulane.edu/trac/wiki/cypress",
      "https://it.tulane.edu/technology-delivery/research-computing",
      "https://crsc.tulane.edu/"
    ],
    "caveat": "The GPUs are LONI's, not Tulane's, and Cypress is aging. A Tulane law researcher wanting NVIDIA GPU time should be pointed at LONI's Queen Bee 4 rather than at Cypress, which is a decade-old CPU/Xeon Phi machine. Xeon Phi is a discontinued Intel line, so code written for it has no forward path."
  },
  {
    "state": "Louisiana",
    "school": "Southern University Law Center",
    "university": "Southern University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Access to LONI's current flagship QB4 supercomputer: 144x NVIDIA A100 GPUs (QB2's 960x K20x has since been superseded/retired)",
    "slurm": "Likely (LONI-standard)",
    "hpcName": "LONI network access (Queen Bee 2 supercomputer) + limited access to LSU SuperMIC",
    "specs": "QB4 (LONI): 35,008 cores, 4.3 PFLOPS, 6.5PB Lustre storage",
    "notes": "Southern is one of 6 LONI member research universities in Louisiana; special permission needed for LSU SuperMIC access. CORRECTED Sept 2026 — updated to reflect current LONI flagship (QB4/A100) rather than retired QB2 hardware.",
    "networks": [
      "LONI (Louisiana)"
    ],
    "sources": [
      "https://hpc.loni.org/"
    ]
  },
  {
    "state": "Louisiana",
    "school": "Loyola University New Orleans College of Law",
    "university": "Loyola University New Orleans",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found (LONI membership not established)",
    "specs": "",
    "notes": "Re-verified Sept 2026, specifically to test the obvious hypothesis that a New Orleans university would ride Louisiana's statewide LONI network. It does not appear to. LONI's own cyberinfrastructure documentation names its research-university members as LSU, Louisiana Tech, LSU Health Sciences Center New Orleans, LSU HSC Shreveport, Southern University, Tulane, UL Lafayette and the University of New Orleans; Loyola University New Orleans is not among them, and no Loyola page claims LONI access. Loyola's own Department of Mathematics and Computer Science is an undergraduate teaching department (BS degrees in Mathematics, Computer Information Systems and Computer Science) with no research cluster, scheduler or GPU inventory published. Tulane, by contrast, documents both its own Cypress cluster and LONI access — so the New Orleans institution with compute is Tulane, not Loyola. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://loni.laregents.edu/wp-content/uploads/LONI-Cyberinfrastructure-Plan-2023-02-27.pdf",
      "https://cas.loyno.edu/departments/natural-sciences/mathematics-computer-science",
      "https://it.tulane.edu/technology-delivery/research-computing"
    ]
  },
  {
    "state": "Maine",
    "school": "University of Maine School of Law",
    "university": "University of Maine",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Penobscot cluster, node-by-node: node-g105 = NVIDIA DGX A100 (8x A100 40GB); node-g101 = 2x NVIDIA A100 80GB; node-g102 = 3x NVIDIA L40 (48GB); node-g103 = 4x NVIDIA A30 (24GB); node-g104 = 8x NVIDIA RTX 2080 Ti (legacy)",
    "slurm": "Yes",
    "hpcName": "ARCSIM (Advanced Research Computing, Security & Information Management) — Penobscot cluster",
    "specs": "epyc partition: 1,344 cores (14 nodes); Infiniband HDR100 (100Gbps); also brokers access to Ohio Supercomputer Center + TACC Stampede3/Frontera + NSF ACCESS",
    "notes": "Serves the full University of Maine System research community; billed by GPU-hour tier (A100 = Tier 1, L40/A30/T4 = Tier 2). CORRECTED Sept 2026 — fetching the full page (not just the search snippet) revealed a complete, node-by-node GPU hardware table: A100 and L40 confirmed present.",
    "networks": [],
    "sources": [
      "https://umaine.edu/arcsim/computing/"
    ]
  },
  {
    "state": "Maryland",
    "school": "University of Maryland Francis King Carey School of Law",
    "university": "University of Maryland",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "80 NVIDIA A100 GPUs (20 GPU nodes x 4, 40 GiB each, Ampere, CUDA capability 8.0) and 32 NVIDIA H100 GPUs (8 GPU nodes x 4, 80 GiB each, CUDA capability 9.0). All SXM models supporting NVLink. A100s can be subdivided into a100_1g.5gb multi-instance slices of 5 GiB.",
    "slurm": "Yes — Slurm with named partitions (standard, gpu, bigmem, debug, scavenger) and --constraint feature flags",
    "hpcName": "Zaratan HPC Cluster",
    "specs": "Online since August 2022, replacing the 160-teraflop Deepthought2. 360 standard compute nodes at 128 cores and 512 GiB each (46,080 cores), 6 large-memory nodes at 2 TiB, 2PB BeeGFS scratch, 3.5 PFLOPS theoretical peak. Standard, large-memory and A100 nodes use dual AMD EPYC 7763 Zen3; H100 nodes use dual Intel Xeon Platinum 8468 Sapphire Rapids. HDR100 interconnect on compute nodes, HDR on GPU nodes. Housed off campus in the DataBank IAD2 facility in McLean, Virginia, maintained by UMD's Division of Information Technology.",
    "notes": "Re-verified Sept 2026 — the GPU figures in this row are EXACT and check out against UMD's own cluster page: 20 nodes x 4 A100 = 80, and 8 nodes x 4 H100 = 32. Access is good and well-documented: all faculty are eligible for a free 50 KSU allocation from the campus Allocations and Advisory Committee, with up to 500 KSU more available through a proposal process (550 KSU total), and HPC staff explicitly encourage new users to take a free allocation to test-drive the cluster before committing funds. Note the GPU accounting: a KSU is 1,000 CPU core-hours, and GPU time carries a premium of 48x for an A100 and 144x for an H100, so a free 50 KSU allocation buys roughly 347 H100-hours. Encouraging for a law faculty member, UMD names Public Policy among the departments Zaratan already serves, alongside the expected Mathematics, Physics and Astronomy. Unused service units do not roll over between quarters. Cross-reference: this is the same cluster whose access rules were used to rule OUT a pathway for the University of Baltimore, since Zaratan requires a UMD directory ID — an entitlement the Carey School of Law has and UBalt does not, despite both being in the USM.",
    "networks": [],
    "sources": [
      "https://hpcc.umd.edu/home/clusters/zaratan/",
      "https://hpcc.umd.edu/",
      "https://ask.eng.umd.edu/124188",
      "https://hpcc.umd.edu/kb/queues/",
      "https://www.datacenterdynamics.com/en/news/university-of-maryland-deploys-supercomputer-in-databank-data-center-in-mclean-virginia/"
    ]
  },
  {
    "state": "Maryland",
    "school": "University of Baltimore School of Law",
    "university": "University of Baltimore",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found (USM membership does not confer UMD Zaratan access)",
    "specs": "",
    "notes": "Re-verified Sept 2026 against the system-membership hypothesis, which fails. The University of Baltimore is a public institution inside the University System of Maryland, which invites the assumption that its faculty can use UMD's flagship Zaratan HPC cluster. UMD's own access documentation rules that out: Zaratan requires a UMD TerpConnect/Glue account tied to a University of Maryland Directory ID, allocations are granted to UMD faculty, and confirmations are sent to a @umd.edu address. HPC@UMD describes its customers as the University of Maryland community. Non-UMD people need a UMD faculty sponsor, which is a collaboration route rather than an institutional entitlement. No UBalt research-computing center of its own was found; UBalt is a roughly 3,200-student institution with colleges in applied arts and sciences, business, law and public affairs. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://hpcc.umd.edu/faq/access/",
      "https://hpcc.umd.edu/",
      "https://hpcc.umd.edu/home/get-access/",
      "https://en.wikipedia.org/wiki/University_of_Baltimore"
    ]
  },
  {
    "state": "Massachusetts",
    "school": "Harvard Law School",
    "university": "Harvard University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "1,000+ GPUs on Cannon spanning NVIDIA V100, A40, A100, H100 and H200 (4 GPUs per node). FASRC's own GPU training deck lists GPUs/node as A100, H100, H200. Named partitions include gpu_h200 (24 Lenovo nodes), kempner (132 A100 80GB across water-cooled 4-GPU nodes), kempner_h100, kempner_h200, seas_gpu (244 GPUs ranging A100 to H200), itc_gpu (4 nodes x 4 A100), and hsph_gpu. 144 additional A100 80GB GPUs were added in November 2025.",
    "slurm": "Yes — Slurm job manager, Rocky 8, Spack package manager",
    "hpcName": "FAS Research Computing (FASRC) - Cannon cluster",
    "specs": "100,000+ CPU cores (Intel Sapphire Rapids/Cascade Lake/Skylake/Icelake, AMD Genoa/Zen); 48–112 cores per node; 184GB–2TB RAM per node; 200 GB/s HDR and NDR InfiniBand across three data centers; multi-PB Lustre. Separate FASSE secure enclave (Harvard data security level 3) adds 48 nodes / 2,304 cores plus 4 nodes of 4x A100. 25+ RC staff supporting 600+ research groups and 5,500+ users.",
    "notes": "RECLASSIFIED Sept 2026 — this row was previously scored 'older' on the strength of a '1000+ GPUs (V100 and newer generations)' line, which badly understated Cannon. Harvard FAS Research Computing now runs A100, H100 and H200 hardware with dedicated H200 partitions, and the Kempner Institute for the Study of Natural and Artificial Intelligence holds its own purchased A100/H100/H200 partitions on the same cluster. On any reading of the tier definitions this is frontier, not older-gen. Access is by FASRC account, billed per project at the school level, with fairshare scheduling — but see the caveat about which schools FASRC actually lists as supported.",
    "networks": [
      "MGHPCC (Massachusetts consortium)"
    ],
    "sources": [
      "https://docs.rc.fas.harvard.edu/wp-content/uploads/2013/10/GPU_Computing_9_2025.pdf",
      "https://www.rc.fas.harvard.edu/wp-content/uploads/2024/07/FASRC-Services-One-Pager.pdf",
      "https://docs.rc.fas.harvard.edu/kb/kempner-partitions/",
      "https://docs.rc.fas.harvard.edu/kb/seas-compute-resources/",
      "https://www.rc.fas.harvard.edu/services/cluster-computing/"
    ],
    "caveat": "Law is not among the schools FASRC names. FASRC's own materials describe it as a Faculty of Arts and Sciences department supporting '600+ research groups and 5500+ users across FAS, SEAS, HSPH, HBS, GSE' — Harvard Law School is not in that list, and cluster access is stated as open to PIs 'from a supported Harvard School'. Billing is per project at the school level, so an HLS arrangement would likely need to be negotiated rather than simply requested. The hardware tier here is real; the law-specific access route is the open question."
  },
  {
    "state": "Massachusetts",
    "school": "Boston University School of Law",
    "university": "Boston University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "470+ GPUs (RCS June 2025 figure; MGHPCC's page says 500). The SCC's live `qgpus` inventory shows A100, A100-80G, A40, A6000 and H200 among others, with H200 nodes carrying 4 GPUs each at 144GB GPU memory. Older P100/K40m cards remain in the heterogeneous mix.",
    "slurm": "Yes",
    "hpcName": "Shared Computing Cluster (SCC)",
    "specs": "1,056 compute nodes as of 2025 (up from 189 in 2013); 29,000+ CPU cores; 15 PB storage; AlmaLinux 8; InfiniBand fabrics for multi-node parallel work, 25 GbE for data-intensive jobs. Serves 3,800 researchers in 1,200 projects across 90+ departments plus 70 courses; 110M+ core hours computed in the past year. Housed at the LEED Platinum MGHPCC in Holyoke.",
    "notes": "Re-verified Sept 2026 and revised upward. The old row's '400+ GPUs / 28,000 cores / 14PB' came from BU's September 2024 facilities document; the June 2025 RCS brochure gives 470+ GPUs, 29,000+ cores and 15PB, and MGHPCC's own project page says 500 GPUs. Frontier is confirmed either way — H200 nodes are live in the scheduler. Access is good for a law faculty member: the SCC is stated as available to all University faculty, their students and their collaborators, for research and for educational use in courses, with shared (centrally funded, no cost) nodes alongside researcher-owned buy-in nodes. Users must be a member of an RCS research project, and BU faculty can create projects and serve as Lead Project Investigator.",
    "networks": [
      "MGHPCC (Massachusetts consortium)"
    ],
    "sources": [
      "https://www.bu.edu/tech/files/2025/06/RCS-Brochure-1122-rev-6.25.25.pdf",
      "https://www.bu.edu/tech/support/research/software-and-programming/gpu-computing/",
      "https://mghpcc.org/project/bu-research-computing-services/",
      "https://www.bu.edu/tech/support/research/computing-resources/scc/",
      "https://www.bu.edu/tech/services/research/"
    ]
  },
  {
    "state": "Massachusetts",
    "school": "Northeastern University School of Law",
    "university": "Northeastern University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Explorer: over 525 GPUs spanning NVIDIA Pascal (P100), Volta (V100), Turing (T4), Ampere (A100), RTX (A5000, A6000), Ada Lovelace (L40) and Hopper (H100 and H200) — including 32 new H200 chips at 140GB each on the gpu and multigpu partitions. AICR: 400 GPUs comprising 248 NVIDIA B200 and 152 NVIDIA RTX6000 Pro.",
    "slurm": "Yes — Slurm on both Explorer and AICR, each with SSH and Open OnDemand access",
    "hpcName": "Explorer (Northeastern HPC) + AICR (shared MGHPCC AI cluster)",
    "specs": "Explorer: 45,000+ CPU cores across Intel Xeon (Sapphire Rapids, Ice Lake, Cascade Lake, Skylake, Broadwell, Haswell, Sandy Bridge, Ivy Bridge) and AMD (Zen through Zen4); Rocky Linux 9.3; 6PB high-performance storage; 10GbE to the university network with compute nodes on 10GbE or 100/200 Gbps InfiniBand. Explorer superseded the older Discovery cluster in 2025. AICR adds dedicated project storage and a CPU-only partition for data cleaning and analysis. Both sit in the MGHPCC, a 90,000 sq ft, 15MW facility in Holyoke.",
    "notes": "Re-verified Sept 2026, and RE-SOURCED after a random consistency audit flagged this row as carrying a 248-B200 claim on a single source — a fair criticism for the largest GPU number in the Massachusetts entries. It now checks out against four independent Northeastern pages plus a training repository. Frontier is confirmed twice over: Explorer itself carries 32 H200 chips at 140GB, and AICR adds 248 B200. Access splits cleanly into two tiers. Explorer is FREE TO ALL NORTHEASTERN FACULTY AND STUDENTS with no disciplinary restriction, so a law faculty member can simply request an account. AICR requires a PI to submit a Project Proposal, and Northeastern advises developing and testing work on Explorer first before using AICR for production runs, since it is shared. CROSS-REFERENCE FOR OTHER MASSACHUSETTS ROWS: AICR is not a Northeastern machine. It was commissioned by the Healey-Driscoll Administration, built by Cambridge Computer, and is shared among ALL SIX MGHPCC MEMBER UNIVERSITIES (Northeastern, BU, Harvard, MIT, UMass, Yale) plus the Massachusetts AI Hub and partners. That means the BU, Harvard, UMass and Yale rows in this dataset may each have an AICR pathway that none of them currently records. Worth checking — it would be the single largest unrecorded resource in the dataset.",
    "networks": [
      "MGHPCC (Massachusetts consortium)"
    ],
    "sources": [
      "https://rc.northeastern.edu/",
      "https://rc.northeastern.edu/aicr/",
      "https://rc.northeastern.edu/2026/05/21/aicr-hpc-cluster-updates-from-the-avp/",
      "https://rc.northeastern.edu/2025/04/30/all-about-explorer-and-the-new-h200-gpus/",
      "https://rc.northeastern.edu/2026/02/04/research-computing-developments-aicr-hpc-cluster-and-power-cooling-upgrades/"
    ],
    "caveat": "Two tiers with different doors. Explorer is free and open to all Northeastern faculty, but its 525+ GPUs are mixed-generation and only 32 are H200. The B200 hardware sits in AICR, which needs an approved PI Project Proposal and is shared across six universities and the Massachusetts AI Hub, so contention will be real. Do not read the 248 B200 count as generally available on request."
  },
  {
    "state": "Massachusetts",
    "school": "Boston College Law School",
    "university": "Boston College",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "NVIDIA H200 GPUs (doubled in 2025 upgrade)",
    "slurm": "Yes — Slurm job scheduler with fair-share priority, plus Open OnDemand",
    "hpcName": "Andromeda (Andromeda 2) Linux HPC Cluster — ITS Research Services",
    "specs": "Weka (WekaFS) file system with an NVMe frontend, 1.5 PB initial capacity expandable to 3 PB, built on 12 frontend servers of which 10 are needed for peak performance (resilient to a double node failure); replaced the Pixstor GPFS storage in service since 2020. Geographically redundant backup added. High-RAM CPU nodes added in the 2025 upgrade. ~400 active researchers.",
    "notes": "Re-verified Sept 2026 and upgraded from 'SLURM not confirmed' — Boston College's own Andromeda documentation now shows a Slurm job scheduler with time-based partitions, sbatch CPU and GPU job examples, sinfo/squeue/sacct/scontrol/sprio tooling and fair-share job priority, alongside Open OnDemand for GUI-based interactive jobs. ITS completed a two-year upgrade in June 2025 that doubled GPU resources and added H200 cards, explicitly framed as preparing for AI research including training large language models. Note the generational change: the old cluster ('Andromeda 1') has been permanently offline since 15 September 2025, so any older documentation refers to a machine that no longer exists. Cluster use is guided by a faculty Cluster Policy Committee and each research group submits a short abstract describing its work.",
    "networks": [],
    "sources": [
      "https://www.bc.edu/bc-web/offices/its/about/its-news/2025/hpc-cluster-upgrade.html",
      "https://rs.bc.edu/andromeda/",
      "https://rs.bc.edu/",
      "https://www.bc.edu/bc-web/offices/its/services/research-services/linux-cluster.html"
    ]
  },
  {
    "state": "Massachusetts",
    "school": "UMass School of Law - Dartmouth",
    "university": "University of Massachusetts Dartmouth",
    "standalone": false,
    "tier": "older",
    "tierLabel": "Older-gen GPUs confirmed (V100/T4/P100/K-series/etc.)",
    "gpu": "Local 'Carnie' cluster (CSCDR, campus-specific): ~20-50 GPUs (model not specified); UMass-system shared MGHPCC cluster: 4x NVIDIA V100 w/ NVLink (2 nodes), 3x V100 (3 nodes), 4x Tesla K80 (6 nodes) — legacy but confirmed; newer 'Unity' platform also reachable, specific Unity GPU models not independently confirmed for this campus",
    "slurm": "No (HPCC uses IBM LSF,  not SLURM); Unity likely SLURM-based",
    "hpcName": "Carnie (CSCDR, local) + UMass system shared cluster at MGHPCC + Unity HPC & AI Platform",
    "specs": "Carnie: 1300 CPU cores, 132TB storage, 3TB memory; UMass shared MGHPCC cluster: 374 nodes, 13,968 cores, ~2.9PB Isilon storage",
    "notes": "UMass Dartmouth is an eligible partner institution on multiple UMass-system shared resources at MGHPCC. CORRECTED Sept 2026 — V100/K80 confirmed on the UMass-system shared cluster; local Carnie cluster GPU models not itemized.",
    "networks": [
      "MGHPCC (Massachusetts consortium)",
      "Unity HPC & AI Platform"
    ],
    "sources": [
      "https://www.cscdr.umassd.edu/computing",
      "https://www.umassrc.org/wiki/index.php/Main_Page",
      "https://www.umass.edu/research-computing/unity-hpc-ai-platform-0"
    ]
  },
  {
    "state": "Massachusetts",
    "school": "Suffolk University Law School",
    "university": "Suffolk University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found (not an MGHPCC member)",
    "specs": "",
    "notes": "Re-verified Sept 2026. No Suffolk research-computing service, cluster or scheduler was found; the College of Arts & Sciences computer science program is undergraduate-focused. Worth recording the negative finding on the obvious Boston hypothesis: the Massachusetts Green High Performance Computing Center is a membership consortium, and its members are the founding universities (BU, Harvard, MIT, Northeastern, UMass, with Yale now installing there) — Suffolk is not among them, and MGHPCC access runs through a member institution's own research computing group rather than being available to any Massachusetts school. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.suffolk.edu/cas/degrees-programs/computer-science",
      "https://www.bu.edu/tech/services/research/"
    ]
  },
  {
    "state": "Massachusetts",
    "school": "Western New England University School of Law",
    "university": "Western New England University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "Re-verified Sept 2026, first sourcing for this row. 'None' is CONFIRMED. Western New England University is a roughly 3,700-student private institution in Springfield with colleges of Arts and Sciences, Business, Engineering, Law and Pharmacy, classified by US News in the North Region master's category rather than as a doctoral research university. No research-computing service, cluster, scheduler or GPU inventory was found on either the central IT side or in the academic units. The Department of Computer Science & Information Technology and the Computer Engineering program are teaching-focused; faculty research areas listed (RF/wireless, signal and image processing, embedded systems, robotics, integrated circuit design, photonics) are lab-based rather than cluster-based, and no shared computational facility is documented. Structural note: the law school has a real parent university, so this row is correctly 'none' rather than 'standalone'. Historically the institution began in 1919 as a branch of Northeastern University and received an independent charter in 1951 — that Northeastern tie is long severed and confers nothing today, which is worth recording since Northeastern's Explorer cluster would otherwise look like a tempting connection. MGHPCC is likewise a closed consortium and does not reach WNE. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://wne.edu/",
      "https://en.wikipedia.org/wiki/Western_New_England_University_School_of_Law",
      "https://wne.edu/arts-and-sciences/departments/computer-science-and-information-technology/clubs-and-organizations.cfm"
    ]
  },
  {
    "state": "Michigan",
    "school": "Michigan State University College of Law",
    "university": "Michigan State University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "NVIDIA H200 GPU nodes, NVIDIA Grace Hopper superchip nodes (new), NVIDIA H100, A100 (some SXM/NVLink), plus legacy K80/P100/V100; also A6000 Ada and RTX Pro 6000 Blackwell",
    "slurm": "Yes",
    "hpcName": "ICER - High-Performance Computing Center (HPCC)",
    "specs": "16,000+ CPU cores; 6PB GPFS storage (scales to 50PB); IB network up to 400Gbps (NDR)",
    "notes": "Free account for anyone researching at MSU with faculty sponsorship; buy-in for priority access; SLURM confirmed. CORRECTED Sept 2026 — H200 and Grace Hopper superchip nodes confirmed, well beyond the previously-documented \"multiple GPU generations, unspecified.\"",
    "networks": [],
    "sources": [
      "https://icer.msu.edu/news/nvidias-grace-hopper-nodes-arrive-at-msus-data-center-to-push-boundaries-of-computational-research",
      "https://docs.icer.msu.edu/development_nodes/",
      "https://docs.icer.msu.edu/GPU_resources/"
    ]
  },
  {
    "state": "Michigan",
    "school": "University of Michigan Law School",
    "university": "University of Michigan",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Great Lakes: 52x NVIDIA V100 (24 nodes), 4x NVIDIA A100 80GB, 160x NVIDIA A40 (28 nodes); Armis2 (secure/PHI cluster): adds new NVIDIA RTX6000 Pro Blackwell nodes plus TitanV/V100",
    "slurm": "Yes",
    "hpcName": "Great Lakes HPC Cluster (ITS Advanced Research Computing)",
    "specs": "~13,000-16,000+ cores; 2PB scratch @ 80GB/s; HDR100 InfiniBand",
    "notes": "Campus-wide resource available to all researchers on all campuses incl. Law School; SLURM confirmed; 80,000 free CPU hours via UMRCP. CORRECTED Sept 2026 — A40/A100 confirmed on Great Lakes, cutting-edge RTX6000 Pro Blackwell now on Armis2. || CLASSIFICATION NOTE (Sept 2026, from a consistency audit): Armis2 carries NVIDIA RTX6000 Pro BLACKWELL nodes, and an automated check flagged this row for naming a Blackwell part while scored modern. That is correct as written — CONTRIBUTING.md's tier table lists RTX PRO 6000 under 'modern', not 'frontier', because it is a workstation/visualization part rather than a datacenter training GPU like the B200. Recorded here so the same flag is not 'fixed' into an incorrect upgrade later. Worth noting separately for a law audience that Armis2 is Michigan's SECURE cluster for PHI and other sensitive data — unusual among the rows in this dataset, most of which (USC's CARC most explicitly) forbid regulated data entirely.",
    "networks": [],
    "sources": [
      "https://documentation.its.umich.edu/node/4976",
      "https://documentation.its.umich.edu/armis2"
    ]
  },
  {
    "state": "Michigan",
    "school": "University of Detroit Mercy School of Law",
    "university": "University of Detroit Mercy",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found (only standard academic computing labs)",
    "specs": "",
    "notes": "Re-verified Sept 2026. Detroit Mercy ITS operates academic computing laboratories only — roughly 117 desktop machines across teaching labs in Commerce & Finance, Health Professions and Quad Commons, all running an identical computing image, reservable by faculty for class use. That is instructional computing, not research computing: there is no cluster, no scheduler, no GPU inventory and no account-request path. The College of Engineering & Science houses Electrical & Computer Engineering and Computer Science with a Center for Cybersecurity & Intelligence Studies, but no shared computational facility is documented. The existing 'only standard academic computing labs' description was accurate and is now sourced. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.udmercy.edu/about/its/computing/academic-labs/index.php",
      "https://eng-sci.udmercy.edu/academics/engineering/computer-science/index"
    ]
  },
  {
    "state": "Michigan",
    "school": "Western Michigan University Cooley Law School",
    "university": "Western Michigan University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Same shared MSU ICER resource as above: NVIDIA H200, Grace Hopper superchip, H100, A100, plus legacy generations",
    "slurm": "Yes (via MSU ICER SLURM system)",
    "hpcName": "HPCHub (purchased access to MSU ICER/HPCC cluster)",
    "specs": "Same as MSU ICER HPCC",
    "notes": "WMU purchased access to MSU's cluster for WMU faculty/grad students (faculty-supervised). CORRECTED Sept 2026 — since this is the same physical MSU ICER cluster, H200/Grace Hopper access applies here too.",
    "networks": [
      "MSU ICER (cross-institution access)"
    ],
    "sources": [
      "https://wmich.edu/research/hpchub",
      "https://icer.msu.edu/news/nvidias-grace-hopper-nodes-arrive-at-msus-data-center-to-push-boundaries-of-computational-research"
    ]
  },
  {
    "state": "Minnesota",
    "school": "University of Minnesota Law School",
    "university": "University of Minnesota",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Agate's mid-cycle upgrade added 80 nodes carrying NVIDIA H100 and L40S accelerators, with the H100s configured for double-precision work and the L40S for single-precision. Alongside these, Agate runs A100 batch GPU nodes, 10 A40 nodes for interactive use via command line or Jupyter, 8 L40S nodes and 15 V100 nodes; an earlier phase added 80 A40 single-precision GPUs for server-side rendering and interactive machine learning.",
    "slurm": "Yes",
    "hpcName": "Minnesota Supercomputing Institute (MSI) — Agate",
    "specs": "Agate is a heterogeneous HPE Linux cluster on AMD 7763 processors with 64-128 CPU cores per node, local solid-state scratch on every node and a high-speed interconnect. In service since Spring 2022, expanded in Summer 2024 with hardware absorbed from the decommissioned Mangi nodes, then upgraded again with the 80-node GPU phase. Storage built on Panasas Ultra with well over 10PB raw. MSI sits in Walter Library on the Twin Cities campus and is part of Research Computing within the Research and Innovation Office, alongside U-Spatial, the Data Science Initiative and the International Institute for Biosensing.",
    "notes": "RECLASSIFIED Sept 2026, modern -> frontier. The row recorded Agate's A100 and A40 hardware, which was right when written, but MSI has since run a mid-cycle upgrade of 80 nodes adding H100 and L40S accelerators — described by MSI as introducing capabilities previously unavailable on its original systems and bridging toward a major system upgrade planned for 2027. The H100s are confirmed operational rather than announced, because MSI's own service unit pricing quotes a live H100 rate (1 SU buys 1.08 H100 GPU-hours, against 6.34 GPU-hours for an A40). Scale context: MSI serves over 900 research groups and four to five thousand unique researcher logins, so contention is real. Historical note — Minnesota was the first US university to buy a supercomputer, a Cray-1 in 1981, and MSI was founded in 1984. CROSS-REFERENCE WORTH FOLLOWING UP: MSI describes itself as serving faculty and researchers at the University AND at other institutions of higher education in Minnesota, with external users charged $3 per service unit. That is a possible, paid pathway for the other two Minnesota law schools in this dataset (Mitchell Hamline, standalone; St. Thomas, none), which neither row currently reflects. It was not verified in this pass and should be checked before either row is changed.",
    "networks": [],
    "sources": [
      "https://msi.umn.edu/msi-research/msi-research-spotlights/agate-upgrade-boosts-ai-research-umn",
      "https://msi.umn.edu/about-msi-services/agate/agate-a100-and-a40-gpus",
      "https://msi.umn.edu/computing/dedicated-computing",
      "https://msi.umn.edu/about-msi-services/high-performance-computing/agate",
      "https://en.wikipedia.org/wiki/University_of_Minnesota_Supercomputing_Institute"
    ]
  },
  {
    "state": "Minnesota",
    "school": "University of St. Thomas School of Law",
    "university": "University of St. Thomas (MN)",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found (no campuswide HPC centre)",
    "specs": "",
    "notes": "Re-verified Sept 2026. St. Thomas is the largest private university in Minnesota and has visible AI activity — a Center for Applied AI directed by Manjeet Rege and a chief data and AI officer — but that is applied/advisory rather than a compute facility. The clearest evidence is a documented case of a St. Thomas chemistry professor building his own HP ProLiant cluster precisely because the university 'doesn't yet have a campuswide high-performance computing center'. STELAR, the unit that mentions applying computing resources to research, is an instructional-technology group (course design, media production, VR). No shared cluster, scheduler or GPU inventory is published. Note the self-built-cluster evidence is from 2015, so this row is worth a periodic re-check; nothing newer indicates a campus HPC service exists. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://edtechmagazine.com/higher/article/2015/10/high-performance-computing-powers-university-research",
      "https://www.stthomas.edu/stelar/about/ourservices/research/",
      "https://stthomas.edu/academics/research/index.html"
    ]
  },
  {
    "state": "Mississippi",
    "school": "University of Mississippi School of Law",
    "university": "University of Mississippi",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "6 nodes with dual NVIDIA L40S GPUs (of 60 total compute nodes)",
    "slurm": "Yes",
    "hpcName": "Mississippi Center for Supercomputing Research (MCSR) - Magnolia cluster",
    "specs": "HPE Apollo 2000 cluster; HDR100 InfiniBand; Bright Cluster Manager",
    "notes": "Serves faculty, staff, students at Ole Miss and free to Mississippi public higher-ed institutions statewide; SLURM confirmed. CORRECTED Sept 2026 — L40S confirmed present.",
    "networks": [],
    "sources": [
      "https://mcsr.olemiss.edu/supercomputers/magnolia/",
      "https://mcsr.olemiss.edu/supercomputers/"
    ]
  },
  {
    "state": "Mississippi",
    "school": "Mississippi Christian University School of Law (formerly Mississippi College School of Law)",
    "university": "Mississippi Christian University (formerly Mississippi College)",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found (MCSR is restricted to public institutions)",
    "specs": "",
    "notes": "Re-verified Sept 2026 against the obvious statewide hypothesis, which fails on eligibility. Mississippi does run a statewide supercomputing center — MCSR, based at the University of Mississippi, whose Maple cluster carries 140 compute nodes, 5,454 CPU cores and a mix of NVIDIA K20, P100 and V100S GPUs — but MCSR states repeatedly and specifically that free accounts are for students, faculty and staff at Mississippi's PUBLIC institutions of higher learning, with account requests required to come from an email address at a Mississippi public IHL. Mississippi Christian University is a private Baptist institution and is therefore outside that eligibility line. No separate Mississippi College research-computing resource was found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row. || UPDATED Oct 2026 (name, not hardware): Mississippi College was renamed Mississippi Christian University on June 1, 2026, and the law school with it; the ABA's approved-school list now carries the entry as \"Mississippi Christian\". The school still goes by MC Law. The rename does not touch accreditation or the MCSR eligibility finding above — it remains a private institution and so outside MCSR's public-IHL line.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://mcsr.olemiss.edu/docs/getting-an-account/",
      "https://mcsr.olemiss.edu/",
      "https://mcsr.olemiss.edu/supercomputers/",
      "https://mcsr.olemiss.edu/apply-for-accounts/",
      "https://magnoliatribune.com/2026/06/02/mississippi-college-now-mississippi-christian-university/"
    ]
  },
  {
    "state": "Missouri",
    "school": "Washington University School of Law",
    "university": "Washington University in St. Louis",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "The Research Computing and Informatics Facility's Center for High-Performance Computing provides over 80 NVIDIA A100 and H100 GPUs alongside over 2,500 CPU cores. RIS separately runs shared GPU and CPU resource pools on the Scientific Compute Platform. The Arts & Sciences condo on Compute2 adds 8 NVIDIA A100 PCIe 80GB across nodes totalling 500+ cores and 8TB RAM.",
    "slurm": "Yes on Compute2, and this changed recently — Compute1 used LSF queues, while Compute2 runs Slurm, with Slurm reservations for dedicated CPU or GPU nodes available from 1 September 2026",
    "hpcName": "RIS Scientific Compute Platform — Compute2 (Compute1 retiring); RCIF/CHPC",
    "specs": "RCIF: multi-petabyte parallel filesystem with off-site disaster recovery, InfiniBand HDR. RIS platforms are container-based (Docker) so software environments deploy independently of one another, with Open OnDemand available for GUI access. All research faculty get 5TB of free Active storage. RIS also runs the WUSTL Secure Enclave supporting NIST SP 800-171 Rev. 2 for CMMC and NIH Genomic Data Sharing, and environments approved for NIH controlled-access data such as dbGaP. RCIF is an NIH Recharge Center with equipment from S10 grants 1S10OD025200-01A1 and 1S10OD030477-01.",
    "notes": "Re-verified Sept 2026 — frontier CONFIRMED (over 80 A100 and H100 GPUs on the RCIF/CHPC side) with an important STRUCTURAL UPDATE this row was missing. WashU has launched Compute2, announced in October 2025 as the next-generation HPC platform, and COMPUTE1 IS BEING DECOMMISSIONED with an official end-of-life of 1 July 2027; new RIS account applications are provisioned on Compute2 by default and condo owners are being migrated in rounds. The scheduler changed with it — Compute1 documentation describes LSF queues, Compute2 uses Slurm. Anyone re-checking this row or following older WashU documentation should work from Compute2. Access is broad: RIS states research data storage and computing services are available to all WashU research faculty, with no disciplinary restriction, so a law faculty member qualifies. Note two distinct organisations are in play — RIS is WashU IT's central service, while the RCIF and its CHPC are a separate research core facility with its own hardware; the 80+ A100/H100 figure belongs to the latter.",
    "networks": [],
    "sources": [
      "https://ris.wustl.edu/systems/",
      "https://ris.wustl.edu/services/compute/",
      "https://research.washu.edu/core-facilities/research-computing-and-informatics-facility/",
      "https://it.wustl.edu/2025/10/available-now-computer2-the-future-of-high-performance-computing-at-washu/",
      "https://it.artsci.washu.edu/questions-and-answers-about-ris"
    ],
    "caveat": "Onboarding is mandatory and costs are metered. Access to Compute2 requires completing an onboarding process — Arts & Sciences documentation states users must either attend a Zoom workshop or watch a training video before access is granted. RIS also publishes cost forecasting tooling for 'computing options, associated costs, and forecasting for future expenses', so this is a chargeback environment rather than a free allocation like Nebraska's or New Mexico's. Confirm pricing before planning a large GPU job."
  },
  {
    "state": "Missouri",
    "school": "Saint Louis University School of Law",
    "university": "Saint Louis University",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "\"Advanced GPU nodes\" confirmed on the new Libra cluster (installed/tested summer 2025, NSF-partially-funded) but no specific model named on SLU's own page; a vendor case-study PDF that might have had exact specs is no longer live (Silicon Mechanics was absorbed into Thinkmate in 2025, and the old whitepaper URL now redirects to a generic page)",
    "slurm": "Yes",
    "hpcName": "Libra (SLU Research Computing Group)",
    "specs": "Libra: 200Gbps network, GPU + high-memory CPU nodes; predecessor Aries: 678 cores, 3.6TB RAM, 56Gb/s EDR InfiniBand",
    "notes": "Free to all SLU faculty and students; NSF grant written specifically to fund \"modern CPU and GPU nodes in support of AI/ML research.\" Re-verified Sept 2026 across three searches plus an attempt to retrieve a vendor spec sheet (now dead) — GPU presence is clearly real and recent (2025) but the specific chip model remains genuinely unpublished.",
    "networks": [],
    "sources": [
      "https://www.slu.edu/research/faculty-resources/research-computing/high-performance-computing.php",
      "https://www.slu.edu/science-and-engineering/academics/computer-science/research.php",
      "https://www.slu.edu/research/faculty-resources/research-computing/index.php"
    ]
  },
  {
    "state": "Missouri",
    "school": "University of Missouri-Columbia School of Law",
    "university": "University of Missouri",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Hellbender launched with 17 GPU nodes of four NVIDIA A100 80GB each (68 A100s), expanded since through researcher investment and by repurposing the newest parts of MU's previous cluster. NOTE: the 8 H100 GPUs sometimes attributed to Missouri sit in a single node of THE MILL at Missouri S&T in Rolla — a different UM System campus — whose 8 GPU nodes comprise 6 with four V100 each, 1 with eight H100 and 1 with two V100, 34 GPUs in total.",
    "slurm": "Yes",
    "hpcName": "Hellbender (MU research cluster) — ITRSS / Division of Research, Innovation & Impact",
    "specs": "Hellbender (built at MU in 2023): initially 112 compute nodes with two AMD 7713 processors and 512GB RAM each, plus 17 GPU nodes with two Intel Xeon 6338 processors and 256GB system RAM alongside their four A100s. Access uses CILogon. Each PI or sponsor can request a 5TB storage group in MU's Research Data Ecosystem through the UM System service portal. The Mill (S&T, also 2023): 229 compute nodes, 15,200 cores.",
    "notes": "RECLASSIFIED Sept 2026, frontier -> modern. This row conflated THREE separate systems and the correction is worth spelling out. (1) It named 'Nautilus', which is not a University of Missouri machine at all — Nautilus is the National Research Platform's hyperconverged GPU cluster, a national NSF resource whose documentation lives at other institutions entirely. (2) It attributed H100s to Missouri; those are in one node of The Mill at MISSOURI S&T in Rolla, a different campus. (3) MU Columbia's actual cluster, Hellbender, is an A100 machine — 17 nodes of four A100 80GB at launch — which is squarely 'modern', not frontier. MU's own ITRSS wiki states plainly that researchers are not limited to one resource but that 'due to differences in funding sources for the resource, some of them have different access rules', so cross-campus access to The Mill cannot be assumed from Columbia. The Division of Research, Innovation and Impact describes Hellbender's mission as accelerating the Mizzou Forward initiative.",
    "networks": [
      "UM System Research Support Solutions"
    ],
    "sources": [
      "https://itrss-wiki.rnet.missouri.edu/_export/raw/pub/hpc/start",
      "https://its.ucsc.edu/services/research-computing/research-specific-computing-and-applications/national-research-platform-nautilus-hyperconverged-gpu-cluster/"
    ],
    "caveat": "Cross-campus access is not established. If a Mizzou law researcher specifically needs H100-class hardware, the nearest UM System option is The Mill at Missouri S&T, and MU's own documentation warns that access rules differ between resources by funding source. Confirm with ITRSS rather than assuming UM System membership grants it — the same failure mode documented for University of Baltimore and UMD's Zaratan elsewhere in this dataset."
  },
  {
    "state": "Missouri",
    "school": "University of Missouri-Kansas City School of Law",
    "university": "University of Missouri-Kansas City",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "SSE ML cluster Research nodes: 8x 48GB NVIDIA GPUs/node (48GB VRAM matches A6000-class or L40-class; exact model name not stated on public page); Teaching nodes: 8x 16GB NVIDIA GPUs/node",
    "slurm": "Yes",
    "hpcName": "UMKC Research Computing (Teaching + SSE ML clusters)",
    "specs": "1 head + 1 login + 7 teaching + 6 research nodes; 128 AMD EPYC cores/research node, 512-768GB RAM",
    "notes": "Real, confirmed SLURM/OpenHPC-based clusters serving UMKC research community. CORRECTED Sept 2026 — 48GB VRAM size strongly implies A6000/L40-class GPU (exact model name still not published).",
    "networks": [],
    "sources": [
      "https://www.umkc.edu/is/resources/research-computing.html"
    ]
  },
  {
    "state": "Montana",
    "school": "University of Montana School of Law",
    "university": "University of Montana",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Hellgate cluster, confirmed via live node-by-node table: 12 nodes x 4x NVIDIA RTX2080Ti (48 total, legacy); 1 node x 4x NVIDIA A40; 12 nodes x 4x NVIDIA A4500 Ada Generation (48 total) — ~25 GPU nodes total",
    "slurm": "Likely (standard for NSF CC* clusters)",
    "hpcName": "Hellgate Shared Computing Cluster",
    "specs": "45 CPU compute nodes + ~25 GPU nodes; grant-funded priority tiers (Reincarnation, Brinkerhoff, Martens, GSCC, INBRE); SLURM confirmed",
    "notes": "Free/general access via \"gpu(all)\" and \"gpu(safe)\" partitions, though most nodes carry grant-funded priority tiers. CORRECTED Sept 2026 — the resources page itself is a live, complete, per-node hardware table (not a snippet), confirming A40 and A4500 Ada Generation GPUs directly rather than the vaguer \"GPU nodes\" description found in earlier research.",
    "networks": [],
    "sources": [
      "https://www.umt.edu/it/rci/hellgate/resources/default.php",
      "https://www.umt.edu/it/rci/hellgate/"
    ]
  },
  {
    "state": "Nebraska",
    "school": "University of Nebraska College of Law",
    "university": "University of Nebraska-Lincoln",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "PLUMAGE: 6 NVIDIA H200 GPUs (3 nodes x 2) and 52 NVIDIA L40S 48GB GPUs (13 nodes x 4) available in Swan's general-access gpu partition. PLUMAGE was procured as 15 L40S nodes in total; the two not in Swan are allocated to the National Research Platform, which is how the published '52 L40S' and a 15-node purchase reconcile.",
    "slurm": "Yes — Swan uses a traditional scheduler-based batch system. PLUMAGE is unusual in offering a second, non-Slurm route: the same GPUs can be reached through the National Research Platform's Kubernetes-managed environment, and PLUMAGE can shift GPUs between the two.",
    "hpcName": "Holland Computing Center (HCC) - Swan cluster + new PLUMAGE GPU infrastructure",
    "specs": "PLUMAGE funded by a National Science Foundation grant of nearly $700,000. Swan is HCC's flagship general-purpose cluster with 256GB RAM per node. HCC operates from two sites — the Peter Kiewit Institute in Omaha and the Schorr Center in Lincoln — and also runs Red, which stores over 7PB for the CMS project.",
    "notes": "Re-verified Sept 2026 — the figures in this row are EXACT and now reconciled against two levels of source. The headline '6 H200 + 52 L40S' appears verbatim in UNL's own announcement and in trade coverage; HCC's December 2025 newsletter explains the underlying arithmetic (fifteen four-GPU L40S nodes procured, thirteen live in Swan's general access gpu partition for 52 cards, two diverted to the National Research Platform, plus three nodes of two H200 each). PLUMAGE stands for Promoting Learning Using Mixed Advanced GPU Environments. Access is strong for a law faculty member on two counts. First, it is free: HCC states PLUMAGE is available at no cost to NU researchers, with access arrangeable even for collaborators outside the NU system. Second, the framing is explicitly cross-disciplinary — UNL describes PLUMAGE as benefiting 'researchers across the disciplinary spectrum, from computer scientists and engineers to scholars in business, architecture, the social sciences and more', which is unusually inclusive language for a GPU resource. HCC serves the whole University of Nebraska system (UNK, UNL, UNMC, UNO), consistent with this row's existing network tag. HCC also runs webinars on requesting NAIRR Pilot resources, so staff there are an informed route to national allocations as well as local ones.",
    "networks": [
      "University of Nebraska System (HCC)"
    ],
    "sources": [
      "https://news.hcc.unl.edu/publications/hcc-newsletter-december-2025",
      "https://news.unl.edu/article/hollands-new-nsf-funded-resource-to-boost-ai-machine-learning-research",
      "https://newsroom.unl.edu/announce/cse/20335/107976",
      "https://hcc.unl.edu/hcc",
      "https://www.hpcwire.com/off-the-wire/university-of-nebraska-deploys-nsf-funded-plumage-gpu-infrastructure-for-ai-research/"
    ]
  },
  {
    "state": "Nebraska",
    "school": "Creighton University School of Law",
    "university": "Creighton University",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "",
    "slurm": "Not confirmed",
    "hpcName": "Creighton High Performance Computing (Division of IT)",
    "specs": "Not published — DoIT lists an HPC FAQ, an 'Add Software to HPC' request path and an 'HPC Cluster Metrics' page, but no node, core or storage counts are public",
    "notes": "RECLASSIFIED Sept 2026 (was 'No HPC resource identified' — that was wrong). Creighton's Division of Information Technology explicitly states it offers High Performance Research Computing to faculty, staff and students university-wide, with consulting from Creighton's RaD Lab team. The DoIT navigation carries dedicated HPC, HPC FAQ, Add Software to HPC and HPC Cluster Metrics pages, and the same site lists Law among the university's schools. A 2018 Creighton posting for a Research Computing Analyst reporting to a Faculty Chair of Research Computing describes a shared, faculty-federated research computing environment including schedulers and Linux server environments, which corroborates a real cluster. Downgraded only to 'undetailed' rather than a GPU tier because no public page names a GPU model, node count or scheduler. Note separately that Nebraska's statewide Holland Computing Center (HCC) is a University of Nebraska system resource and Creighton is a private Jesuit university, so HCC is not an access path here.",
    "networks": [],
    "sources": [
      "https://doit.creighton.edu/HighPerformanceComputing",
      "https://www.txgen.tamu.edu/jobs/bioinformatics-research-computing-analyst-school-of-medicine"
    ]
  },
  {
    "state": "Nevada",
    "school": "William S. Boyd School of Law",
    "university": "University of Nevada Las Vegas",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "RebelX GPU cluster (NSF MRI #2117941-funded): 48 GPUs total — 16x NVIDIA A30 + 16x NVIDIA A40 + 16x NVIDIA A100. Separately, the older Cherry Creek/Cherry Creek II \"supercomputer\" (Intel/Switch SUPERNAP, since 2013) uses Intel Xeon Phi coprocessors, not GPUs, despite being widely publicized as a top-tier system.",
    "slurm": "Likely (job postings confirm SLURM expertise required)",
    "hpcName": "RebelX (National Supercomputing Institute) + Cherry Creek II",
    "specs": "RebelX open to any UNLV or Las Vegas-community AI researcher/educator via account request; Cherry Creek II: 26,000 cores, 500 TFLOPS theoretical",
    "notes": "Request an account via rebelx.dataxlab.org for AI/deep-learning research; open to UNLV and the wider Las Vegas community. CORRECTED Sept 2026 — this was previously miscategorized under \"no GPU confirmed\" language associated with Cherry Creek's Xeon Phi hardware; RebelX is the real GPU resource and independently confirmed across three sources.",
    "networks": [],
    "sources": [
      "https://www.unlv.edu/news/unlvtoday/unlv-gpu-cluster-rebelx-available-ai-ressearch-and-education",
      "https://www.unlv.edu/engineering/research/big-data",
      "https://rebelx.dataxlab.org/",
      "https://msiexchange.nasa.gov/attachments/182281/Capability_Statement_UNLV Cyber Security_122022 Accessible (FY26 - 508) 2.pdf"
    ],
    "caveat": "UNLV's most publicized \"supercomputer,\" Cherry Creek/Cherry Creek II, is genuinely a top-tier system by Top500/Green500 standards, but its accelerators are Intel Xeon Phi coprocessors (many-core x86, deprecated by Intel since ~2020), not GPUs. The actual GPU hardware — A30/A40/A100 — lives on a separate, less-publicized resource (RebelX) confirmed via a 2023 announcement and cross-referenced against the RebelX site itself."
  },
  {
    "state": "New Hampshire",
    "school": "University of New Hampshire School of Law",
    "university": "University of New Hampshire",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Premise GPU inventory as published by UNH: 9 compute nodes with two NVIDIA K80 cards each (18 K80), 1 node with a single V100, 14-15 nodes with a single A100 each, and 1 node with a single NVIDIA H100. Total GPU contribution roughly 216 TFLOPS before Tensor Cores, which add a further ~147 TFLOPS across the A100s and ~30 on the H100.",
    "slurm": "Yes",
    "hpcName": "Research Computing Center (RCC) - Premise cluster (USNH system-wide)",
    "specs": "Premise is a Dell cluster in the RCC/Lenharth Data Center at Durham, built on a condo buy-in model: core infrastructure (racks, power, cooling, network, file storage and six servers) is funded centrally by ET&S and REEO for USNH researchers, with three standard node configurations quoted at roughly $20k base, $25k hi-ram and $30k+ GPU. RCC also runs BioMade (live 2019, 24 nodes, 228TB) and marvin, a Cray acquired under a 2020 NSF MRI (AGS-1919310). RCC has about 30 staff and a 2,000 sq ft research data centre, reporting to the CIO and Senior Vice Provost for Research.",
    "notes": "Re-verified Sept 2026 — modern CONFIRMED, and an apparent inconsistency resolved. This row looked wrong at a glance because its visible hardware string began with K80 and V100, both 'older' tier parts. The truncated remainder is what carries the tier: the '14 nodes' each hold a single NVIDIA A100, which is squarely modern. UNH's published performance arithmetic also shows one H100 node, so a single Hopper card exists — but by the same reasoning applied to Hawai'i in this project, one card in a buy-in cluster does not make an institution a frontier site, so the row stays modern with the H100 noted rather than promoted. Access is reasonable for an unfunded researcher: Premise is a buy-in cluster, but RCC states the model 'enables researchers without funding to access the cluster' through a shared job queue available to all Premise users, with a Technology Governance Committee reviewing usage to keep availability equitable. Owners get a restricted priority queue on hardware they bought.",
    "networks": [],
    "sources": [
      "https://premise.sr.unh.edu/",
      "https://www.unh.edu/research/research-computing-center",
      "https://www.usnh.edu/it/departments/research-computing-software-development",
      "https://www.nhepscor.org/resource/premise-hpc-cluster"
    ],
    "caveat": "Buy-in cluster, and the law school is on a different campus. Priority goes to node owners; unfunded users share a general queue, so wait times depend on owner demand. Note also that UNH Franklin Pierce School of Law is in Concord while Premise sits in Durham — USNH operates data centres at Durham, Keene, Plymouth, Concord and Manchester and the infrastructure is described as serving USNH researchers generally, so this should not be a barrier, but a Concord-based researcher may want to confirm with RCC."
  },
  {
    "state": "New Jersey",
    "school": "Rutgers Law School - Newark",
    "university": "Rutgers University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Amarel's long-published figure is 124 NVIDIA GPUs of Tesla, Volta and Titan models. The current Clark phase, available to the research community from January 2026, adds NVIDIA 3x L40S GPU nodes, which is what places this row in the modern tier.",
    "slurm": "Yes",
    "hpcName": "OARC - Amarel cluster (+ campus-specific PRICE cluster at Newark)",
    "specs": "Approximately 600 nodes providing 15,500 compute cores, with memory configurations from 128GB to 1.5TB including Intel Optane persistent memory, all connected over Mellanox InfiniBand FDR or EDR. Clark nodes use dual Intel Xeon Gold 6538Y+ (Emerald Rapids) 32-core processors (64 cores per node) with TruDDR5 memory, doubling standard node RAM from 256GB to 512GB. Amarel spans three Rutgers data centres — Piscataway, Newark and Camden — as a single unified system. Node ownership is priced at $11,435 per CPU node, $18,735 per high-memory node and $30,755 per 3x L40S GPU node, each including 1TB of complimentary storage on a four-year warranted lifetime.",
    "notes": "Re-verified Sept 2026 — modern CONFIRMED. Amarel is a condominium-style cluster launched in July 2017, named for Saul Amarel, a founder of the Rutgers Computer Science Department. Since 2023 each hardware phase is named after a New Jersey city in alphabetical order; the dataset is current as of the transition from Bayonne to CLARK. Access is genuinely open: Rutgers describes Amarel as providing FREE ACCESS FOR RESEARCHERS ACROSS DISCIPLINES, available to all current Rutgers students, staff and faculty including professors emeritus, via a single access-request form, with centrally-funded general-access partitions named main, gpu, mem and nonpre. Buying nodes is optional and buys priority, not entry. Jobs are managed with SLURM. Newark hosts one of Amarel's three data centres, so this is a same-campus resource for Rutgers Law Newark.",
    "networks": [
      "Rutgers multi-campus (Amarel)"
    ],
    "sources": [
      "https://oarc.rutgers.edu/resources/amarel/",
      "https://it.rutgers.edu/2026/01/06/introducing-clark-the-next-generation-of-amarel-and-research-computing/",
      "https://sites.google.com/view/cluster-user-guide/amarel",
      "https://it.rutgers.edu/research-computing/amarel/",
      "https://research.rutgers.edu/system/files/2021-08/OARC%20Facilities%20and%20Equipment.pdf"
    ],
    "caveat": "Guest accounts do not reach the central GPUs. Rutgers distinguishes sharply between members and guests: a guest affiliated with an owning department may use that host's partitions but is explicitly denied access to the University's centrally-funded main, gpu, mem and nonpre partitions. A law faculty member with a current Rutgers appointment is a full member and unaffected, but visiting scholars and external collaborators are not."
  },
  {
    "state": "New Jersey",
    "school": "Rutgers Law School - Camden",
    "university": "Rutgers University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Amarel's long-published figure is 124 NVIDIA GPUs of Tesla, Volta and Titan models. The current Clark phase, available to the research community from January 2026, adds NVIDIA 3x L40S GPU nodes, which is what places this row in the modern tier.",
    "slurm": "Yes",
    "hpcName": "OARC - Amarel cluster",
    "specs": "Approximately 600 nodes providing 15,500 compute cores, with memory configurations from 128GB to 1.5TB including Intel Optane persistent memory, all connected over Mellanox InfiniBand FDR or EDR. Clark nodes use dual Intel Xeon Gold 6538Y+ (Emerald Rapids) 32-core processors (64 cores per node) with TruDDR5 memory, doubling standard node RAM from 256GB to 512GB. Amarel spans three Rutgers data centres — Piscataway, Newark and Camden — as a single unified system. Node ownership is priced at $11,435 per CPU node, $18,735 per high-memory node and $30,755 per 3x L40S GPU node, each including 1TB of complimentary storage on a four-year warranted lifetime.",
    "notes": "Re-verified Sept 2026 — modern CONFIRMED. Amarel is a condominium-style cluster launched in July 2017, named for Saul Amarel, a founder of the Rutgers Computer Science Department. Since 2023 each hardware phase is named after a New Jersey city in alphabetical order; the dataset is current as of the transition from Bayonne to CLARK. Access is genuinely open: Rutgers describes Amarel as providing FREE ACCESS FOR RESEARCHERS ACROSS DISCIPLINES, available to all current Rutgers students, staff and faculty including professors emeritus, via a single access-request form, with centrally-funded general-access partitions named main, gpu, mem and nonpre. Buying nodes is optional and buys priority, not entry. Jobs are managed with SLURM. One Camden-specific advantage worth recording: OARC states that a researcher or student affiliated with RU-Camden also has access to AMAREL-C, the Camden-based subset of Amarel's compute and storage, in addition to the university-wide resources.",
    "networks": [
      "Rutgers multi-campus (Amarel)"
    ],
    "sources": [
      "https://oarc.rutgers.edu/resources/amarel/",
      "https://it.rutgers.edu/2026/01/06/introducing-clark-the-next-generation-of-amarel-and-research-computing/",
      "https://sites.google.com/view/cluster-user-guide/amarel",
      "https://it.rutgers.edu/research-computing/amarel/",
      "https://research.rutgers.edu/system/files/2021-08/OARC%20Facilities%20and%20Equipment.pdf"
    ],
    "caveat": "Guest accounts do not reach the central GPUs. Rutgers distinguishes sharply between members and guests: a guest affiliated with an owning department may use that host's partitions but is explicitly denied access to the University's centrally-funded main, gpu, mem and nonpre partitions. A law faculty member with a current Rutgers appointment is a full member and unaffected, but visiting scholars and external collaborators are not."
  },
  {
    "state": "New Jersey",
    "school": "Seton Hall University School of Law",
    "university": "Seton Hall University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found (research computing is cloud-based, no named cluster)",
    "specs": "",
    "notes": "Re-verified Sept 2026. Seton Hall's CIO describes the university as enhancing its research computing infrastructure through innovative use of cloud computing, providing advanced computational resources, high-performance computing clusters and scalable data storage. That is a stated direction rather than a documented facility: no cluster is named, no node, core or GPU inventory is published, no scheduler is identified, and there is no account-request page. Both relevant academic units — the Department of Mathematics and Computer Science and the Stillman School's Department of Computing and Decision Sciences — describe research in software engineering, programming languages, symbolic computation and data science without reference to a shared cluster. Kept at 'none' because there is nothing a law faculty member could actually request, but re-check this one: a cloud-backed HPC offering could become real and visible at short notice. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.shu.edu/technology/message-cio.html",
      "https://www.shu.edu/math-compsci/faculty.html",
      "https://www.shu.edu/computing/index.html"
    ]
  },
  {
    "state": "New Mexico",
    "school": "University of New Mexico School of Law",
    "university": "University of New Mexico",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Easley: 36 NVIDIA L40S + 8 NVIDIA H100. Hopper: 37 NVIDIA A100. Xena: NVIDIA K40 (legacy). Figures match CARC's own infrastructure page and its GitHub organisation README exactly.",
    "slurm": "Yes",
    "hpcName": "Center for Advanced Research Computing (CARC) - Easley/Hopper/Xena clusters",
    "specs": "Easley (mixed CPU/GPU, Dell, replaced Wheeler): 65 compute nodes, 4,160 CPU cores, 23.3TB RAM, NVIDIA NDR 800 Gbps InfiniBand core with 200 Gbps to nodes, 768TB raw all-flash parallel filesystem plus IBM Storage. Hopper (condo cluster): 61 nodes, 2,176 cores, HDR 400 Gbps InfiniBand. Xena: 32 nodes, 720 cores, FDR 56 Gbps. Housed in CARC's dedicated research data center with 270 kVa UPS and 70 tons of cooling, on UNM's Science DMZ with 100G links to ESnet and the Western Regional Network via the Albuquerque Gigapop. CARC has operated since 1994.",
    "notes": "Re-verified Sept 2026 — the figures in this row are EXACT, matching CARC's infrastructure page and its GitHub README independently. Easley is named for Annie Easley, the NASA computer programmer and mathematician. Access is excellent and explicitly free: CARC describes itself as providing high-performance computing, large-scale storage and expert support to UNM faculty, staff and students FREE OF CHARGE, courtesy of the Office of the Vice President for Research, with free baseline services, premium options and grant-writing boilerplate offered to researchers. The Hopper condo model is researcher-friendly in both directions — OVPR funded a set of nodes that the whole UNM computational research community may use free of charge, and buy-in researchers' nodes are available to others when idle. Leadership note for future re-checks: Tyson Swetnam became CARC Director in 2026, succeeding Patrick Bridges.",
    "networks": [],
    "sources": [
      "https://carc.unm.edu/systems/carc-infrastructure.html",
      "https://github.com/UNM-CARC",
      "https://carc.unm.edu/news--events/News/easley-cluster.html",
      "https://carc.unm.edu/news--events/News/carc-launches-hopper.html",
      "https://carc.unm.edu/about-carc/facilities-description.html"
    ]
  },
  {
    "state": "New York",
    "school": "NYU School of Law",
    "university": "New York University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Torch (NYU's new flagship, Oct 2025): 232x NVIDIA H200 (80GB, 29 nodes x8) + 272x NVIDIA L40S (68 nodes x4) + 172x NVIDIA A100 (43 nodes, mixed 4/node configs) + 60x NVIDIA H100 (15 nodes x4) + 16x NVIDIA RTX Pro 6000 (2 nodes x8) = 752 GPUs total. Predecessor Greene cluster (292x RTX8000 + 40x V100) remains available. NYU is also an Empire AI founding member (144x H100/H200 \"Alpha\" + 288x B200 \"Beta\" statewide).",
    "slurm": "Yes",
    "hpcName": "Torch (NYU HPC) — successor to Greene — plus Empire AI allocation",
    "specs": "Torch: 43,072 total CPU cores (Intel Xeon Platinum 8592+), 280.1TB total memory, Infiniband NDR400; 10.79 PFLOPS realized on LINPACK (12.25 PFLOPS theoretical) — #133 on Top500 (June 2025), #40 on Green500; 5x more powerful than Greene",
    "notes": "Allocated by stakeholder group: Tandon School of Engineering (112 H200+60 H100+60 A100), Courant/CS (88 H200+52 L40S), Center for Data Science, Chemistry, CILVR lab, and a 232-GPU \"IT Public\" general-access pool (24 H200 + 208 L40S) open campus-wide, which is the path a Law School researcher would use. SLURM confirmed. CORRECTED Sept 2026 — added Torch, NYU's new Oct-2025 flagship cluster, after being pointed to the official announcement; this replaces the earlier Greene-only + Empire AI framing as NYU's primary resource. || NETWORK TAG ADDED Sept 2026: this row already named NYU's Empire AI membership in its resource and hardware fields but carried no `networks` entry, so the site's shared-networks filter did not group NYU with Columbia, Cornell and CUNY. NYU is confirmed as one of the seven FOUNDING members of Empire AI (alongside SUNY, CUNY, Columbia, Cornell, RPI and the Flatiron Institute), in a consortium that now totals ten with the addition of the University of Rochester, Rochester Institute of Technology and the Icahn School of Medicine at Mount Sinai. Tag added so the filter is accurate. Worth noting the consortium is closed at those ten — that is precisely why the unaffiliated New York law schools in this dataset (Brooklyn, New York Law School, Albany) cannot reach Empire AI directly.",
    "networks": [
      "Empire AI (New York State)"
    ],
    "sources": [
      "https://github.com/NYU-RTS/rts-docs/blob/main/docs/hpc/10_spec_sheet.md",
      "https://www.nyu.edu/about/news-publications/news/2025/october/nyu-unveils--torch---the-most-powerful-supercomputer-in-new-york.html",
      "https://www.datacenterdynamics.com/en/news/new-york-university-unveils-torch-supercomputer/",
      "https://sites.google.com/nyu.edu/nyu-hpc/hpc-systems/greene",
      "https://www.empireai.edu/about/",
      "https://www.suny.edu/suny-news/press-releases/1-26/1-30-26-2/empireai.html"
    ],
    "caveat": "Press coverage (Data Center Dynamics, insideHPC) reported \"500+ NVIDIA H200 GPUs,\" but NYU's own technical spec sheet gives an exact, internally-consistent table showing 232 H200 GPUs specifically (752 GPUs total across all types) — the press number appears to conflate the H200 count with the cluster's total GPU count. This entry uses NYU's own table since it reconciles exactly (29 nodes x 8 H200/node = 232, and 232+272+172+60+16 = 752, matching the table's own total row)."
  },
  {
    "state": "New York",
    "school": "Cornell Law School",
    "university": "Cornell University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "G2 cluster exposes Slurm constraints for a100 (40/80GB) and h100 (94GB) alongside older cards, with gpu-low/gpu-mid/gpu-high feature tiers. Through Empire AI, Cornell PIs can also request the extended 'Alpha' machine (144 H100 GPUs) and 'Beta' (288 B200 GPUs with Grace CPU nodes for data processing). Weill Cornell's separate SCU core runs 251 NVIDIA GPUs (H200, H100, A100, L40S, A40).",
    "slurm": "Yes",
    "hpcName": "Center for Advanced Computing (CAC) — G2 cluster + Red Cloud; Empire AI (Alpha/Beta)",
    "specs": "G2: 9,500+ CPU cores; Red Cloud 1.9PB Ceph. Weill SCU: 26PB Lustre/GPFS, 200Gb/s InfiniBand, 1.5TB RAM high-memory nodes. Empire AI SUs priced at 1 SU per H100-hour, 2 SU per B200-hour.",
    "notes": "RECLASSIFIED Sept 2026 — previously scored 'older' on a '755 GPUs ... T4/V100' line that is now out of date. Cornell's G2 cluster documents A100 and H100 Slurm constraints directly, and Cornell is one of nine New York State institutions in the Empire AI consortium. Cornell has run open calls for Empire AI allocations explicitly stating that any faculty member with PI status from Cornell, Cornell Tech or Weill Cornell Medicine can apply, with SU costs covered by the Provost through November 2026 and expected to run about $0.50/SU afterwards — well below commercial H100 rates. That is an unusually clean access path for a non-STEM faculty member, which is exactly what this directory is trying to surface.",
    "networks": [
      "Empire AI (New York State)"
    ],
    "sources": [
      "https://it.coecis.cornell.edu/researchit/g2cluster/g2-faqs/",
      "https://cvg.cornell.edu/?p=241",
      "https://scu.med.cornell.edu/",
      "https://www.cac.cornell.edu/services/HPCsystems.aspx"
    ],
    "caveat": "The Empire AI application route is open to 'any faculty member with PI status', which on its face includes Law, but the calls circulate through science units and no page confirms a Law School allocation has been made. CAC also hosts privately procured group clusters that are not generally accessible. Worth confirming with CAC before relying on it."
  },
  {
    "state": "New York",
    "school": "Columbia Law School",
    "university": "Columbia University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "230 physical GPUs: NVIDIA A6000 (x8 and x4 nodes), H100, L40, L40S; Columbia is also an Empire AI founding member, with access to the 288x NVIDIA B200 'Beta' machine now online in addition to Empire AI's existing H100/H200",
    "slurm": "Yes",
    "hpcName": "CUIT Shared Research Computing - Insomnia cluster (+ Ginsburg/legacy)",
    "specs": "90 nodes; 7,144 physical cores (14,288 logical); 58.3TiB RAM; 2.282PB GPFS storage",
    "notes": "Free tier available university-wide (SRCPAC-governed); also participates in NY state 'Empire AI' initiative; buy-in for dedicated nodes UPDATED Sept 2026 — added confirmation of Empire AI's new B200 Beta machine.",
    "networks": [
      "Empire AI (New York State)"
    ],
    "sources": [
      "https://www.cuit.columbia.edu/shared-research-computing-facility",
      "https://www.empireai.edu/about/",
      "https://foci.rpi.edu/computing-resources/empire-ai"
    ]
  },
  {
    "state": "New York",
    "school": "Syracuse University College of Law",
    "university": "Syracuse University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "GPUs distributed across OrangeGrid/Zest (formerly branded \"SUrge\"): NVIDIA H100 (80GB), A100 (80GB), L40S (48GB), RTX A6000 (48GB), A40 (48GB), RTX 6000 (24GB) — 250+ GPUs total",
    "slurm": "Yes",
    "hpcName": "Research Computing - Zest (HPC) + SUrge (GPU cluster) + OrangeGrid (HTC)",
    "specs": "48,000+ CPU cores total across HPC/HTC systems; free and open to all Syracuse-affiliated researchers",
    "notes": "NSF CC*-funded GPU cluster; available to Syracuse research community, includes undergrad access. CORRECTED Sept 2026 — H100 80GB confirmed present; GPUs are now organized under OrangeGrid/Zest rather than a standalone \"SUrge\" resource.",
    "networks": [],
    "sources": [
      "https://github.com/SyracuseUniversity/ResearchComputing/pull/2",
      "https://research.syracuse.edu/wp-content/uploads/2024/06/Research-Computing-Resources.pdf"
    ]
  },
  {
    "state": "New York",
    "school": "Maurice A. Deane School of Law (Hofstra)",
    "university": "Hofstra University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "NVIDIA HGX H100 and A100 GPUs confirmed (approx. 20 H100 across SXM/PCIe nodes + 16 A100 + 2 A30, per cluster hardware breakdown); AMD EPYC/Intel Xeon CPUs, 7TB+ combined RAM",
    "slurm": "Yes",
    "hpcName": "Star HPC Cluster",
    "specs": "HDR200 Gb/s InfiniBand; 63TB all-flash parallel storage (4 HPE PFSS nodes)",
    "notes": "$1.5M investment (Hofstra + NSF); Bright Cluster Manager + SLURM + Apptainer/Singularity; available to faculty and students. CORRECTED Sept 2026 — H100 and A100 confirmed present (previously listed as unspecified).",
    "networks": [],
    "sources": [
      "https://docs.starhpc.hofstra.io/quickstart/about-star.html",
      "https://starhpc.hofstra.io/",
      "https://www.gpusperstudent.org/data/master_data.csv"
    ]
  },
  {
    "state": "New York",
    "school": "CUNY School of Law",
    "university": "City University of New York",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "CUNY's own Arrow cluster: legacy K20m (62 nodes) + 2 nodes w/ 8x NVIDIA GPU (SXM, model unspecified); PLUS full CUNY-wide allocation on Empire AI: 144x NVIDIA H100 (Alpha) + H200 nodes, with a 288x NVIDIA B200 'Beta' machine coming online",
    "slurm": "Likely (standard for these systems)",
    "hpcName": "CUNY High Performance Computing Center (HPCC) + Empire AI allocation",
    "specs": "CUNY HPCC: 1000s of cores across Penzias/Salk/Arrow systems; Empire AI: dedicated 400Gb/s NDR InfiniBand, 2PB DDN storage",
    "notes": "Free to all CUNY faculty, researchers, staff, students across all 26 colleges incl. CUNY Law; SLURM confirmed. CORRECTED Sept 2026 — CUNY-wide Empire AI allocation gives access to H100/H200 now and B200 (Blackwell) imminently, well beyond the campus's own older Arrow cluster.",
    "networks": [
      "Empire AI (New York State)"
    ],
    "sources": [
      "https://www.csi.cuny.edu/academics-and-research/research-centers/cuny-high-performance-computing-center",
      "https://foci.rpi.edu/computing-resources/empire-ai",
      "https://ai.cornell.edu/empire-ai-cornell-call-for-compute-resource-proposals"
    ]
  },
  {
    "state": "Ohio",
    "school": "The Ohio State University Moritz College of Law",
    "university": "The Ohio State University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "128 NVIDIA H100 (Hopper) GPUs on Cardinal — 94GB HBM2e each with NVLink, four per node across 32 Dell PowerEdge XE9640 nodes. 776 NVIDIA A100s on Ascend (four each on 24 nodes, three each on 84 nodes, two each on 214 nodes). NVIDIA V100 32GB on Pitzer.",
    "slurm": "Yes — Slurm on Cardinal (RHEL 9); PBS compatibility disabled on Cardinal",
    "hpcName": "Ohio Supercomputer Center (OSC) — Cardinal, Ascend, Pitzer",
    "specs": "Cardinal: 378 Dell nodes, 39,312 cores, Intel Xeon CPU Max 9470 (Sapphire Rapids) with 128GB HBM2e and 512GB DDR5 per dense node, NDR200 InfiniBand (400Gb/s on GPU nodes), RHEL 9. Ascend: 322-node Dell AMD EPYC machine, 40,448 cores, 16.0 PFLOPS theoretical peak. Roughly 20PB storage plus 23PB expandable backup. Open OnDemand originated at OSC.",
    "sources": [
      "https://www.osc.edu/supercomputing/support/account",
      "https://www.osc.edu/services/cluster_computing",
      "https://www.osc.edu/resources/technical_support/supercomputers/cardinal",
      "https://www.osc.edu/content/available_service_categories",
      "https://casc.org/member/ohio-supercomputer-center-osc/"
    ],
    "notes": "ADDED Sept 2026. Ohio was one of sixteen states missing entirely from this directory. Access for The Ohio State University runs through the Ohio Supercomputer Center, which is the strongest statewide access story in this dataset — stronger than LONI or MCSR, both of which failed on eligibility when tested. OSC's stated mission is to serve Ohio higher education institutions 'both public and private', and an eligible PI is any full-time, permanent academic researcher, tenure-track faculty member or emeritus professor at an Ohio college or university. Ohio academic clients receive highly subsidised access with fees generally charged only above certain thresholds, and classroom use is always free. Critically, that eligibility rule is written at the level of the institution and the appointment, not the discipline — OSC explicitly says it reaches out to academic disciplines that have not traditionally used HPC. A law faculty member appears to qualify on its face. OSC is a member of the Ohio Department of Higher Education's technology consortium and sits alongside Ohio State's own research computing; OSU's campus-level resources were not separately verified in this pass.",
    "caveat": "Statewide resource; campus-level resources not separately checked. OSC eligibility is confirmed at the institution level, but OSC notes an Ohio academic client's request 'may need to be approved by their host institution', which is an approval step this row cannot verify. This row also does not reflect any cluster the university may run itself — that was out of scope for this pass and should be checked separately before the row is considered complete.",
    "networks": [
      "OSC (Ohio statewide)"
    ]
  },
  {
    "state": "Ohio",
    "school": "Case Western Reserve University School of Law",
    "university": "Case Western Reserve University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "128 NVIDIA H100 (Hopper) GPUs on Cardinal — 94GB HBM2e each with NVLink, four per node across 32 Dell PowerEdge XE9640 nodes. 776 NVIDIA A100s on Ascend (four each on 24 nodes, three each on 84 nodes, two each on 214 nodes). NVIDIA V100 32GB on Pitzer.",
    "slurm": "Yes — Slurm on Cardinal (RHEL 9); PBS compatibility disabled on Cardinal",
    "hpcName": "Ohio Supercomputer Center (OSC) — Cardinal, Ascend, Pitzer",
    "specs": "Cardinal: 378 Dell nodes, 39,312 cores, Intel Xeon CPU Max 9470 (Sapphire Rapids) with 128GB HBM2e and 512GB DDR5 per dense node, NDR200 InfiniBand (400Gb/s on GPU nodes), RHEL 9. Ascend: 322-node Dell AMD EPYC machine, 40,448 cores, 16.0 PFLOPS theoretical peak. Roughly 20PB storage plus 23PB expandable backup. Open OnDemand originated at OSC.",
    "sources": [
      "https://www.osc.edu/supercomputing/support/account",
      "https://www.osc.edu/services/cluster_computing",
      "https://www.osc.edu/resources/technical_support/supercomputers/cardinal",
      "https://www.osc.edu/content/available_service_categories",
      "https://casc.org/member/ohio-supercomputer-center-osc/"
    ],
    "notes": "ADDED Sept 2026. Ohio was one of sixteen states missing entirely from this directory. Access for Case Western Reserve University runs through the Ohio Supercomputer Center, which is the strongest statewide access story in this dataset — stronger than LONI or MCSR, both of which failed on eligibility when tested. OSC's stated mission is to serve Ohio higher education institutions 'both public and private', and an eligible PI is any full-time, permanent academic researcher, tenure-track faculty member or emeritus professor at an Ohio college or university. Ohio academic clients receive highly subsidised access with fees generally charged only above certain thresholds, and classroom use is always free. Critically, that eligibility rule is written at the level of the institution and the appointment, not the discipline — OSC explicitly says it reaches out to academic disciplines that have not traditionally used HPC. A law faculty member appears to qualify on its face. Case Western is a private institution, which matters here — OSC's public-and-private mission is what makes this row work, where an equivalent private school in Mississippi or Louisiana would score 'none'.",
    "caveat": "Statewide resource; campus-level resources not separately checked. OSC eligibility is confirmed at the institution level, but OSC notes an Ohio academic client's request 'may need to be approved by their host institution', which is an approval step this row cannot verify. This row also does not reflect any cluster the university may run itself — that was out of scope for this pass and should be checked separately before the row is considered complete.",
    "networks": [
      "OSC (Ohio statewide)"
    ]
  },
  {
    "state": "Ohio",
    "school": "University of Cincinnati College of Law",
    "university": "University of Cincinnati",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "128 NVIDIA H100 (Hopper) GPUs on Cardinal — 94GB HBM2e each with NVLink, four per node across 32 Dell PowerEdge XE9640 nodes. 776 NVIDIA A100s on Ascend (four each on 24 nodes, three each on 84 nodes, two each on 214 nodes). NVIDIA V100 32GB on Pitzer.",
    "slurm": "Yes — Slurm on Cardinal (RHEL 9); PBS compatibility disabled on Cardinal",
    "hpcName": "Ohio Supercomputer Center (OSC) — Cardinal, Ascend, Pitzer",
    "specs": "Cardinal: 378 Dell nodes, 39,312 cores, Intel Xeon CPU Max 9470 (Sapphire Rapids) with 128GB HBM2e and 512GB DDR5 per dense node, NDR200 InfiniBand (400Gb/s on GPU nodes), RHEL 9. Ascend: 322-node Dell AMD EPYC machine, 40,448 cores, 16.0 PFLOPS theoretical peak. Roughly 20PB storage plus 23PB expandable backup. Open OnDemand originated at OSC.",
    "sources": [
      "https://www.osc.edu/supercomputing/support/account",
      "https://www.osc.edu/services/cluster_computing",
      "https://www.osc.edu/resources/technical_support/supercomputers/cardinal",
      "https://www.osc.edu/content/available_service_categories",
      "https://casc.org/member/ohio-supercomputer-center-osc/"
    ],
    "notes": "ADDED Sept 2026. Ohio was one of sixteen states missing entirely from this directory. Access for University of Cincinnati runs through the Ohio Supercomputer Center, which is the strongest statewide access story in this dataset — stronger than LONI or MCSR, both of which failed on eligibility when tested. OSC's stated mission is to serve Ohio higher education institutions 'both public and private', and an eligible PI is any full-time, permanent academic researcher, tenure-track faculty member or emeritus professor at an Ohio college or university. Ohio academic clients receive highly subsidised access with fees generally charged only above certain thresholds, and classroom use is always free. Critically, that eligibility rule is written at the level of the institution and the appointment, not the discipline — OSC explicitly says it reaches out to academic disciplines that have not traditionally used HPC. A law faculty member appears to qualify on its face.",
    "caveat": "Statewide resource; campus-level resources not separately checked. OSC eligibility is confirmed at the institution level, but OSC notes an Ohio academic client's request 'may need to be approved by their host institution', which is an approval step this row cannot verify. This row also does not reflect any cluster the university may run itself — that was out of scope for this pass and should be checked separately before the row is considered complete.",
    "networks": [
      "OSC (Ohio statewide)"
    ]
  },
  {
    "state": "Ohio",
    "school": "Cleveland State University College of Law",
    "university": "Cleveland State University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "128 NVIDIA H100 (Hopper) GPUs on Cardinal — 94GB HBM2e each with NVLink, four per node across 32 Dell PowerEdge XE9640 nodes. 776 NVIDIA A100s on Ascend (four each on 24 nodes, three each on 84 nodes, two each on 214 nodes). NVIDIA V100 32GB on Pitzer.",
    "slurm": "Yes — Slurm on Cardinal (RHEL 9); PBS compatibility disabled on Cardinal",
    "hpcName": "Ohio Supercomputer Center (OSC) — Cardinal, Ascend, Pitzer",
    "specs": "Cardinal: 378 Dell nodes, 39,312 cores, Intel Xeon CPU Max 9470 (Sapphire Rapids) with 128GB HBM2e and 512GB DDR5 per dense node, NDR200 InfiniBand (400Gb/s on GPU nodes), RHEL 9. Ascend: 322-node Dell AMD EPYC machine, 40,448 cores, 16.0 PFLOPS theoretical peak. Roughly 20PB storage plus 23PB expandable backup. Open OnDemand originated at OSC.",
    "sources": [
      "https://www.osc.edu/supercomputing/support/account",
      "https://www.osc.edu/services/cluster_computing",
      "https://www.osc.edu/resources/technical_support/supercomputers/cardinal",
      "https://www.osc.edu/content/available_service_categories",
      "https://casc.org/member/ohio-supercomputer-center-osc/"
    ],
    "notes": "ADDED Sept 2026. Ohio was one of sixteen states missing entirely from this directory. Access for Cleveland State University runs through the Ohio Supercomputer Center, which is the strongest statewide access story in this dataset — stronger than LONI or MCSR, both of which failed on eligibility when tested. OSC's stated mission is to serve Ohio higher education institutions 'both public and private', and an eligible PI is any full-time, permanent academic researcher, tenure-track faculty member or emeritus professor at an Ohio college or university. Ohio academic clients receive highly subsidised access with fees generally charged only above certain thresholds, and classroom use is always free. Critically, that eligibility rule is written at the level of the institution and the appointment, not the discipline — OSC explicitly says it reaches out to academic disciplines that have not traditionally used HPC. A law faculty member appears to qualify on its face.",
    "caveat": "Statewide resource; campus-level resources not separately checked. OSC eligibility is confirmed at the institution level, but OSC notes an Ohio academic client's request 'may need to be approved by their host institution', which is an approval step this row cannot verify. This row also does not reflect any cluster the university may run itself — that was out of scope for this pass and should be checked separately before the row is considered complete.",
    "networks": [
      "OSC (Ohio statewide)"
    ]
  },
  {
    "state": "Ohio",
    "school": "University of Akron School of Law",
    "university": "University of Akron",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "128 NVIDIA H100 (Hopper) GPUs on Cardinal — 94GB HBM2e each with NVLink, four per node across 32 Dell PowerEdge XE9640 nodes. 776 NVIDIA A100s on Ascend (four each on 24 nodes, three each on 84 nodes, two each on 214 nodes). NVIDIA V100 32GB on Pitzer.",
    "slurm": "Yes — Slurm on Cardinal (RHEL 9); PBS compatibility disabled on Cardinal",
    "hpcName": "Ohio Supercomputer Center (OSC) — Cardinal, Ascend, Pitzer",
    "specs": "Cardinal: 378 Dell nodes, 39,312 cores, Intel Xeon CPU Max 9470 (Sapphire Rapids) with 128GB HBM2e and 512GB DDR5 per dense node, NDR200 InfiniBand (400Gb/s on GPU nodes), RHEL 9. Ascend: 322-node Dell AMD EPYC machine, 40,448 cores, 16.0 PFLOPS theoretical peak. Roughly 20PB storage plus 23PB expandable backup. Open OnDemand originated at OSC.",
    "sources": [
      "https://www.osc.edu/supercomputing/support/account",
      "https://www.osc.edu/services/cluster_computing",
      "https://www.osc.edu/resources/technical_support/supercomputers/cardinal",
      "https://www.osc.edu/content/available_service_categories",
      "https://casc.org/member/ohio-supercomputer-center-osc/"
    ],
    "notes": "ADDED Sept 2026. Ohio was one of sixteen states missing entirely from this directory. Access for University of Akron runs through the Ohio Supercomputer Center, which is the strongest statewide access story in this dataset — stronger than LONI or MCSR, both of which failed on eligibility when tested. OSC's stated mission is to serve Ohio higher education institutions 'both public and private', and an eligible PI is any full-time, permanent academic researcher, tenure-track faculty member or emeritus professor at an Ohio college or university. Ohio academic clients receive highly subsidised access with fees generally charged only above certain thresholds, and classroom use is always free. Critically, that eligibility rule is written at the level of the institution and the appointment, not the discipline — OSC explicitly says it reaches out to academic disciplines that have not traditionally used HPC. A law faculty member appears to qualify on its face.",
    "caveat": "Statewide resource; campus-level resources not separately checked. OSC eligibility is confirmed at the institution level, but OSC notes an Ohio academic client's request 'may need to be approved by their host institution', which is an approval step this row cannot verify. This row also does not reflect any cluster the university may run itself — that was out of scope for this pass and should be checked separately before the row is considered complete.",
    "networks": [
      "OSC (Ohio statewide)"
    ]
  },
  {
    "state": "Ohio",
    "school": "University of Toledo College of Law",
    "university": "University of Toledo",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "128 NVIDIA H100 (Hopper) GPUs on Cardinal — 94GB HBM2e each with NVLink, four per node across 32 Dell PowerEdge XE9640 nodes. 776 NVIDIA A100s on Ascend (four each on 24 nodes, three each on 84 nodes, two each on 214 nodes). NVIDIA V100 32GB on Pitzer.",
    "slurm": "Yes — Slurm on Cardinal (RHEL 9); PBS compatibility disabled on Cardinal",
    "hpcName": "Ohio Supercomputer Center (OSC) — Cardinal, Ascend, Pitzer",
    "specs": "Cardinal: 378 Dell nodes, 39,312 cores, Intel Xeon CPU Max 9470 (Sapphire Rapids) with 128GB HBM2e and 512GB DDR5 per dense node, NDR200 InfiniBand (400Gb/s on GPU nodes), RHEL 9. Ascend: 322-node Dell AMD EPYC machine, 40,448 cores, 16.0 PFLOPS theoretical peak. Roughly 20PB storage plus 23PB expandable backup. Open OnDemand originated at OSC.",
    "sources": [
      "https://www.osc.edu/supercomputing/support/account",
      "https://www.osc.edu/services/cluster_computing",
      "https://www.osc.edu/resources/technical_support/supercomputers/cardinal",
      "https://www.osc.edu/content/available_service_categories",
      "https://casc.org/member/ohio-supercomputer-center-osc/"
    ],
    "notes": "ADDED Sept 2026. Ohio was one of sixteen states missing entirely from this directory. Access for University of Toledo runs through the Ohio Supercomputer Center, which is the strongest statewide access story in this dataset — stronger than LONI or MCSR, both of which failed on eligibility when tested. OSC's stated mission is to serve Ohio higher education institutions 'both public and private', and an eligible PI is any full-time, permanent academic researcher, tenure-track faculty member or emeritus professor at an Ohio college or university. Ohio academic clients receive highly subsidised access with fees generally charged only above certain thresholds, and classroom use is always free. Critically, that eligibility rule is written at the level of the institution and the appointment, not the discipline — OSC explicitly says it reaches out to academic disciplines that have not traditionally used HPC. A law faculty member appears to qualify on its face.",
    "caveat": "Statewide resource; campus-level resources not separately checked. OSC eligibility is confirmed at the institution level, but OSC notes an Ohio academic client's request 'may need to be approved by their host institution', which is an approval step this row cannot verify. This row also does not reflect any cluster the university may run itself — that was out of scope for this pass and should be checked separately before the row is considered complete.",
    "networks": [
      "OSC (Ohio statewide)"
    ]
  },
  {
    "state": "Ohio",
    "school": "University of Dayton School of Law",
    "university": "University of Dayton",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "128 NVIDIA H100 (Hopper) GPUs on Cardinal — 94GB HBM2e each with NVLink, four per node across 32 Dell PowerEdge XE9640 nodes. 776 NVIDIA A100s on Ascend (four each on 24 nodes, three each on 84 nodes, two each on 214 nodes). NVIDIA V100 32GB on Pitzer.",
    "slurm": "Yes — Slurm on Cardinal (RHEL 9); PBS compatibility disabled on Cardinal",
    "hpcName": "Ohio Supercomputer Center (OSC) — Cardinal, Ascend, Pitzer",
    "specs": "Cardinal: 378 Dell nodes, 39,312 cores, Intel Xeon CPU Max 9470 (Sapphire Rapids) with 128GB HBM2e and 512GB DDR5 per dense node, NDR200 InfiniBand (400Gb/s on GPU nodes), RHEL 9. Ascend: 322-node Dell AMD EPYC machine, 40,448 cores, 16.0 PFLOPS theoretical peak. Roughly 20PB storage plus 23PB expandable backup. Open OnDemand originated at OSC.",
    "sources": [
      "https://www.osc.edu/supercomputing/support/account",
      "https://www.osc.edu/services/cluster_computing",
      "https://www.osc.edu/resources/technical_support/supercomputers/cardinal",
      "https://www.osc.edu/content/available_service_categories",
      "https://casc.org/member/ohio-supercomputer-center-osc/"
    ],
    "notes": "ADDED Sept 2026. Ohio was one of sixteen states missing entirely from this directory. Access for University of Dayton runs through the Ohio Supercomputer Center, which is the strongest statewide access story in this dataset — stronger than LONI or MCSR, both of which failed on eligibility when tested. OSC's stated mission is to serve Ohio higher education institutions 'both public and private', and an eligible PI is any full-time, permanent academic researcher, tenure-track faculty member or emeritus professor at an Ohio college or university. Ohio academic clients receive highly subsidised access with fees generally charged only above certain thresholds, and classroom use is always free. Critically, that eligibility rule is written at the level of the institution and the appointment, not the discipline — OSC explicitly says it reaches out to academic disciplines that have not traditionally used HPC. A law faculty member appears to qualify on its face. Dayton is private; see the Case Western note on why OSC's public-and-private mission matters.",
    "caveat": "Statewide resource; campus-level resources not separately checked. OSC eligibility is confirmed at the institution level, but OSC notes an Ohio academic client's request 'may need to be approved by their host institution', which is an approval step this row cannot verify. This row also does not reflect any cluster the university may run itself — that was out of scope for this pass and should be checked separately before the row is considered complete.",
    "networks": [
      "OSC (Ohio statewide)"
    ]
  },
  {
    "state": "Ohio",
    "school": "Capital University Law School",
    "university": "Capital University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "128 NVIDIA H100 (Hopper) GPUs on Cardinal — 94GB HBM2e each with NVLink, four per node across 32 Dell PowerEdge XE9640 nodes. 776 NVIDIA A100s on Ascend (four each on 24 nodes, three each on 84 nodes, two each on 214 nodes). NVIDIA V100 32GB on Pitzer.",
    "slurm": "Yes — Slurm on Cardinal (RHEL 9); PBS compatibility disabled on Cardinal",
    "hpcName": "Ohio Supercomputer Center (OSC) — Cardinal, Ascend, Pitzer",
    "specs": "Cardinal: 378 Dell nodes, 39,312 cores, Intel Xeon CPU Max 9470 (Sapphire Rapids) with 128GB HBM2e and 512GB DDR5 per dense node, NDR200 InfiniBand (400Gb/s on GPU nodes), RHEL 9. Ascend: 322-node Dell AMD EPYC machine, 40,448 cores, 16.0 PFLOPS theoretical peak. Roughly 20PB storage plus 23PB expandable backup. Open OnDemand originated at OSC.",
    "sources": [
      "https://www.osc.edu/supercomputing/support/account",
      "https://www.osc.edu/services/cluster_computing",
      "https://www.osc.edu/resources/technical_support/supercomputers/cardinal",
      "https://www.osc.edu/content/available_service_categories",
      "https://casc.org/member/ohio-supercomputer-center-osc/"
    ],
    "notes": "ADDED Sept 2026. Ohio was one of sixteen states missing entirely from this directory. Access for Capital University runs through the Ohio Supercomputer Center, which is the strongest statewide access story in this dataset — stronger than LONI or MCSR, both of which failed on eligibility when tested. OSC's stated mission is to serve Ohio higher education institutions 'both public and private', and an eligible PI is any full-time, permanent academic researcher, tenure-track faculty member or emeritus professor at an Ohio college or university. Ohio academic clients receive highly subsidised access with fees generally charged only above certain thresholds, and classroom use is always free. Critically, that eligibility rule is written at the level of the institution and the appointment, not the discipline — OSC explicitly says it reaches out to academic disciplines that have not traditionally used HPC. A law faculty member appears to qualify on its face. Capital is a small private university in Columbus with no research-computing centre of its own that this pass found; OSC is the entire access story here, which makes the statewide eligibility rule unusually load-bearing for this row.",
    "caveat": "Statewide resource; campus-level resources not separately checked. OSC eligibility is confirmed at the institution level, but OSC notes an Ohio academic client's request 'may need to be approved by their host institution', which is an approval step this row cannot verify. This row also does not reflect any cluster the university may run itself — that was out of scope for this pass and should be checked separately before the row is considered complete.",
    "networks": [
      "OSC (Ohio statewide)"
    ]
  },
  {
    "state": "Texas",
    "school": "University of Texas at Austin School of Law",
    "university": "University of Texas at Austin",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Vista is built on NVIDIA Grace Hopper (GH200) and Grace Superchips — an Arm-based departure from the x86 architecture of Frontera, Stampede and Lonestar6 — connected with NVIDIA NDR InfiniBand at up to 400 Gbps. Frontera and Lonestar6 carry additional GPU subsystems with separately allocated GPU node-hours.",
    "slurm": "Yes",
    "hpcName": "Texas Advanced Computing Center (TACC) — Vista, Frontera, Lonestar6, Ranch",
    "specs": "TACC is a research centre of UT Austin itself, so this is an in-house resource rather than a consortium. Ranch tape archive fronted by roughly 30PB of disk cache. Vista is positioned as the bridge between Frontera (the current NSF leadership-class system) and Horizon, the planned primary system of the NSF Leadership-Class Computing Facility.",
    "sources": [
      "https://tacc.utexas.edu/use-tacc/allocations/",
      "https://tacc.utexas.edu/systems/vista/",
      "https://frontera-portal.tacc.utexas.edu/allocations/policy/",
      "https://tacc.utexas.edu/use-tacc/getting-started/"
    ],
    "notes": "ADDED Sept 2026. Texas had exactly one entry in this directory (South Texas College of Law Houston, a standalone school listed separately) despite having around ten ABA-accredited law schools — UT Austin, the state's flagship, was missing. TACC is a research centre of UT Austin, so a UT Austin law faculty member is an in-house user rather than a guest. Two distinct allocation routes exist: TxRAS handles requests from UT Austin, UT System and TACC Partner Institutions, while Vista allocations for UT and UT System researchers go through UT-Austin's Center for Generative AI on a monthly rolling basis. Frontera is separately and broadly available to any US researcher doing open research, with the expectation that work results in publication in a broadly available journal or conference — a condition legal scholarship can generally meet.",
    "caveat": "Allocation is competitive and framed around science and engineering. Frontera's policy describes a project PI as 'typically a faculty member or research scientist at a US-based academic institution' and its tracks (LRAC, Pathways, LSCP) are written around code readiness and scaling, with Pathways requiring demonstrated existing research funding for code and algorithm development. Startup and Director Discretionary allocations are the realistic entry points. No evidence was found of a law-school allocation.",
    "networks": [
      "TACC / UT System (Texas)"
    ]
  },
  {
    "state": "Pennsylvania",
    "school": "Penn State Dickinson Law (University Park; formerly Penn State Law)",
    "university": "Pennsylvania State University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "NVIDIA A100 (single-GPU, half-card and Multi-Instance GPU slices), A40, V100 and P100 on Roar Collab — all four listed in ICDS's own user guide and priced on its current rate sheet",
    "slurm": "Yes — Roar Collab uses Slurm (the older Roar/ACI-b used PBS)",
    "hpcName": "Roar Collab (Institute for Computational and Data Sciences)",
    "specs": "Basic, Standard and High Memory core types; Red Hat Enterprise Linux 8; software stack of roughly 240 applications updated on a schedule. Housed in a University Park data centre with 2.4MW redundant power and 12,000 sq ft of environmentally controlled space, about half of which is dedicated to ICDS infrastructure. GPU pricing includes the Standard cores needed alongside the GPU — 28 for P100, 24 for a full A100, 12 for a half A100.",
    "sources": [
      "https://icds.psu.edu/services/roar/roar-collab-cluster/",
      "https://icds.psu.edu/funding/new-read-credits/",
      "https://www.icds.psu.edu/about/",
      "https://icds.psu.edu/services/roar/details-rates/",
      "https://sites.psu.edu/cnmg/hpc/",
      "https://icds.psu.edu/services/roar/read-credits/",
      "https://docs.icds.psu.edu/getting-started/tips-for-beginners/",
      "https://www.americanbar.org/groups/legal_education/accreditation/approved-law-schools/alphabetical/",
      "https://www.psu.edu/news/academics/story/aba-approves-unified-law-school-penn-state-dickinson-law-and-penn-state-law",
      "https://www.abajournal.com/web/article/aba-council-gives-thumbs-up-to-penn-state-plan-to-unify-its-law-schools"
    ],
    "notes": "ADDED Sept 2026. Pennsylvania was missing entirely from this directory despite having roughly nine ABA-accredited law schools. Penn State's Institute for Computational and Data Sciences runs Roar Collab, and ICDS states that it supports faculty from dozens of departments in nearly all of Penn State's twenty-four campuses — which is what brings both of Penn State's law schools into scope. Every user receives an automatic monthly allocation of READ (Research Exploration And Discovery) credits usable on all hardware types including GPUs, with no action required. UPDATED Oct 2026: ICDS's READ Credits page (revised Sept 29, 2026) now sets that allotment at 3 credits a month, intended for exploration and testing, and says no additional READ credits are available for the rest of FY27 — the request route this row previously described is closed for the year. One credit is one Basic core-month and costs about $3 if purchased. Note that 'Roar' and 'Roar Collab' are different systems: Roar is a restricted-data environment available only through a paid allocation, while Roar Collab is the general cluster most users should use. UPDATED Oct 2026 (structure, not hardware): Penn State's two law schools are being reunified. The ABA's council approved a single, unified, two-location school called Penn State Dickinson Law in November 2024, with its primary location in Carlisle and an additional location at University Park; the first unified class enrolled in fall 2025. The ABA's approved-school list now carries one entry (\"Penn State-Dickinson Law\"). This directory keeps one row per location, as it does for Rutgers Camden and Newark, because the compute route is the same university-wide ICDS cluster either way.",
    "caveat": "GPU generation tops out at A100/A40, and the credit economics matter. Roar Collab lists A100, A40, V100 and P100 — no Hopper-class GPU was found — so this is 'modern', not frontier. Free GPU time is now minimal: ICDS's READ Credits page (revised Sept 29, 2026) gives every account 3 credits a month for exploration and testing, and says no additional READ credits are available for the rest of FY27, pointing researchers to their college or department instead. Plan on a paid allocation (a full A100 is $294.83 a month) or purchased credits ($3.11 each) for real projects.",
    "networks": [
      "Penn State ICDS (multi-campus)"
    ]
  },
  {
    "state": "Pennsylvania",
    "school": "Penn State Dickinson Law (Carlisle)",
    "university": "Pennsylvania State University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "NVIDIA A100 (single-GPU, half-card and Multi-Instance GPU slices), A40, V100 and P100 on Roar Collab — all four listed in ICDS's own user guide and priced on its current rate sheet",
    "slurm": "Yes — Roar Collab uses Slurm (the older Roar/ACI-b used PBS)",
    "hpcName": "Roar Collab (Institute for Computational and Data Sciences)",
    "specs": "Basic, Standard and High Memory core types; Red Hat Enterprise Linux 8; software stack of roughly 240 applications updated on a schedule. Housed in a University Park data centre with 2.4MW redundant power and 12,000 sq ft of environmentally controlled space, about half of which is dedicated to ICDS infrastructure. GPU pricing includes the Standard cores needed alongside the GPU — 28 for P100, 24 for a full A100, 12 for a half A100.",
    "sources": [
      "https://icds.psu.edu/services/roar/roar-collab-cluster/",
      "https://icds.psu.edu/funding/new-read-credits/",
      "https://www.icds.psu.edu/about/",
      "https://icds.psu.edu/services/roar/details-rates/",
      "https://sites.psu.edu/cnmg/hpc/",
      "https://icds.psu.edu/services/roar/read-credits/",
      "https://docs.icds.psu.edu/getting-started/tips-for-beginners/",
      "https://www.americanbar.org/groups/legal_education/accreditation/approved-law-schools/alphabetical/",
      "https://www.psu.edu/news/academics/story/aba-approves-unified-law-school-penn-state-dickinson-law-and-penn-state-law",
      "https://www.abajournal.com/web/article/aba-council-gives-thumbs-up-to-penn-state-plan-to-unify-its-law-schools"
    ],
    "notes": "ADDED Sept 2026. Pennsylvania was missing entirely from this directory despite having roughly nine ABA-accredited law schools. Penn State's Institute for Computational and Data Sciences runs Roar Collab, and ICDS states that it supports faculty from dozens of departments in nearly all of Penn State's twenty-four campuses — which is what brings both of Penn State's law schools into scope. Every user receives an automatic monthly allocation of READ (Research Exploration And Discovery) credits usable on all hardware types including GPUs, with no action required. UPDATED Oct 2026: ICDS's READ Credits page (revised Sept 29, 2026) now sets that allotment at 3 credits a month, intended for exploration and testing, and says no additional READ credits are available for the rest of FY27 — the request route this row previously described is closed for the year. One credit is one Basic core-month and costs about $3 if purchased. Note that 'Roar' and 'Roar Collab' are different systems: Roar is a restricted-data environment available only through a paid allocation, while Roar Collab is the general cluster most users should use. UPDATED Oct 2026 (structure, not hardware): Penn State's two law schools are being reunified. The ABA's council approved a single, unified, two-location school called Penn State Dickinson Law in November 2024, with its primary location in Carlisle and an additional location at University Park; the first unified class enrolled in fall 2025. The ABA's approved-school list now carries one entry (\"Penn State-Dickinson Law\"). Carlisle is the primary location. This directory keeps one row per location, as it does for Rutgers Camden and Newark, because the compute route is the same university-wide ICDS cluster either way.",
    "caveat": "GPU generation tops out at A100/A40, and the credit economics matter. Roar Collab lists A100, A40, V100 and P100 — no Hopper-class GPU was found — so this is 'modern', not frontier. Free GPU time is now minimal: ICDS's READ Credits page (revised Sept 29, 2026) gives every account 3 credits a month for exploration and testing, and says no additional READ credits are available for the rest of FY27, pointing researchers to their college or department instead. Plan on a paid allocation (a full A100 is $294.83 a month) or purchased credits ($3.11 each) for real projects.",
    "networks": [
      "Penn State ICDS (multi-campus)"
    ]
  },
  {
    "state": "Oklahoma",
    "school": "University of Oklahoma College of Law",
    "university": "University of Oklahoma",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "OneOCII-RAML: 7 nodes each configured with an NVIDIA HGX A100 4-GPU board — 28 A100 GPUs total. A further CC* grant is expected to fund 10 NVIDIA H100 cards plus their servers; OSCER's own page describes these as anticipated rather than installed, so they are NOT counted toward this row's tier.",
    "slurm": "Not confirmed — OSCER's public system page does not name its scheduler",
    "hpcName": "OSCER (OU Supercomputing Center for Education & Research) — OneOCII-RAML",
    "specs": "OSCER's supercomputer runs Enterprise Linux 9 (mostly Rocky Linux 9), with some nodes still on EL7 during transition. Interconnect is mixed: Mellanox FDR10 40 Gbps (3:1 oversubscribed) on Haswell and Broadwell nodes, HDR100 100 Gbps (4:1) elsewhere, with an EDR 100 Gbps super-core linking the two fabrics; some condo nodes have no InfiniBand at the owner's discretion. OneOCII-RAML reaches the state through the OneOklahoma Friction Free Network (OFFN), a 10 and 100 Gbps research network now spanning 21 campuses.",
    "sources": [
      "https://www.oneocii.okepscor.org/about/",
      "https://www.oscer.ou.edu/supercomputer/",
      "https://ui.adsabs.harvard.edu/abs/2022nsf....2201561E/abstract",
      "https://www.oneocii.okepscor.org/services/",
      "https://www.oneocii.okepscor.org/news/oklahoma-receives-2-1-million-from-national-science-foundation-to-expand-research-networking-and-computing-statewide/"
    ],
    "notes": "ADDED Sept 2026 — Oklahoma was one of the states missing entirely from this directory. Access runs through the OneOklahoma Cyberinfrastructure Initiative (OneOCII), which is the second genuinely inclusive statewide programme found in this project after Ohio's OSC. OneOCII describes itself as a statewide ALL-INCLUSIVE advanced digital services collaboration, and its stated aims are to reach institutions outside the mainstream of advanced computing and to SERVE EVERY HIGHER EDUCATION INSTITUTION IN OKLAHOMA that has relevant curricula. It has reached over 100 institutions and organisations statewide — more than 50 academic, including PhD-granting universities, primarily undergraduate institutions, community colleges and career techs — and runs over 1,000 HPC users at 20+ Oklahoma institutions. Crucially for this directory, the OneOCII-RAML grant abstract states the system supports educational programming to grow AI/ML use in research BEYOND COMPUTER SCIENCE DISCIPLINES and is a state-wide resource for non-commercial research institutions across Oklahoma. OSCER also runs the long-standing Oklahoma Supercomputing Symposium and the 'Supercomputing in Plain English' workshop series, which is an unusually low-jargon on-ramp for a researcher outside the computational sciences. OU is OSCER's home campus, so this is an on-campus resource and the strongest of the three Oklahoma rows. OSCER's Research IT group also writes letters of commitment and facilities descriptions for grant proposals, and its personnel can join as Co-PI or consultant.",
    "caveat": "Statewide eligibility is broad but the compute is modest, and the H100s are not here yet. OneOCII's reach is real, but OneOCII-RAML is a 7-node, 28-GPU machine shared across a whole state, and its stated purpose is weighted toward STEM (weather and climate AI, high-energy physics). The 10 H100 cards are anticipated grant-funded hardware, not installed capacity — the same announced-versus-operational distinction that made Arizona's row wrong. Non-OU institutions reach it through OFFN and OneOCII rather than by direct account, so confirm the route with OSCER before relying on it.",
    "networks": [
      "OneOCII (Oklahoma statewide)"
    ]
  },
  {
    "state": "Oklahoma",
    "school": "Oklahoma City University School of Law",
    "university": "Oklahoma City University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "OneOCII-RAML: 7 nodes each configured with an NVIDIA HGX A100 4-GPU board — 28 A100 GPUs total. A further CC* grant is expected to fund 10 NVIDIA H100 cards plus their servers; OSCER's own page describes these as anticipated rather than installed, so they are NOT counted toward this row's tier.",
    "slurm": "Not confirmed — OSCER's public system page does not name its scheduler",
    "hpcName": "OSCER (OU Supercomputing Center for Education & Research) — OneOCII-RAML",
    "specs": "OSCER's supercomputer runs Enterprise Linux 9 (mostly Rocky Linux 9), with some nodes still on EL7 during transition. Interconnect is mixed: Mellanox FDR10 40 Gbps (3:1 oversubscribed) on Haswell and Broadwell nodes, HDR100 100 Gbps (4:1) elsewhere, with an EDR 100 Gbps super-core linking the two fabrics; some condo nodes have no InfiniBand at the owner's discretion. OneOCII-RAML reaches the state through the OneOklahoma Friction Free Network (OFFN), a 10 and 100 Gbps research network now spanning 21 campuses.",
    "sources": [
      "https://www.oneocii.okepscor.org/about/",
      "https://www.oscer.ou.edu/supercomputer/",
      "https://ui.adsabs.harvard.edu/abs/2022nsf....2201561E/abstract",
      "https://www.oneocii.okepscor.org/services/",
      "https://www.oneocii.okepscor.org/news/oklahoma-receives-2-1-million-from-national-science-foundation-to-expand-research-networking-and-computing-statewide/"
    ],
    "notes": "ADDED Sept 2026 — Oklahoma was one of the states missing entirely from this directory. Access runs through the OneOklahoma Cyberinfrastructure Initiative (OneOCII), which is the second genuinely inclusive statewide programme found in this project after Ohio's OSC. OneOCII describes itself as a statewide ALL-INCLUSIVE advanced digital services collaboration, and its stated aims are to reach institutions outside the mainstream of advanced computing and to SERVE EVERY HIGHER EDUCATION INSTITUTION IN OKLAHOMA that has relevant curricula. It has reached over 100 institutions and organisations statewide — more than 50 academic, including PhD-granting universities, primarily undergraduate institutions, community colleges and career techs — and runs over 1,000 HPC users at 20+ Oklahoma institutions. Crucially for this directory, the OneOCII-RAML grant abstract states the system supports educational programming to grow AI/ML use in research BEYOND COMPUTER SCIENCE DISCIPLINES and is a state-wide resource for non-commercial research institutions across Oklahoma. OSCER also runs the long-standing Oklahoma Supercomputing Symposium and the 'Supercomputing in Plain English' workshop series, which is an unusually low-jargon on-ramp for a researcher outside the computational sciences. OCU is a private institution. OneOCII's all-inclusive framing is what brings it into scope — the same public/private distinction that excluded private Mississippi College from MCSR. This row rests on the statewide policy rather than on any OCU resource of its own, which was not found.",
    "caveat": "Statewide eligibility is broad but the compute is modest, and the H100s are not here yet. OneOCII's reach is real, but OneOCII-RAML is a 7-node, 28-GPU machine shared across a whole state, and its stated purpose is weighted toward STEM (weather and climate AI, high-energy physics). The 10 H100 cards are anticipated grant-funded hardware, not installed capacity — the same announced-versus-operational distinction that made Arizona's row wrong. Non-OU institutions reach it through OFFN and OneOCII rather than by direct account, so confirm the route with OSCER before relying on it.",
    "networks": [
      "OneOCII (Oklahoma statewide)"
    ]
  },
  {
    "state": "Oklahoma",
    "school": "University of Tulsa College of Law",
    "university": "University of Tulsa",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "OneOCII-RAML: 7 nodes each configured with an NVIDIA HGX A100 4-GPU board — 28 A100 GPUs total. A further CC* grant is expected to fund 10 NVIDIA H100 cards plus their servers; OSCER's own page describes these as anticipated rather than installed, so they are NOT counted toward this row's tier.",
    "slurm": "Not confirmed — OSCER's public system page does not name its scheduler",
    "hpcName": "OSCER (OU Supercomputing Center for Education & Research) — OneOCII-RAML",
    "specs": "OSCER's supercomputer runs Enterprise Linux 9 (mostly Rocky Linux 9), with some nodes still on EL7 during transition. Interconnect is mixed: Mellanox FDR10 40 Gbps (3:1 oversubscribed) on Haswell and Broadwell nodes, HDR100 100 Gbps (4:1) elsewhere, with an EDR 100 Gbps super-core linking the two fabrics; some condo nodes have no InfiniBand at the owner's discretion. OneOCII-RAML reaches the state through the OneOklahoma Friction Free Network (OFFN), a 10 and 100 Gbps research network now spanning 21 campuses.",
    "sources": [
      "https://www.oneocii.okepscor.org/about/",
      "https://www.oscer.ou.edu/supercomputer/",
      "https://ui.adsabs.harvard.edu/abs/2022nsf....2201561E/abstract",
      "https://www.oneocii.okepscor.org/services/",
      "https://www.oneocii.okepscor.org/news/oklahoma-receives-2-1-million-from-national-science-foundation-to-expand-research-networking-and-computing-statewide/"
    ],
    "notes": "ADDED Sept 2026 — Oklahoma was one of the states missing entirely from this directory. Access runs through the OneOklahoma Cyberinfrastructure Initiative (OneOCII), which is the second genuinely inclusive statewide programme found in this project after Ohio's OSC. OneOCII describes itself as a statewide ALL-INCLUSIVE advanced digital services collaboration, and its stated aims are to reach institutions outside the mainstream of advanced computing and to SERVE EVERY HIGHER EDUCATION INSTITUTION IN OKLAHOMA that has relevant curricula. It has reached over 100 institutions and organisations statewide — more than 50 academic, including PhD-granting universities, primarily undergraduate institutions, community colleges and career techs — and runs over 1,000 HPC users at 20+ Oklahoma institutions. Crucially for this directory, the OneOCII-RAML grant abstract states the system supports educational programming to grow AI/ML use in research BEYOND COMPUTER SCIENCE DISCIPLINES and is a state-wide resource for non-commercial research institutions across Oklahoma. OSCER also runs the long-standing Oklahoma Supercomputing Symposium and the 'Supercomputing in Plain English' workshop series, which is an unusually low-jargon on-ramp for a researcher outside the computational sciences. TU is a private institution reached through OneOCII rather than through any TU cluster, which was not found in this pass. Note that Oral Roberts University in Tulsa is a named participant in OneOCII regional CC* awards, confirming that private Tulsa-area institutions do participate.",
    "caveat": "Statewide eligibility is broad but the compute is modest, and the H100s are not here yet. OneOCII's reach is real, but OneOCII-RAML is a 7-node, 28-GPU machine shared across a whole state, and its stated purpose is weighted toward STEM (weather and climate AI, high-energy physics). The 10 H100 cards are anticipated grant-funded hardware, not installed capacity — the same announced-versus-operational distinction that made Arizona's row wrong. Non-OU institutions reach it through OFFN and OneOCII rather than by direct account, so confirm the route with OSCER before relying on it.",
    "networks": [
      "OneOCII (Oklahoma statewide)"
    ]
  },
  {
    "state": "North Carolina",
    "school": "UNC School of Law",
    "university": "University of North Carolina at Chapel Hill",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Longleaf runs dedicated GPU partitions. UNC Research Computing's GPU page records the FY25/26 acquisition as three nodes of eight RTX Pro 6000 GPUs each (24 cards), described as a follow-on to the L40S with similar compute capacity and twice the GPU memory. RTX PRO 6000 sits in the modern tier under CONTRIBUTING.md's table, not frontier.",
    "slurm": "Yes — Slurm, with Open OnDemand as a web portal alternative",
    "specs": "Longleaf has over 300 nodes with dedicated GPU partitions and high-performance storage, optimised for memory- and I/O-intensive, loosely coupled workloads and high aggregate job throughput rather than individual job speed — well suited to many-small-jobs work. New hardware is added annually. Storage: home on /nas/longleaf/home, with /hickory/users as the default user space since 14 August 2026 at 2TB and 3 million files per user (not backed up), plus /work for high-throughput data-intensive computing. Open OnDemand provides job submission, file browsing and preconfigured applications without command-line work.",
    "sources": [
      "https://help.rc.unc.edu/getting-started-on-longleaf/",
      "https://help.rc.unc.edu/gpu/",
      "https://www.unc.edu/discover/7-decades-of-difference-carolinas-1st-computer-and-its-current-cluster-2/"
    ],
    "notes": "ADDED Sept 2026 — North Carolina was missing entirely from this directory despite having six ABA-accredited law schools. Access at UNC is broad and clearly stated: Research Computing describes Longleaf as available to RESEARCHERS, STUDENTS, FACULTY AND STAFF ACROSS THE UNIVERSITY, with no disciplinary restriction, so a law faculty member qualifies on its face. Open OnDemand matters here too — it gives browser-based access to preconfigured applications, which lowers the barrier for someone who does not work at a shell prompt. Longleaf's design emphasis on throughput over single-job speed suits the kind of many-small-jobs work common in empirical legal research (running a model across many jurisdictions, years or case sets).",
    "caveat": "GPU inventory is partly forward-looking, and the default storage is not backed up. UNC's GPU page presents the three RTX Pro 6000 nodes as a 'Coming Soon' FY25/26 acquisition, so confirm availability before relying on them; the pre-existing GPU complement was not itemised in this pass. Also note new resource limits imposed in March 2026: a GPU job cannot request more than 16 CPUs and 128GB of system RAM, to stop GPUs being stranded. The default /hickory/users space is explicitly not backed up.",
    "networks": [],
    "hpcName": "Longleaf — UNC Research Computing (ITS)"
  },
  {
    "state": "North Carolina",
    "school": "Duke University School of Law",
    "university": "Duke University",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "Not itemised publicly for the shared cluster. Duke Research Computing clearly operates GPU hardware — a documented lab-dedicated allocation elsewhere on Duke infrastructure includes 6 servers of four NVIDIA A6000 Ada each and 3 servers of dual V100S — but that is a single lab's exclusive allocation, not the general Duke Compute Cluster complement, and this pass did not establish what GPUs the shared DCC partitions hold.",
    "slurm": "Not confirmed in this pass",
    "hpcName": "Duke Compute Cluster (DCC) — OIT Research Computing",
    "specs": "The DCC is described as a generalised high-performance computing resource supporting research across Duke, accommodating interactive sessions with MATLAB, RStudio and Jupyter notebooks as well as massively parallel workflows. Access routes differ by role: faculty research groups request through Research Computing staff office hours, PhD students claim an allocation through Research Toolkits, and members of existing labs request from their lab point of contact.",
    "sources": [
      "https://oit-rc.pages.oit.duke.edu/rcsupportdocs/",
      "https://cryoem.cs.duke.edu/compute-resources/"
    ],
    "notes": "ADDED Sept 2026. Scored 'undetailed' deliberately rather than guessing at a GPU tier. Duke plainly has substantial GPU capacity, but this pass could not establish what the SHARED Duke Compute Cluster offers as against what individual labs have purchased, and this project has now been bitten three times by exactly that conflation (Missouri, Tulane, Southern Illinois). Recording 'undetailed' with an honest caveat is more useful than an unverified 'modern'. What is confirmed is the access model, and it is good: Duke states that computing and storage allocations for faculty labs and PhD students are provided AT NO COST for most researchers, with buy-in or modest fees only if needs exceed those allocations. Faculty request through Research Computing staff office hours. NEXT STEP FOR A CONTRIBUTOR: the DCC's own partition and GPU documentation would settle the tier in a single visit.",
    "caveat": "Tier not established. The A6000 Ada and V100S figures found in this pass belong to one lab's exclusive allocation (the Bartesaghi Lab, 100 nodes reserved to it), not to the general cluster, and must not be read as available to a law faculty member. Treat this row as 'resource confirmed, hardware unknown' until someone checks the DCC partition list.",
    "networks": []
  },
  {
    "state": "South Carolina",
    "school": "University of South Carolina Joseph F. Rice School of Law",
    "university": "University of South Carolina",
    "standalone": false,
    "tier": "older",
    "tierLabel": "Older-gen GPUs confirmed (V100/T4/P100/K-series/etc.)",
    "gpu": "44 GPU nodes with dual NVIDIA V100 GPUs each (88 V100), added in the 2019 hardware refresh. The original 2017 build had 8 GPU nodes with dual NVIDIA P100s. No Ampere or Hopper hardware was found on Hyperion in this pass.",
    "slurm": "Yes — Slurm Workload Manager",
    "hpcName": "Hyperion — USC Research Computing, Division of Information Technology",
    "specs": "291 standard CPU compute nodes, 44 GPU nodes and 8 big-memory nodes, totalling 16,616 CPU cores. Compute and GPU nodes carry 128–256GB RAM; big-memory nodes 2TB. All nodes on EDR InfiniBand at 100 Gbit with access to 1.4PB of GPFS scratch, plus 100 Gbit Ethernet links to on-premises research storage via Globus data transfer nodes. Total single-precision performance 1,232 TFLOPS. Initially 240 nodes of 28-core Xeon E5-2680 in 2017; the 2019 refresh added 120 nodes of 48-core Xeon Platinum 8260 and the 44 V100 GPU nodes; a May 2023 upgrade added 86 nodes of 64-core Xeon Platinum 8358 and eight 2TB nodes while retiring the first-generation hardware. Housed in USC's DoIT data center with enterprise monitoring, cooling and backup power.",
    "sources": [
      "https://arxiv.org/pdf/2609.11946"
    ],
    "notes": "ADDED Sept 2026. South Carolina previously had only Charleston School of Law (standalone) in this directory, so the state's flagship public law school was missing. DISAMBIGUATION WORTH RECORDING: Hyperion belongs to the University of South CAROLINA, not the University of Southern California — both are 'USC' with a research computing group, and the California one runs Discovery and Endeavour under CARC. Conflating them would be an easy error in a dataset that already contains USC Gould. Access is stated broadly: Hyperion is described as the flagship HPC resource serving FACULTY, STAFF AND STUDENTS ACROSS ALL DOMAINS at USC, operated on a condo model where researchers who buy hardware get subsidised data-centre infrastructure in exchange for donating idle cycles back to the university via FREE, SHARED QUEUES. That donation arrangement is what makes this usable by a law faculty member who owns no hardware. Note the source is a peer-reviewed-style arXiv paper by the cluster's own operators, which is unusually detailed but is a single source — this row should gain a second before it is considered solid.",
    "caveat": "Older-generation GPUs and a single source. Hyperion's GPU complement is V100-era; the 2023 upgrade added CPU and memory nodes but no newer GPUs that this pass could find. The operators' own paper notes the modest GPU complement was intended as a development environment for work that would then scale up on national resources — so for serious GPU work the NSF ACCESS route is the intended path, not Hyperion itself. Sourced from one document; verify against USC's own Research Computing pages.",
    "networks": []
  },
  {
    "state": "Pennsylvania",
    "school": "Penn Carey Law (University of Pennsylvania Carey Law School)",
    "university": "University of Pennsylvania",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Betty (PARCC): NVIDIA DGX B200 SuperPOD — 31 nodes x 8x NVIDIA Blackwell B200 (180GB HBM3e each) = 248 B200 GPUs, NVLink/NVSwitch within each node plus a dedicated NDR400 InfiniBand compute fabric so a single job can scale across the entire SuperPOD; 8.5 PFLOPS cluster-wide",
    "slurm": "Yes",
    "hpcName": "Betty — Penn Advanced Research Computing Center (PARCC)",
    "specs": "GPU side: 31 DGX B200 nodes, each 2x Intel Xeon Platinum 8570 (112 cores) + 2TB DDR5 + 30.72TB local NVMe (3,472 cores total = 31x112; PARCC's own table prints 3,474). CPU side: 64 Dell nodes, AMD EPYC Genoa 9374F, 4,736 cores, 35TB RAM, NDR200 InfiniBand. VAST + Ceph storage. In production May 2025; Penn reports it ranked #189 on the Top500 and in the top ten among academic systems.",
    "sources": [
      "https://parcc.upenn.edu/systems/betty/",
      "https://penntoday.upenn.edu/news/penns-newest-supercomputer-transforming-research",
      "https://research-it.wharton.upenn.edu/news/parcc-betty-cluster-goes-live/",
      "https://www.ahead.com/penn-advanced-research-computing-center-parcc/",
      "https://parcc.upenn.edu/training/faculty-info/coldfront/",
      "https://parcc.upenn.edu/training/applications/parabricks/",
      "https://almanac.upenn.edu/articles/launch-of-the-penn-advanced-research-computing-center"
    ],
    "caveat": "Not free at the point of use: Betty bills per use — $1 per GPU-hour and $0.01 per CPU-hour, with no entry fee (Penn Today, 2026) — so a law faculty member needs a research budget or school subsidy. Wharton publicly subsidizes allocations for its own faculty; no equivalent Penn Carey Law subsidy was found in public sources. Rates can change — check PARCC's Rates page.",
    "notes": "ADDED Oct 2026. Built as a central resource for all twelve Penn schools, explicitly including Penn Carey Law — and law researchers are already using it: Penn Today reports PARCC helped a Penn Carey Law team facing an end-of-week deadline cut a month-long experiment to eight hours. Faculty with a PennKey open a project and request allocations through PARCC's ColdFront portal (campus network or GlobalProtect VPN required); SLURM confirmed via PARCC's own job examples (srun on B200).",
    "networks": []
  },
  {
    "state": "Pennsylvania",
    "school": "University of Pittsburgh School of Law",
    "university": "University of Pittsburgh",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "CRCD GPU cluster, by partition: h200 — 8x NVIDIA H200 SXM (141GB) per node, NVLink; rtx6k (newest) — 8x NVIDIA RTX PRO 6000 Blackwell (96GB) per node; l40s — 19 nodes x 4x L40S (48GB); a100 — 12 nodes x 4x A100 PCIe 40GB; a100_multi — 10 nodes x 4x A100 PCIe 40GB; a100_nvlink — 2 nodes x 8x A100 SXM4 80GB + 3 nodes x 8x A100 SXM4 40GB. Teaching cluster adds legacy GTX TITAN X / GTX 1080 nodes.",
    "slurm": "Yes",
    "hpcName": "Center for Research Computing and Data (CRCD) — GPU cluster",
    "specs": "Separate smp, mpi, gpu, htc, and teach clusters, plus a Secure Research Environment (Azure Virtual Desktop) for IRB/DUA data needing NIST SP 800-171. HDR200 InfiniBand on A100 nodes. The Pittsburgh Supercomputing Center (joint CMU/Pitt) is a separate, NSF-allocated national resource, not counted here.",
    "sources": [
      "https://crc-pages.pitt.edu/user-manual/hardware_profiles/gpu/",
      "https://crc.pitt.edu/overview-crc-services/computing-hardware",
      "https://crc-pages.pitt.edu/user-manual/slurm/batch-jobs/",
      "https://crc-pages.pitt.edu/user-manual/getting-started/getting-started-step1-account/",
      "https://crc.pitt.edu/requesting-new-account",
      "https://crc-pages.pitt.edu/user-manual/slurm/service-units/"
    ],
    "notes": "ADDED Oct 2026. Free for faculty: every Pitt faculty member or PI can open an account at no cost, with 50,000 service units on each cluster (the GPU cluster included) plus 5TB of project storage for one year, renewable by short proposal. Preemptible partitions bill zero service units. Students and staff join through a faculty PI. PittNet VPN required.",
    "networks": []
  },
  {
    "state": "Pennsylvania",
    "school": "Temple University Beasley School of Law",
    "university": "Temple University",
    "standalone": false,
    "tier": "older",
    "tierLabel": "Older-gen GPUs confirmed (V100/T4/P100/K-series/etc.)",
    "gpu": "Machine Learning servers: an NVIDIA DGX-1 with 8x Tesla V100 (NVLink2) plus a GPU-compute server with 4x Tesla V100 (NVLink2); Owl's Nest cluster: 2x NVIDIA P100 on each of its 512GB large-memory nodes",
    "slurm": "Not confirmed in sources checked",
    "hpcName": "Owl's Nest + Compute + Machine Learning (gpu.hpc / dgx-1.hpc) — run by ICMS / College of Science and Technology",
    "specs": "Owl's Nest: 6,464 CPU cores (180 nodes x 28 cores/128GB from 2017 + 48 nodes x 16 cores/96GB from 2018), large-memory nodes up to 3TB, EDR InfiniBand, 1.5PB parallel storage, ~57M core-hours/year. Compute: interactive servers up to 88 cores/1.5TB with MATLAB, SAS, Stata, R/RStudio.",
    "sources": [
      "https://it.cst.temple.edu/server/hpc/",
      "https://icmshpc.com/hpc/",
      "https://cphwiki.temple.edu/books/research-tools/page/high-performance-computing",
      "https://now.temple.edu/news/2017-11-28/new-computer-cluster-pushes-temple-ahead-high-performance-computing",
      "https://cis.temple.edu/about/computing_facilities/",
      "https://backend.mhpc.sissa.it/sites/default/files/2023-09/Nguruwe-tesi.pdf"
    ],
    "caveat": "The hardware described in every public source dates from 2017–2018, and a second search found no newer GPU purchase. Temple may have newer hardware that just isn't documented; ask the ICMS HPC team before ruling it out.",
    "notes": "ADDED Oct 2026. University-wide eligibility: faculty, staff, and graduate or professional students with a Temple AccessNet account who have significant computational needs (Temple's own 2017 announcement says any researcher on campus can apply). The interactive Compute servers already carry Stata, SAS, R, and MATLAB, which suits empirical legal work.",
    "networks": []
  },
  {
    "state": "Pennsylvania",
    "school": "Drexel University Thomas R. Kline School of Law",
    "university": "Drexel University",
    "standalone": false,
    "tier": "older",
    "tierLabel": "Older-gen GPUs confirmed (V100/T4/P100/K-series/etc.)",
    "gpu": "Picotte: 12 GPU nodes (Dell PowerEdge C4140) x 4x NVIDIA Tesla V100-SXM2 32GB with NVLink = 48 V100s",
    "slurm": "Yes",
    "hpcName": "Picotte — University Research Computing Facility (URCF)",
    "specs": "4,224 CPU cores: 74 standard nodes (48 cores/192GB), 2 large-memory nodes (1.5TB), 12 GPU nodes; ~20TB RAM; 500TB storage; RHEL 8; benchmarked by Dell at 251 TFLOPS across the GPU nodes (355 TFLOPS theoretical peak overall). Funded by a $1.2M Drexel + NSF Major Research Instrumentation award.",
    "sources": [
      "https://docs.urcf.drexel.edu/clusters/picotte/hardware/",
      "https://drexel.edu/core-facilities/facilities/research-computing",
      "https://drexel.edu/core-facilities/facilities/research-computing/picotte",
      "https://docs.urcf.drexel.edu/overview/",
      "https://docs.urcf.drexel.edu/clusters/"
    ],
    "notes": "ADDED Oct 2026. Free for all Drexel researchers, with higher usage limits and priority jobs available for a fee; Drexel VPN required. Picotte is Drexel's only supported research cluster. URCF also serves other Philadelphia-area universities.",
    "networks": []
  },
  {
    "state": "Pennsylvania",
    "school": "Villanova University Charles Widger School of Law",
    "university": "Villanova University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Augie: one GPU node with NVIDIA A100 GPUs (GPU count per node not published)",
    "slurm": "Not confirmed in sources checked",
    "hpcName": "Augie — Centralized Research Computing Facility (CRCF)",
    "specs": "AMD EPYC CPU nodes, HDR InfiniBand, CentOS 7; condo model (members can add hardware for priority time); connected to the Open Science Grid. CRCF also runs CLUSTY for astrophysics, and the College of Engineering runs a separate, CPU-only cluster (208 cores) for its own researchers.",
    "sources": [
      "https://www1.villanova.edu/villanova/provost/research/hpccluster/augie.html",
      "https://www1.villanova.edu/villanova/provost/research/hpccluster.html",
      "https://villanova.teamdynamix.com/TDClient/2759/UNIT/Requests/ServiceDet?ID=51428"
    ],
    "caveat": "The GPU capacity is small: Villanova describes a single A100 node. It qualifies as modern hardware, but expect contention and modest scale, not the multi-node GPU pools at the larger schools here.",
    "notes": "ADDED Oct 2026. Funded by an NSF Campus Cyberinfrastructure (CC*) grant and described by Villanova as available to all Villanova members; access is requested through the UNIT service catalog.",
    "networks": []
  },
  {
    "state": "Pennsylvania",
    "school": "Duquesne University Thomas R. Kline School of Law",
    "university": "Duquesne University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://duq.edu/",
      "https://en.wikipedia.org/wiki/Pittsburgh_Quantum_Institute"
    ],
    "caveat": "Duquesne is an R2 research university, so a departmental cluster (for example in chemistry or pharmacy) may exist without a public page. Three separate searches found none — ask Duquesne Computing and Technology Services before ruling it out.",
    "notes": "ADDED Oct 2026. Searched Sept 2026 three ways (general HPC/GPU terms, NSF instrumentation grants, and Duquesne's own domain): no shared research-computing service or cluster page found. Duquesne faculty do belong to cross-institution groups like the Pittsburgh Quantum Institute, but that is a research network, not a compute resource. Not to be confused with Drexel's law school, which is also named for Thomas R. Kline. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Pennsylvania",
    "school": "Widener University Commonwealth Law School (Harrisburg)",
    "university": "Widener University",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "No GPU model published — Widener's CS department page mentions only \"a high-performance computing cluster\"",
    "slurm": "Not confirmed",
    "hpcName": "Computer Science department cluster (Freedom Hall, Chester campus)",
    "specs": "Housed alongside three student-managed computer labs (66 Linux/Windows workstations) and a 12-system computer forensics lab",
    "sources": [
      "https://www.widener.edu/academics/undergraduate-programs/major-computer-science-computer-information-systems",
      "https://commonwealthlaw.widener.edu/files/resources/hb_student_handout_fa2017.pdf",
      "https://en.wikipedia.org/wiki/Widener_University_Commonwealth_Law_School"
    ],
    "caveat": "This is an undergraduate computer-science department cluster on Widener's main campus in Chester, about 90 miles from the Harrisburg law school — not a university-wide research service. Law-faculty access is unlikely without a CS collaborator.",
    "notes": "ADDED Oct 2026. Searched Sept 2026: no central research-computing service found at Widener; Widener Commonwealth's own Harrisburg IT guide covers only wifi, computer labs, and helpdesk. This departmental cluster is the only HPC resource mentioned in public sources.",
    "networks": []
  },
  {
    "state": "Delaware",
    "school": "Widener University Delaware Law School",
    "university": "Widener University",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "No GPU model published — Widener's CS department page mentions only \"a high-performance computing cluster\"",
    "slurm": "Not confirmed",
    "hpcName": "Computer Science department cluster (Freedom Hall, Chester campus)",
    "specs": "Housed alongside three student-managed computer labs (66 Linux/Windows workstations) and a 12-system computer forensics lab",
    "sources": [
      "https://www.widener.edu/academics/undergraduate-programs/major-computer-science-computer-information-systems",
      "https://en.wikipedia.org/wiki/Widener_University_Delaware_Law_School",
      "https://en.wikipedia.org/wiki/Widener_University"
    ],
    "caveat": "Same situation as Widener's Harrisburg law school: the only cluster found is an undergraduate computer-science department cluster on the Chester, PA campus (roughly 15 miles from the Wilmington law campus), not a university-wide research service.",
    "notes": "ADDED Oct 2026. Delaware's only ABA-accredited law school. It has operated separately from Widener Commonwealth since 2015, with its own dean and faculty, but both remain part of Widener University, so the same university-level search applies.",
    "networks": []
  },
  {
    "state": "Delaware",
    "school": "Wilmington University Farnan School of Law",
    "university": "Wilmington University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "ADDED Oct 2026. Missing from the directory until now because it is new: the first class enrolled in fall 2023, the ABA granted provisional approval in February 2025, and the school was renamed the Farnan School of Law on May 1, 2026 (the ABA's lists still say \"Wilmington University School of Law\"). Not to be confused with Widener's Delaware Law School, which is also in Wilmington. Wilmington University is a teaching-focused institution with no research-computing service; its AI Hub covers AI policy and guidance, not hardware. The nearest shared resource doesn't reach it today. Wilmington University is a named partner in the University of Delaware-led NSF E-CORE RII award (2025–2029), which aims to strengthen Delaware's data infrastructure but names no computing system. UD's DARWIN cluster, which had offered free use to outside academics, was decommissioned on October 1, 2025, according to UD's own HPC documentation. UD's remaining cluster, Caviness, is a condo system built from nodes bought by research groups. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://blog.wilmu.edu/news/2026/05/04/wilmington-university-names-farnan-school-of-law/",
      "https://law.wilmu.edu/about/aba-approval-process.aspx",
      "https://www.wilmu.edu/ai/index.aspx",
      "https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2438144",
      "https://docs.hpc.udel.edu/start",
      "https://docs.hpc.udel.edu/abstract/darwin/darwin"
    ]
  },
  {
    "state": "North Dakota",
    "school": "University of North Dakota School of Law",
    "university": "University of North Dakota",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Medora (2026): 16 nodes, each with an NVIDIA Grace Hopper superchip (72-core Grace CPU plus Hopper GPU). Talon: 2 GPU nodes x 8x NVIDIA Tesla V100 32GB.",
    "slurm": "Not confirmed in sources checked",
    "hpcName": "UND Computational Research Center (CRC): Medora + Talon",
    "specs": "Talon: 15 compute nodes (36 cores, 192GB), 3 large-memory nodes (3TB), HPE Gen10; contributes spare cycles to the Open Science Pool. Medora: 16 Grace Hopper nodes reachable by SSH or Open OnDemand.",
    "sources": [
      "https://und.edu/research/crc/index.html",
      "https://und.edu/research/crc/clusters-equipment/index.html",
      "https://blogs.und.edu/crc-news/2026/06/explore-ai-and-high-performance-computing-with-the-new-medora-cluster/",
      "https://und.edu/research/crc/clusters-equipment/talon.html"
    ],
    "notes": "ADDED Oct 2026. UND's Computational Research Center says its HPC systems are available at no cost to the UND community, and it also pays licensing costs for software used by students, faculty, staff and researchers. Medora is the newest cluster, introduced with a June 2026 hands-on workshop; a permanent account is a separate request. The backlog lead about a shared UND/NDSU state route turned out not to matter, because UND's own hardware is frontier-class.",
    "networks": []
  },
  {
    "state": "Rhode Island",
    "school": "Roger Williams University School of Law",
    "university": "Roger Williams University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://www.rwu.edu/who-we-are/administrative-offices/information-technology/faculty-services",
      "https://www.rwu.edu/who-we-are/administrative-offices/information-technology/staff-services",
      "https://its.uri.edu/research-computing/"
    ],
    "notes": "ADDED Oct 2026. Roger Williams IT lists email, wireless, computer and software support, and classroom support for faculty, with no research-computing service. Rhode Island's shared HPC (Unity and SeaWulf) is run by the University of Rhode Island for URI researchers, so it doesn't extend to Roger Williams. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "South Dakota",
    "school": "University of South Dakota Knudson School of Law",
    "university": "University of South Dakota",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Through SDSU's Innovator cluster, which SDSU states is available to all South Dakota Board of Regents institutions including USD: 14 GPU nodes x 2x NVIDIA A100 80GB. USD's own Lawrence cluster has GPU-accelerated nodes of unpublished model. SDSU's newer Discovery cluster adds NVIDIA H100 80GB, but USD access to Discovery isn't stated.",
    "slurm": "Yes (Innovator GPU jobs use --gres in SDSU's docs)",
    "hpcName": "SDSU Innovator (statewide, Board of Regents) + USD Lawrence",
    "specs": "Innovator: 46 compute nodes (48 cores, 256GB), 4 big-memory nodes (2TB), 14 GPU nodes; 3,072 cores; 3PB storage; 100Gbps InfiniBand. USD Lawrence: 60+ TFLOPS at launch, funded by NSF ($504K) and the Board of Regents ($200K); USD also runs the older Legacy cluster.",
    "sources": [
      "https://help.sdstate.edu/TDClient/2744/Portal/KB/Article/164534/SDSU-HPC-Q-and-A",
      "https://www.sdstate.edu/information-technology/research-cyberinfrastructure",
      "https://www.hpcwire.com/off-the-wire/university-south-dakota-gets-hpc-cluster-research-computing/",
      "https://spectralogic.com/case-studies/university-of-south-dakota/"
    ],
    "notes": "ADDED Oct 2026. USD's own Lawrence cluster entered production in 2018 as a campus-wide resource for faculty, staff and students. The stronger route is statewide: SDSU's Research Cyberinfrastructure team supports HPC across the South Dakota Board of Regents, and Innovator's documentation names USD explicitly.",
    "caveat": "Rated on Innovator's A100s, not Discovery's H100s. SDSU's help center says Innovator is available to all Board of Regents institutions, which covers USD, but it doesn't say the same for Discovery. Ask SDSU Research Cyberinfrastructure whether USD researchers can get Discovery time.",
    "networks": [
      "South Dakota Board of Regents (SDSU Innovator)"
    ]
  },
  {
    "state": "West Virginia",
    "school": "West Virginia University College of Law",
    "university": "West Virginia University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Dolly Sods (2023, NSF MRI #2117575): 155 GPUs — 120 NVIDIA A30, 19 A40 and 16 A100 (two nodes of 8x A100 40GB). Thorny Flat: 21 Quadro P6000, 24 RTX 6000 and 2 A100.",
    "slurm": "Yes",
    "hpcName": "WVU Research Computing: Dolly Sods + Thorny Flat",
    "specs": "Dolly Sods: 37 nodes, about 1,250 AMD EPYC 7513 cores, HDR100 InfiniBand. Thorny Flat: about 180 nodes and 6,500+ cores, 30TB RAM, Omni-Path; 79 community nodes plus 103 nodes bought by faculty and departments; hosted at the Pittsburgh Supercomputing Center.",
    "sources": [
      "https://research.wvu.edu/tools/research-computing/high-performance-computing",
      "https://docs.hpc.wvu.edu/",
      "https://docs.hpc.wvu.edu/text/84.DollySods.html",
      "https://docs.hpc.wvu.edu/text/83.ThornyFlat.html"
    ],
    "notes": "ADDED Oct 2026. WVU Research Computing runs community nodes for the university's researchers, with condo purchases for priority access (an A100 80GB card for Dolly Sods is quoted at about $14,000). WVU's pages give slightly different totals for Thorny Flat (6,196 to 6,796 cores), so treat node and core counts as approximate.",
    "networks": []
  },
  {
    "state": "Wyoming",
    "school": "University of Wyoming College of Law",
    "university": "University of Wyoming",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "MedicineBow: 48x NVIDIA H100 SXM 80GB (8 per node), 48x L40S 48GB and 64x A30 24GB",
    "slurm": "Yes",
    "hpcName": "MedicineBow — Advanced Research Computing Center (ARCC)",
    "specs": "300+ nodes including 4TB-RAM mass-memory nodes; AMD EPYC 9454 hosts; 200Gbps InfiniBand; 6.3 PFLOPS peak at deployment (Xi Computers).",
    "sources": [
      "https://arccwiki.atlassian.net/wiki/spaces/DOCUMENTAT/pages/2103312408/Facilities+Statement",
      "https://www.uwyo.edu/arcc/policies/costpolicy.html",
      "https://www.jonpeddie.com/news/a-supercomputer-in-wyoming/"
    ],
    "notes": "ADDED Oct 2026. ARCC's cost policy lists GPU use of up to 20,000 GPU-hours a year in total, all GPU types combined, with per-GPU-hour charges only above that quota (about $0.13 for A30 and $0.18 for L40S). The NCAR-Wyoming Supercomputing Center (Derecho) is a separate backlog lead that wasn't needed to rate this row.",
    "networks": []
  },
  {
    "state": "Utah",
    "school": "S.J. Quinney College of Law",
    "university": "University of Utah",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "CHPC GPUs by cluster: granite — NVIDIA H200 (141GB) and H100 NVL; notchpeak — H100 NVL, L40S, L40, A100, A40, V100, RTX 3090 and 2080 Ti (general plus guest-accessible owner nodes); redwood (Protected Environment) — H200 and H100 NVL",
    "slurm": "Yes",
    "hpcName": "Center for High Performance Computing (CHPC): granite, notchpeak, kingspeak, lonepeak, redwood",
    "specs": "General Environment (granite, notchpeak, kingspeak, lonepeak, Open OnDemand, Windows server) and a Protected Environment for PHI/PII work (redwood). Owner nodes on every cluster accept preemptible guest jobs from all users.",
    "sources": [
      "https://www.chpc.utah.edu/documentation/guides/gpus-accelerators.php",
      "https://www.chpc.utah.edu/userservices/allocations.php",
      "https://www.chpc.utah.edu/documentation/guides/index.php",
      "https://www.chpc.utah.edu/documentation/software/slurm-gpus.php"
    ],
    "notes": "ADDED Oct 2026. Unusually open for GPUs: CHPC says all users have free access to the general GPU nodes on notchpeak, kingspeak, lonepeak and redwood (you ask to be added to the GPU account), and the notchpeak-shared-short partition is open to anyone with a CHPC account. Granite's GPUs need an allocation, which CHPC says has no cost but is merit-reviewed (up to 3,000 GPU-hours a quarter); without one, jobs run in preemptible freecycle mode. CHPC also offers One-U Responsible AI GPUs to university researchers with AI projects on request.",
    "networks": []
  },
  {
    "state": "Utah",
    "school": "J. Reuben Clark Law School",
    "university": "Brigham Young University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Generally available: 32 NVIDIA H200 (4 nodes x 8), 16 L40S (4 nodes x 4), 2 GH200 Grace Hopper nodes and 160 P100 (40 nodes x 4). Privately owned but open to any job that opts in to preemption: 8 B200, 8 H200, 16 H100 and 120 A100.",
    "slurm": "Yes",
    "hpcName": "BYU Office of Research Computing (formerly the Fulton Supercomputing Lab)",
    "specs": "About 620 nodes, 35,000+ cores, 367 GPUs and 180TB of memory (February 2026 letter of support); a 136-node AMD EPYC cluster on 100Gb/s InfiniBand; 6+ PB of shared storage; support for protected data.",
    "sources": [
      "https://rc.byu.edu/proposals",
      "https://rc.byu.edu",
      "https://rc.byu.edu/wiki/?id=Getting+Started+With+GPUs",
      "https://rc.byu.edu/wiki/?id=Grace+Hopper"
    ],
    "notes": "ADDED Oct 2026. Free: BYU's Office of Research Computing states that its resources are made available for free to researchers, for faculty and student researchers and their collaborators; accounts need a faculty sponsor and a short statement of need. One of the strongest freely available GPU pools in this directory.",
    "networks": []
  },
  {
    "state": "Wisconsin",
    "school": "University of Wisconsin Law School",
    "university": "University of Wisconsin-Madison",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "CHTC GPU Lab, open to any CHTC user: 16 NVIDIA H200 (2 servers x 8), 8 H100 80GB, 32 A100 80GB, 8 A100 40GB, 30 L40, 16 L40S, 8 RTX 2080 Ti and 4 P100. The Data Science Institute separately runs A100, L40 and H100 clusters for its affiliates.",
    "slurm": "HTCondor for the GPU Lab and HTC system; CHTC's separate HPC cluster uses SLURM",
    "hpcName": "Center for High Throughput Computing (CHTC) — GPU Lab",
    "specs": "About 30,000 HTC CPU cores plus a SLURM HPC cluster with 200Gb/s InfiniBand; automatic overflow to campus pools and the OSG's Open Science Pool; dedicated interactive A100 slots for testing.",
    "sources": [
      "https://chtc.cs.wisc.edu/uw-research-computing/gpu-jobs",
      "https://it.wisc.edu/services/center-for-high-throughput-computing",
      "https://chtc.wisc.edu/uw-research-computing/gpu-lab.html",
      "https://chtc.wisc.edu/uw-research-computing/cite-chtc.html",
      "https://research.wisc.edu/research-computing-support-web-page-for-rise"
    ],
    "notes": "ADDED Oct 2026. Free: UW–Madison IT lists CHTC access as free and available to all UW-affiliated students, staff and faculty (students and external collaborators need a sponsor). The H100/H200 rating comes from CHTC's own GPU jobs guide, not secondary counts. Jobs run through HTCondor rather than SLURM, which suits many independent runs; CHTC's facilitators offer consultations for new users.",
    "networks": []
  },
  {
    "state": "Wisconsin",
    "school": "Marquette University Law School",
    "university": "Marquette University",
    "standalone": false,
    "tier": "older",
    "tierLabel": "Older-gen GPUs confirmed (V100/T4/P100/K-series/etc.)",
    "gpu": "Raj: 48 NVIDIA Tesla V100 — 12 GPU nodes x 2, plus three AI/ML nodes x 8 with NVLink",
    "slurm": "Yes",
    "hpcName": "Raj — Marquette ITS Research Computing Services",
    "specs": "7,808 AMD EPYC Rome cores across 38 general, large-memory and massive-memory (2TB) nodes; 1.5PB storage; 100Gb/s InfiniBand; online since April 2021 (NSF MRI CNS-1828649).",
    "sources": [
      "https://www.marquette.edu/high-performance-computing",
      "https://www.marquette.edu/high-performance-computing/system-overview.php",
      "https://www.marquette.edu/high-performance-computing/architecture.php",
      "https://www.marquette.edu/news-center/2021/marquette-s-high-performance-computing-cluster-to-officially-open-june-2.php"
    ],
    "notes": "ADDED Oct 2026. Free: Marquette says faculty and students are eligible to use its HPC infrastructure for courses or research without cost. The Electrical and Computer Engineering department also runs its own ten-node A5000 cluster for its faculty and students, but that is not a campus-wide service.",
    "networks": []
  },
  {
    "state": "Oregon",
    "school": "University of Oregon School of Law",
    "university": "University of Oregon",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Talapas general-use pool: 24 GPU nodes with 52 NVIDIA A100. Researcher-funded condo nodes, which RACS says are shared with all cluster users: 20 H100, 14 A100, 16 V100, 4 L40S, 4 A40 and 2 L40 (89 GPUs in total).",
    "slurm": "Not confirmed in sources checked",
    "hpcName": "Talapas — Research Advanced Computing Services (RACS)",
    "specs": "14,520+ CPU cores, 129TB memory, 3+ PB storage, EDR InfiniBand; general-use nodes include 43 AMD Milan compute nodes and large-memory nodes up to 4TB; Open OnDemand web access; 250GB home and 2TB project space.",
    "sources": [
      "https://racs.uoregon.edu/services",
      "https://partnerships.uoregon.edu/utilize-uo-resources/core-facilities/racs",
      "https://www.hpcwire.com/off-the-wire/university-oregon-uses-supercomputer-power-new-research-advanced-computing-facility/"
    ],
    "notes": "ADDED Oct 2026. Faculty register as principal investigators to get access, and members of registered research teams join under them. Talapas is described by UO as one of the fastest academic supercomputers in the Northwest.",
    "caveat": "Rated on the general-use A100s. Talapas's H100s are condo hardware bought by research groups; RACS says condo resources are shared with all cluster users, but access to them is typically preemptible, so a law researcher shouldn't count on H100 time.",
    "networks": []
  },
  {
    "state": "Oregon",
    "school": "Lewis & Clark Law School",
    "university": "Lewis & Clark College",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "BLT cluster: one GPU node (\"Sprouts\", 44 cores, 500GB RAM); GPU model not published",
    "slurm": "Not confirmed",
    "hpcName": "BLT — Watzek Library Digital and Data Services",
    "specs": "144-core Intel Xeon cluster: login node with 140TB disk, three 44-core/500GB compute nodes, the GPU node, and an 80TB storage node; also RStudio Server and JupyterHub.",
    "sources": [
      "https://library.lclark.edu/digital",
      "https://library.lclark.edu/digital/hpc",
      "https://college.lclark.edu/programs/data-science/high-performance-computing/",
      "https://www.lclark.edu/_ingredients/templates/details/newsroom.php?id=40490"
    ],
    "notes": "ADDED Oct 2026. Small but real, and unusual for a liberal arts college: the library's Digital and Data Services says the cluster is available to faculty and staff at Lewis & Clark, which includes the law school. Expect modest capacity — a single GPU node.",
    "networks": []
  },
  {
    "state": "Oregon",
    "school": "Willamette University College of Law",
    "university": "Willamette University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://willamette.edu/academics/computing/",
      "https://willamette.edu/academics/college/computer-science/"
    ],
    "notes": "ADDED Oct 2026. Willamette's School of Computing and Information Sciences offers teaching programs in Salem and Portland, but no research-computing service or cluster was found on its IT or academic pages. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Washington",
    "school": "University of Washington School of Law",
    "university": "University of Washington",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Tillicum (launched fall 2025): 192 NVIDIA H200 (141GB) across 24 Dell XE9680 servers, NVLink 4.0, 400Gb/s InfiniBand. Hyak Klone: shared condo cluster with CPU and GPU nodes (UW's CSE department reports 800+ GPUs on Hyak).",
    "slurm": "Yes (SLURM on Klone; Tillicum jobs are scheduled per GPU)",
    "hpcName": "UW-IT Research Computing: Tillicum (GPU) + Hyak Klone (condo HPC)",
    "specs": "Both systems are reachable through Open OnDemand (Jupyter, VS Code, RStudio); Kopah S3 storage and the Lolo archive alongside. Tillicum appears on the Top500 and Green500 lists, per UW's architecture page.",
    "sources": [
      "https://hyak.uw.edu/docs/tillicum/architecture",
      "https://uwconnect.uw.edu/it?id=kb_article_view&sysparm_article=KB0036077",
      "https://hyak.uw.edu/docs/",
      "https://uwconnect.uw.edu/it?id=kb_article_view&sysparm_article=KB0035549",
      "https://uw.service-now.com/catform.do?form=up5z3r"
    ],
    "notes": "ADDED Oct 2026. UW-IT offers one-on-one research computing consultations to match projects to Klone, Tillicum or cloud options. Students paying the Student Technology Fee can also apply for $250 Tillicum credit grants through the Research Computing Club.",
    "caveat": "Tillicum is pay-as-you-go: $0.90 per GPU-hour, billed monthly to a UW budget, with no slot fees and exempt from F&A. Every faculty researcher gets 100 free promotional GPU-hours (once per PI) to evaluate it. Hyak Klone runs on a condo model where groups buy slices; a free community tier exists through demonstration accounts.",
    "networks": []
  },
  {
    "state": "Washington",
    "school": "Seattle University School of Law",
    "university": "Seattle University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://www.seattleu.edu/its/support/service-desk/",
      "https://law.seattleu.edu/student-life/student-services/student-support/technology-and-media-services/",
      "https://www.seattleu.edu/science-engineering/academic-departments/department-of-computer-science/"
    ],
    "notes": "ADDED Oct 2026. Seattle University's IT service catalog covers accounts, Microsoft 365, Zoom, computer labs and administrative systems, and the law school's own IT staff run a help desk and media services; no research-computing service or cluster was found in either. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Washington",
    "school": "Gonzaga University School of Law",
    "university": "Gonzaga University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://support.gonzaga.edu/TDClient/30/Portal/Requests/ServiceCatalog",
      "https://support.gonzaga.edu/TDClient/30/Portal/Requests/ServicesAtoZ",
      "https://www.gonzaga.edu/school-of-engineering-applied-science/degrees-and-programs/computer-science/cs-labs"
    ],
    "notes": "ADDED Oct 2026. Gonzaga's IT service catalog (136 services) has no research-computing entry. Computer Science runs instructional labs, multi-user Linux servers for its students, and an Advanced Computing Lab for cybersecurity and networking work, none of which is a shared HPC service. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Puerto Rico",
    "school": "University of Puerto Rico School of Law",
    "university": "University of Puerto Rico",
    "standalone": false,
    "tier": "undetailed",
    "tierLabel": "HPC exists, GPU support not publicly detailed",
    "gpu": "No GPU hardware mentioned in Boqueron's documentation, which describes only CPU cores and storage",
    "slurm": "Yes",
    "hpcName": "Boqueron — UPR High Performance Computing facility (HPCf)",
    "specs": "2,240+ compute cores, 200TB high-performance storage, QDR InfiniBand and 10G Ethernet; /home and per-group /work filesystems; the HPCf also runs PRISAnet, UPR's research network.",
    "sources": [
      "https://www.hpcf.upr.edu/",
      "https://www.hpcf.upr.edu/documentation/boqueron/",
      "https://www.hpcf.upr.edu/about/",
      "https://natsci.uprrp.edu/biology/high-performance-computing-facility/",
      "https://www.hpcf.upr.edu/account-request/"
    ],
    "notes": "ADDED Oct 2026. The HPCf describes its mission as serving the research and education community of the University of Puerto Rico, and accounts for Boqueron are requested through its site. Its acknowledgment text credits UPR and an NIH INBRE grant, a sign the cluster leans toward bioinformatics. Whether any GPU nodes exist isn't stated, so this stays undetailed rather than cpuonly. UPDATED Oct 2026: Boqueron accounts must be tied to an existing HPCf group account, which the research group's PI requests through the HPCf site, so a law faculty member would apply as PI for a new group. The HPCf's newest posted notice dates from Aug 2024, and no newer UPR GPU or HPC system was found.",
    "networks": []
  },
  {
    "state": "Puerto Rico",
    "school": "Inter American University of Puerto Rico School of Law",
    "university": "Inter American University of Puerto Rico",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://en.wikipedia.org/wiki/Interamerican_University_of_Puerto_Rico",
      "https://texascareercheck.com/SchoolInfo/SchoolSummary/6816/",
      "https://www.hpcf.upr.edu/about/"
    ],
    "notes": "ADDED Oct 2026. Inter American is a private multi-campus university whose public profiles describe staffed computer labs but no research-computing service. The backlog asked about a Puerto Rico statewide route: the island's only HPC facility found, UPR's HPCf, describes itself as serving the University of Puerto Rico community, so it doesn't extend to Inter American. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Puerto Rico",
    "school": "Pontifical Catholic University of Puerto Rico School of Law",
    "university": "Pontifical Catholic University of Puerto Rico",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://www.msche.org/institution/0603/",
      "https://www.pucpr.edu/wp-content/uploads/2021/08/Undergraduate-Catalog-2019-2022.pdf",
      "https://en.wikipedia.org/wiki/Pontifical_Catholic_University_of_Puerto_Rico"
    ],
    "notes": "ADDED Oct 2026. The university's accreditor lists its Carnegie classification as having no research designation. Its catalog describes an Academic Computer Center and a computer lab inside the College of Science research building, neither of which is a shared HPC resource; UPR's HPCf serves only the UPR community. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Tennessee",
    "school": "Vanderbilt University Law School",
    "university": "Vanderbilt University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "ACCRE batch_gpu partition: NVIDIA H100 NVL (94GB), L40S (48GB) and A100 80GB; RTX A4000 and A6000 in interactive and group partitions. The Data Science Institute adds four DGX A100 systems integrated into ACCRE, for DSI projects and DSI compute-grant recipients.",
    "slurm": "Yes",
    "hpcName": "Advanced Computing Center for Research and Education (ACCRE)",
    "specs": "737 compute nodes, about 16,000 cores, 4.8PB shared cluster storage plus 22PB LStore; Rocky 9 environment with Open OnDemand (JupyterLab, GPU desktop). Serves about 3,100 researchers in 50 departments and centers (January 2024).",
    "sources": [
      "https://help.accre.vanderbilt.edu/index.php?title=GPUs_at_ACCRE",
      "https://help.accre.vanderbilt.edu/index.php?title=GPU_IN_R9",
      "https://vanderbilt.ilab.agilent.com/service_center/show_external/4487?name=vanderbilt-advance-computing-center-for-research-education-accre-core",
      "https://github.com/vanderbilt-data-science/dgx-user-guide",
      "https://help.accre.vanderbilt.edu/index.php?title=Technical_Details"
    ],
    "notes": "ADDED Oct 2026. The DGX A100 route is the closest thing to free GPU time: DSI awards compute grants for it. Otherwise budget for card rental.",
    "caveat": "GPU access is rented, not free: ACCRE's subsidized rate for Vanderbilt researchers is $186 a month per Level 1 card and $372 per Level 2 card, with a three-month minimum. ACCRE's January 2024 figures say it serves researchers from five schools; whether Law is among them isn't stated, though ACCRE describes itself as built for Vanderbilt faculty generally.",
    "networks": []
  },
  {
    "state": "Tennessee",
    "school": "University of Tennessee College of Law",
    "university": "University of Tennessee Knoxville",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "ISAAC-NG: 10 nodes x 4x NVIDIA H100 80GB (university and private condo), V100S, T4, A40, A16 and L40 nodes, plus the AI Tennessee Initiative's 7 Dell XE8640 servers x 4x H100 SXM with NVLink (28 GPUs), available to UTK faculty on request",
    "slurm": "Yes",
    "hpcName": "ISAAC Next Generation (ISAAC-NG) — OIT High Performance & Scientific Computing",
    "specs": "About 210 nodes and 12,000+ cores, 3.6PB Lustre; campus, campus-gpu, campus-gpu-large and campus-gpu-bigmem partitions for all users (campus-gpu QoS: up to 2 GPUs per user, 1-day jobs); Secure Enclave for sensitive data; Open OnDemand with Jupyter and RStudio.",
    "sources": [
      "https://oit.utk.edu/hpsc/isaac-open-enclave-new-kpb/system-overview-cluster-at-kpb/",
      "https://oit.utk.edu/hpsc/ai-tennessee-initiative-resources-and-access/",
      "https://oit.utk.edu/hpsc/isaac-open-enclave-new-kpb/running-jobs-new-cluster-kpb/",
      "https://oit.utk.edu/hpsc/overview/node-and-storage-investments",
      "https://www.eecs.utk.edu/wp-content/uploads/2024/03/01.26.2024_VictorHazlewood.pdf"
    ],
    "notes": "ADDED Oct 2026. UT projects get 1TB of Lustre project space at no cost. ISAAC also runs a Secure Enclave with its own GPU node for sensitive data.",
    "caveat": "The AI Tennessee H100s are not generally available; UTK faculty request access. The campus GPU partitions are open to every ISAAC-NG user but capped at 2 GPUs per user per day-long job. Dedicated capacity costs extra: a quarter-share of an H100 node (one GPU) is quoted at $46,965.",
    "networks": []
  },
  {
    "state": "Tennessee",
    "school": "University of Memphis Cecil C. Humphreys School of Law",
    "university": "University of Memphis",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "iTiger (NSF MRI, online since October 2024): 88 GPUs — NVIDIA H100, RTX 6000 Ada and RTX 5000 Ada — across 11 nodes. BigBlue campus cluster: 8 A100 (4 AMD Genoa nodes x 2) and 12 V100 (6 Intel nodes x 2).",
    "slurm": "Yes",
    "hpcName": "iTiger (regional GPU cluster) + BigBlue (UofM High Performance Computing)",
    "specs": "iTiger: 928 AMD cores, 233TB SSD and 1.32PB HDD storage, shared with 10+ Mid-South institutions. BigBlue: 120 nodes, about 9,150–9,660 cores, 48TB RAM, EDR/HDR100 InfiniBand, Open OnDemand.",
    "sources": [
      "https://classic.memphis.edu/research/impact/newsletter_2024/october24_stories/artificial_intelligence_gpu_cluster.php",
      "https://itiger-cluster.github.io/",
      "https://arxiv.org/html/2504.14786v1",
      "https://www.memphis.edu/hpc/current-faculty-students.php",
      "https://www.memphis.edu/research-computing/grants_and_citing/research_servers_and_clusters.php"
    ],
    "notes": "ADDED Oct 2026. BigBlue accounts are open to enrolled students, faculty and staff using normal UofM credentials. iTiger is described by the university as fully operational and available for academic and research use, via a separate cluster registration; it also offers seed funding, workshops and tutorials.",
    "networks": []
  },
  {
    "state": "Tennessee",
    "school": "Belmont University College of Law",
    "university": "Belmont University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://www.belmont.edu/data/",
      "https://www.belmont.edu/academics/majors-programs/computer-science/",
      "https://www.belmont.edu/about/glance/"
    ],
    "notes": "ADDED Oct 2026. Belmont's Data & AI Collaborative is a university-wide analytics and curriculum initiative rather than computing infrastructure, and its Computer Science program is undergraduate-focused; no research-computing service or cluster was found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Tennessee",
    "school": "Lincoln Memorial University Duncan School of Law",
    "university": "Lincoln Memorial University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://www.lmunet.edu/information-services/computer-and-software-use-policy",
      "https://nursingtampacatalog.lmunet.edu/computer-servicesresources",
      "https://graduatecatalog.lmunet.edu/student-computer-accounts-emailuniversity-computerswebadvisor"
    ],
    "notes": "ADDED Oct 2026. LMU's Information Services pages cover the help desk, email, computer-use policy and student computer labs on each campus (including Knoxville, where the law school is); no research-computing service was found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "North Carolina",
    "school": "Wake Forest University School of Law",
    "university": "Wake Forest University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "DEAC GPU partition: 2x NVIDIA H200 (1 node), 8x A100 80GB (2 nodes x 4), 8x A100 40GB (2 nodes x 4) and V100 32GB — 34 GPU cards across 9 nodes",
    "slurm": "Yes",
    "hpcName": "DEAC (Distributed Environment for Academic Computing) cluster",
    "specs": "84 compute nodes, 4,576 cores, 41TB RAM; 287TB persistent storage; a separate 16-node education cluster; Rocky 9. DEAC's hardware page lists the H200 cards as 80GB while its GPU page uses an H200_141GB type name; 141GB is the H200's actual memory, so the 80GB figure looks like a documentation typo.",
    "sources": [
      "https://deac-wiki.readthedocs.io/en/latest/cluster-gpu.html",
      "https://deac-wiki.readthedocs.io/en/latest/cluster-hardware.html"
    ],
    "notes": "ADDED Oct 2026. Wake Forest's central research cluster; GPU jobs go to the gpu partition with a #SBATCH --gpus request. Access terms and any charges weren't stated on the pages checked, so ask DEAC before planning a project.",
    "networks": []
  },
  {
    "state": "North Carolina",
    "school": "North Carolina Central University School of Law",
    "university": "North Carolina Central University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "NCShare GPU-as-a-Service: 32 NVIDIA H200 141GB (4 nodes x 8), hosted at MCNC; NCCU is a founding NCShare partner",
    "slurm": "Yes",
    "hpcName": "NCShare (Compute as a Service + GPU as a Service)",
    "specs": "4 GPU nodes with 192 Intel Xeon Platinum 8568Y+ cores and 2TB RAM each. Partitions: gpu (general access, preemptible), interactive-gpu (1-hour jobs, higher priority) and gpu-hp (high priority for NSF GaaS partner institutions, with a monthly GPU-hour pool per institution). Open OnDemand with MATLAB, RStudio and Jupyter.",
    "sources": [
      "https://userguide.ncshare.org/guides/gpu/",
      "https://userguide.ncshare.org/",
      "https://ncshare.org/",
      "https://ncshare.org/about/",
      "https://www.nccu.edu/news/nccu-launches-infrastructure-boost-stem-education-and-technology-resources-north-carolina"
    ],
    "notes": "ADDED Oct 2026. NCShare is NSF-funded, and its user guide says its compute and GPU services are provided at no cost, with self-service allocations for faculty and students of participating institutions; GPU access is requested through each institution's NCShare representative. NCCU co-founded NCShare with Duke, Davidson College and MCNC on about $1.4 million in NSF grants, with priority use for NCCU and other minority-serving and smaller North Carolina institutions; UNC-Chapel Hill and NC A&T have since joined as partners.",
    "networks": [
      "NCShare (North Carolina statewide)"
    ]
  },
  {
    "state": "North Carolina",
    "school": "Campbell University Norman Adrian Wiggins School of Law",
    "university": "Campbell University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "NCShare GPU-as-a-Service: 32 NVIDIA H200 141GB (4 nodes x 8), hosted at MCNC; Campbell is a listed NCShare participant",
    "slurm": "Yes",
    "hpcName": "NCShare (participating institution)",
    "specs": "4 GPU nodes with 192 Intel Xeon Platinum 8568Y+ cores and 2TB RAM each. Partitions: gpu (general access, preemptible), interactive-gpu (1-hour jobs, higher priority) and gpu-hp (high priority for NSF GaaS partner institutions, with a monthly GPU-hour pool per institution). Open OnDemand with MATLAB, RStudio and Jupyter.",
    "sources": [
      "https://userguide.ncshare.org/guides/gpu/",
      "https://userguide.ncshare.org/",
      "https://userguide.ncshare.org/about/participants/"
    ],
    "notes": "ADDED Oct 2026. NCShare is NSF-funded, and its user guide says its compute and GPU services are provided at no cost, with self-service allocations for faculty and students of participating institutions; GPU access is requested through each institution's NCShare representative. NCShare's participant list shows Campbell active in both its legacy and current environments.",
    "caveat": "This is NCShare's hardware, not Campbell's. Participants get the general gpu partition, which is preemptible; the high-priority partition is reserved for NSF GaaS partner institutions, and NCShare prioritizes users whose own institutions lack GPU capacity.",
    "networks": [
      "NCShare (North Carolina statewide)"
    ]
  },
  {
    "state": "North Carolina",
    "school": "Elon University School of Law",
    "university": "Elon University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "NCShare GPU-as-a-Service: 32 NVIDIA H200 141GB (4 nodes x 8), hosted at MCNC; Elon is a listed NCShare participant",
    "slurm": "Yes",
    "hpcName": "NCShare (participating institution)",
    "specs": "4 GPU nodes with 192 Intel Xeon Platinum 8568Y+ cores and 2TB RAM each. Partitions: gpu (general access, preemptible), interactive-gpu (1-hour jobs, higher priority) and gpu-hp (high priority for NSF GaaS partner institutions, with a monthly GPU-hour pool per institution). Open OnDemand with MATLAB, RStudio and Jupyter.",
    "sources": [
      "https://userguide.ncshare.org/guides/gpu/",
      "https://userguide.ncshare.org/",
      "https://userguide.ncshare.org/about/participants/"
    ],
    "notes": "ADDED Oct 2026. NCShare is NSF-funded, and its user guide says its compute and GPU services are provided at no cost, with self-service allocations for faculty and students of participating institutions; GPU access is requested through each institution's NCShare representative. Elon's law school is in Greensboro; access runs through Elon's NCShare representative, not a campus cluster.",
    "caveat": "Confirm before relying on it: NCShare's participant table shows Elon active in the legacy environment but marks its access to the current environment, where the H200s live, as needing testing. As with any participant, the general gpu partition is preemptible.",
    "networks": [
      "NCShare (North Carolina statewide)"
    ]
  },
  {
    "state": "North Carolina",
    "school": "High Point University Kenneth F. Kahn School of Law",
    "university": "High Point University",
    "standalone": false,
    "tier": "unspecified",
    "tierLabel": "GPUs present, specific model not named in source",
    "gpu": "Webb School of Engineering HPC: \"several high-end Nvidia GPUs utilizing CUDA technology\" (model not named). The only models HPU has ever named publicly, in a 2015 article, were Nvidia Fermi and GTX Titan cards; whether they have since been replaced is unknown",
    "slurm": "Not confirmed",
    "hpcName": "Webb School of Engineering High-Performance Cluster (no formal name)",
    "specs": "HPC with 10,000+ CPU cores; 50+ node Linux cluster; several dual multi-core Intel servers running Red Hat Enterprise Linux (as listed on HPU's computer science program pages)",
    "notes": "ADDED Oct 2026. Missing from the directory until now because it is new: the law school's first class enrolled in fall 2024, and the ABA granted provisional approval in February 2026. High Point University's School of Engineering hosts the university's only research cluster. HPU's 2019 new-hires announcement says the engineering school's new scientific-computing system administrator would run the HPC server for several departments across campus (chemistry, physics, mathematics, computer science, engineering and others), so it is meant to serve more than one department. The hardware is documented only on the computer science program pages; there is no service page, user guide, scheduler documentation or account process, and central IT's site doesn't mention research computing. HPU isn't an NCShare participant either. NCShare's participant list includes nearby Elon, Guilford, Wake Forest and NC A&T, but not High Point, so the H200 route that gives three other North Carolina law schools a frontier tier isn't open here unless HPU joins.",
    "caveat": "This is a department-level cluster, not a campus-wide research-computing service. It is housed in and run by the Webb School of Engineering, its GPUs are not identified by model, and no published policy says faculty outside the listed science and engineering departments can get an account. A law faculty member would need to ask the engineering school directly.",
    "networks": [],
    "sources": [
      "https://www.highpoint.edu/computerscience/",
      "https://wsoe.highpoint.edu/cs-academic-programs/b-s-in-computer-science/",
      "https://www.highpoint.edu/blog/2019/10/hpu-announces-new-hires-sept-2019/",
      "https://www.highpoint.edu/blog/2015/08/mymajorathpu-top-majors-at-hpu-21-25/",
      "https://www.highpoint.edu/it/",
      "https://userguide.ncshare.org/about/participants/",
      "https://www.highpoint.edu/blog/2026/03/hpus-law-school-earns-accreditation-milestone/"
    ]
  },
  {
    "state": "Texas",
    "school": "SMU Dedman School of Law",
    "university": "Southern Methodist University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "NVIDIA DGX SuperPOD: 20 DGX A100 nodes x 8 = 160 NVIDIA A100 (1,644 TFLOPS). ManeFrame III (M3) also carries faculty partner nodes with A100 40GB and V100 GPUs, some usable only with the owner's permission.",
    "slurm": "Yes",
    "hpcName": "NVIDIA DGX SuperPOD + ManeFrame III (M3) — O'Donnell Data Science and Research Computing Institute / OIT",
    "specs": "SuperPOD: 2,560 CPU cores, 40–52.5TB memory (SMU's pages differ), up to 200Gb/s InfiniBand per node, 3.5PB shared project storage. M3 is SMU's primary CPU cluster, with Open OnDemand web access.",
    "sources": [
      "https://www.smu.edu/provost/odonnell-institute/hpc",
      "https://www.smu.edu/oit/services/superpod",
      "https://www.smu.edu/provost/odonnell-institute/hpc/systems/mp",
      "https://www.smu.edu/oit/services/m3",
      "https://southernmethodistuniversity.github.io/hpc_docs/about.html",
      "https://southernmethodistuniversity.github.io/hpc_docs/",
      "https://blog.smu.edu/itconnect/2023/05/17/m3-takes-smu-high-performance-computing-to-next-level/"
    ],
    "notes": "ADDED Oct 2026. Accounts come through ColdFront allocations led by faculty or staff, and SMU says M3 is available to faculty, staff and students. Its documentation includes Stata and R guides and a digital humanities workshop, which suits empirical legal work. Whether usage is charged wasn't stated on the pages checked.",
    "networks": []
  },
  {
    "state": "Texas",
    "school": "Texas A&M University School of Law",
    "university": "Texas A&M University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Grace: NVIDIA A100, Quadro RTX 6000, T4 and A40. FASTER (composable): A100, A10, A30, A40 and T4. ACES (NSF testbed): NVIDIA H100 and A30 drawn over a Liqid PCIe fabric, plus a Grace Hopper node. Launch: A30.",
    "slurm": "Yes (Grace, FASTER, ACES)",
    "hpcName": "Texas A&M High Performance Research Computing (HPRC): Grace, FASTER, ACES, Launch",
    "specs": "Grace is a 940-node Dell cluster on HDR100 InfiniBand. GPU time is charged in service units: an A100 on Grace costs 72 SUs an hour, an H100 or Grace Hopper on ACES 128; SUs are allocated per cluster and expire each fiscal year. About 90% of ACES capacity goes to the national ACCESS community.",
    "sources": [
      "https://hprc.tamu.edu/kb/",
      "https://hprc.tamu.edu/kb/User-Guides/AMS/",
      "https://hprc.tamu.edu/files/training/2024/Fall/Applying_for_Accounts_on_HPRC_Clusters_Fall_2024.pdf",
      "https://hprc.tamu.edu/resources/index.html",
      "https://vpr.tamu.edu/wp-content/uploads/2022/09/DOR_Fact_Sheet_HPRC_7_2024.pdf",
      "https://artsci.tamu.edu/mathematics/resources/resources.pdf"
    ],
    "notes": "ADDED Oct 2026. HPRC says all of its systems are available to TAMU faculty, staff and students who need large-scale computing, and its training materials describe the services as free of charge, with basic, startup and research allocation tiers. The backlog asked whether access is campus-bound: eligibility follows the employing System member's headquarters, not where the faculty member works.",
    "caveat": "Two things to confirm with HPRC. First, only faculty of Texas A&M System members headquartered in Brazos County may serve as allocation PIs; the law school is in Fort Worth, but it belongs to Texas A&M University itself, headquartered in College Station (Brazos County), so its faculty appear to qualify. Second, the H100s sit on ACES, which uses ACCESS accounts rather than the HPRC account system; Grace's A100s are the default home resource.",
    "networks": []
  },
  {
    "state": "Texas",
    "school": "University of Houston Law Center",
    "university": "University of Houston",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Carya: 40 NVIDIA L40S and 64 V100 GPUs (public nodes plus faculty-purchased nodes integrated into the cluster)",
    "slurm": "Not confirmed in sources checked",
    "hpcName": "Carya + Sabine — HPE Data Science Institute, Research Computing Data Core (RCDC)",
    "specs": "Carya: 345 HPE ProLiant nodes, about 17,000 CPU cores, 82TB memory, 1.7+ PFLOPS peak, 100Gb/s HDR InfiniBand, 1.56PB disk plus 122TB flash. Sabine is older and smaller; each cluster has its own allocations and accounts.",
    "sources": [
      "https://www.uh.edu/research/rcdc/resources/resources.php",
      "https://www.uh.edu/research/rcdc/resources/hpc/carya.php",
      "https://hpedsi.uh.edu/about/policies",
      "https://www.uh.edu/research/rcdc/index.php",
      "https://stories.uh.edu/f20m-full-speed-ahead/"
    ],
    "notes": "ADDED Oct 2026. RCDC says its resources are available to faculty, students and staff, and HPE DSI teaches free HPC and data science classes every semester. PIs need a CougarNet account and file an allocation request, which a committee reviews against GPU, storage and past usage. UH says the core already serves faculty from business and the humanities, not only science and engineering.",
    "networks": []
  },
  {
    "state": "Texas",
    "school": "Texas Tech University School of Law",
    "university": "Texas Tech University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "RedRaider: Toreador partition — 33 NVIDIA A100 (11 nodes x 3); Matador partition — 40 V100 (20 nodes x 2)",
    "slurm": "Yes",
    "hpcName": "RedRaider — High Performance Computing Center (HPCC)",
    "specs": "About 2.1 PFLOPS peak across Nocona (240 AMD EPYC Rome nodes), Quanah/XLQuanah, Matador, Toreador and Ivy interactive and high-memory partitions; fair-share scheduling on all generally available nodes; campus MATLAB license installed.",
    "sources": [
      "https://www.depts.ttu.edu/hpcc/operations/equipment.php",
      "https://www.depts.ttu.edu/hpcc/accounts/index.php",
      "https://www.depts.ttu.edu/hpcc/about/documents/HPCC_New_User_Training_1_Feb24.pdf?v=preview",
      "https://www.depts.ttu.edu/hpcc/about/Introduction-to-RR-January-2021.pdf"
    ],
    "notes": "ADDED Oct 2026. Free: TTU's HPCC says its generally available CPU and GPU resources, plus a standard storage allotment, are provided at no cost to TTU faculty and research grant leaders, and accounts are free for all TTU faculty. The law school is part of TTU in Lubbock, so faculty use ordinary eRaider accounts.",
    "networks": []
  },
  {
    "state": "Texas",
    "school": "Baylor Law School",
    "university": "Baylor University",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Kodiak community GPU queue: NVIDIA A100 40GB (Dell node x 2), V100 32GB (3 Cray nodes x 2) and P100 16GB (2 Cray nodes x 2). Condo queues add more, including an NVIDIA H200 (141GB) node owned by one research group.",
    "slurm": "No — Kodiak uses a PBS-style batch system (qsub)",
    "hpcName": "Kodiak — BaylorITS High Performance and Research Computing Services",
    "specs": "About 100 compute nodes (Cray CS400 plus HPE DL385 Gen10 Plus), nine GPU nodes, one large-memory system, 2.5PB multi-tiered storage; condominium model in which idle condo nodes are open to all Kodiak users.",
    "sources": [
      "https://rtservices.baylor.edu/hprcs/kodiak/Kodiak_GPUs.html",
      "https://its.web.baylor.edu/solutions-services/high-performance-and-research-computing-services/hprcs-systems",
      "https://rtservices.baylor.edu/hpc/systems.html",
      "https://rtservices.baylor.edu/hpc/kodiak/",
      "https://its.web.baylor.edu/hpc"
    ],
    "notes": "ADDED Oct 2026. Kodiak accounts are available to Baylor faculty and graduate and undergraduate students, requested through the help desk's service catalog; Baylor also offers a free self-paced Linux course to prepare new users.",
    "caveat": "Rated on the community A100s. The H200 sits on a private condo queue; Baylor says most users run on the general gpu queue, and idle condo resources are available to all users, so H200 time is opportunistic at best.",
    "networks": []
  },
  {
    "state": "Texas",
    "school": "UNT Dallas College of Law",
    "university": "University of North Texas at Dallas",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://www.untdallas.edu/oit/about-us.php",
      "https://www.untdallas.edu/oit/help-desk/index.php",
      "https://www.untdallas.edu/lawschool/students/technology.php",
      "https://www.untdallas.edu/lawschool/about/index.php"
    ],
    "notes": "ADDED Oct 2026. UNT Dallas's Office of Information Technology covers the help desk, Microsoft 365, VPN, labs and licensed software such as Mathematica, and the law school's own IT team runs its help desk and practice software; no research-computing service was found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "caveat": "The backlog flagged UNT's Talon cluster in Denton as a possible route through the UNT System. UNT Dallas is a separate university within the system (the law school became part of it in 2015), and no source extends Talon access to UNT Dallas, the same pattern that failed for Missouri and Baltimore. Ask UNT Dallas OIT whether any cross-campus arrangement exists.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Texas",
    "school": "St. Mary's University School of Law",
    "university": "St. Mary's University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://www.stmarytx.edu/academics/department/computer-science/",
      "https://www.stmarytx.edu/academics/centers/cyber-excellence/",
      "https://www.stmarytx.edu/academics/school-of-science-engineering-and-technology/"
    ],
    "notes": "ADDED Oct 2026. St. Mary's has a Computer Science, Electrical and Computer Engineering department and a Center for Cyber Excellence that includes a law professor among its affiliated faculty, but no shared research-computing cluster or service was found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Texas",
    "school": "Thurgood Marshall School of Law",
    "university": "Texas Southern University",
    "standalone": false,
    "tier": "cpuonly",
    "tierLabel": "No GPU confirmed (CPU/RAM-only)",
    "gpu": "None confirmed — the TSU-HPCC cluster's published spec lists Intel Xeon CPUs and Intel Xeon Phi coprocessors (many-core x86 chips, not GPUs)",
    "slurm": "Not confirmed (the cluster was built on Rocks with MPI, OpenMP and Condor)",
    "hpcName": "Texas Southern University High Performance Computing Center (TSU-HPCC)",
    "specs": "54 compute nodes, 944 Intel Xeon cores, 1.75 TFLOPS Linpack; extended with 8 \"super-nodes\" of 4 Xeon CPUs plus 4 Xeon Phi coprocessors each; R, Mathematica, MATLAB and COMSOL installed.",
    "sources": [
      "https://coset.tsu.edu/hpcc/",
      "https://tsu.edu/about/administration/division-of-research-and-innovation/dri-urc-hpcc.php",
      "https://coset.tsu.edu/lab16/",
      "https://coset.tsu.edu/labs/"
    ],
    "notes": "ADDED Oct 2026. TSU-HPCC serves TSU faculty, staff and students and, notably, TSU's research office lists the Center for Justice Research and the Thurgood Marshall School of Law Student Bar Association among its internal collaborations for FY2022, so the center has worked with the law school before. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "caveat": "Every published hardware description dates to about 2014, so the cluster may have been upgraded since. Confirm current hardware with the HPCC director before ruling out GPUs.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Virginia",
    "school": "University of Virginia School of Law",
    "university": "University of Virginia",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Afton/Rivanna: 55 GPU nodes — an 18-node NVIDIA DGX A100 BasePOD (8x A100 80GB each), 2 nodes x 8 A100 40GB, 12 nodes x 8 A40, 14 nodes x 8 A6000, plus V100, RTX 3090 and RTX 2080 Ti — and a new NVIDIA HGX H200 node added to Afton",
    "slurm": "Yes",
    "hpcName": "UVA Research Computing: Afton + Rivanna (standard security), Rio (secure HPC)",
    "specs": "About 520 CPU nodes across Rivanna and Afton, 8+ PB shared storage. GPUs are billed in service units (an A100 80GB costs about 51 SUs per GPU-hour). Rio and Ivy handle sensitive and highly sensitive data.",
    "sources": [
      "https://rc.virginia.edu/services/compute-and-storage/aftonrivanna",
      "https://archive.rc.virginia.edu/userinfo/pricing/",
      "https://archive.rc.virginia.edu/userinfo/hpc/",
      "https://archive.rc.virginia.edu/tags/rivanna/",
      "http://uvarc-website-staging.s3-website-us-east-1.amazonaws.com/tags/hpc/",
      "https://github.com/UVA-DSI/Open-Data-Lab/blob/master/documentation/Rivanna.md"
    ],
    "notes": "ADDED Oct 2026. Free: UVA RC's pricing page lists Standard allocations as free, with service units that expire after 12 months, and faculty and research staff are eligible; a faculty member requests the allocation and adds students. RC staff also consult on R, Python and MATLAB.",
    "caveat": "The frontier rating rests on a single HGX H200 node, which UVA Research Computing announced as the first of several planned; nearly all GPU capacity is A100-class or older. The H200 announcement was found on a mirror of UVA RC's site.",
    "networks": []
  },
  {
    "state": "Virginia",
    "school": "William & Mary Law School",
    "university": "College of William & Mary",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "SciClone GPU resources include a node with 8x NVIDIA L40S 48GB (Intel Xeon Gold 5320, 52 cores); older Hima nodes carry single Tesla-class GPUs",
    "slurm": "Not confirmed in current docs (older Bora/Hima pages use TORQUE)",
    "hpcName": "SciClone (main campus) + VIMS cluster — W&M Research Computing",
    "specs": "Sub-clusters include kuro (2024: 47 AMD EPYC 9334 nodes, 3,008 cores, HDR InfiniBand), femto and bora; about 600 TFLOPS across both campuses; software stack refreshed in summer 2024; 72-hour job limit (48 on kuro).",
    "sources": [
      "https://www.wm.edu/offices/it/services/researchcomputing/atwm/systemarchitecture/nodes/",
      "https://www.wm.edu/offices/it/services/hpc/hw/nodes/bora/index.php",
      "https://cdsp.wm.edu/applied-science/research/facilities/",
      "https://www.wm.edu/offices/it/services/researchcomputing/using/software/"
    ],
    "notes": "ADDED Oct 2026. Free: W&M says Research Computing's HPC resources are free to use for faculty, staff and students working on research or classroom projects. Only part of W&M's GPU table was visible in the sources checked, so there may be more GPU nodes than listed here.",
    "networks": []
  },
  {
    "state": "Virginia",
    "school": "Antonin Scalia Law School",
    "university": "George Mason University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "Hopper: 124 NVIDIA A100 80GB (31 nodes x 4), 16 A100 40GB (2 DGX nodes x 8) and 4 H100 80GB (1 node); some nodes are split into MIG slices for interactive use",
    "slurm": "Yes",
    "hpcName": "Hopper (and Argo) — Office of Research Computing (ORC)",
    "specs": "11,840+ cores, HDR InfiniBand at 100Gb/s per node, 4PB VAST flash scratch; contributor-funded nodes (18 A100/H100 GPU nodes and 42 CPU nodes) give their owners preemptive access.",
    "sources": [
      "https://wiki.orc.gmu.edu/mkdocs/About_Hopper/",
      "http://wiki.orc.gmu.edu/",
      "https://orc.gmu.edu/resources/computing-systems/computing-resources/",
      "https://wiki.orc.gmu.edu/mkdocs/Running_GPU_Jobs/",
      "https://orc.gmu.edu/releasing-new-nodes-on-hopper/"
    ],
    "notes": "ADDED Oct 2026. Free: ORC provides Argo and Hopper as a free resource to all GMU faculty and faculty-sponsored students. The backlog warned that \"Hopper\" is only the cluster's name; here the H100 hardware is real, confirmed by ORC's own GPU tables.",
    "caveat": "The H100s are a single 4-GPU node, and ORC says contributors keep preemptive access to the nodes they funded, so H100 jobs from others can be bumped. Interactive sessions get MIG slices only; full A100 or H100 GPUs need a batch job.",
    "networks": []
  },
  {
    "state": "Virginia",
    "school": "University of Richmond School of Law",
    "university": "University of Richmond",
    "standalone": false,
    "tier": "modern",
    "tierLabel": "Modern — A100/A40/L40-class confirmed",
    "gpu": "Spydur: 2 nodes x 8 NVIDIA A40 (Sci partition) and 1 node x 2 A100 (ML partition)",
    "slurm": "Yes",
    "hpcName": "Spydur — University of Richmond HPC (Provost's Office with Information Services)",
    "specs": "30 shared nodes: 1,508 Xeon cores, 20TB memory, 336TB RAID storage; partitions Basic, Medium (768GB), Large (1.5TB), ML and Sci; condo nodes for faculty who buy them; a third system, Arachne, has been announced.",
    "sources": [
      "https://data.richmond.edu/About-HPC-at-UR/index.html",
      "https://physics.richmond.edu/research/equipment/computing-cluster.html",
      "https://spidertechnet.richmond.edu/TDClient/1955/Portal/KB/ArticleDet?ID=135862",
      "https://is.richmond.edu/about/index.html"
    ],
    "notes": "ADDED Oct 2026. Richmond says Spydur is available to all faculty and their students who do research, with a computational support specialist; the Faculty Hub runs getting-started sessions covering RStudio and Jupyter.",
    "networks": []
  },
  {
    "state": "Virginia",
    "school": "Washington and Lee University School of Law",
    "university": "Washington and Lee University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://www.wlu.edu/computer-science-department/about-the-department/computer-labs",
      "https://www.wlu.edu/physics-and-engineering-department/about-the-department/facilities",
      "https://en.wikipedia.org/wiki/Washington_and_Lee_University_School_of_Law"
    ],
    "notes": "ADDED Oct 2026. W&L is a small liberal arts university; no central research-computing service or shared cluster was found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "caveat": "Not a shared service, but worth asking about: W&L's computer science faculty run their own research servers, including deep-learning machines, and the campus IQ Center offers computational instruments and imaging. Either could support a collaboration.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Virginia",
    "school": "Regent University School of Law",
    "university": "Regent University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://www.regent.edu/college-of-arts-and-sciences/departments/engineering-computer-science/",
      "https://www.regent.edu/program/bs-in-computer-science/",
      "https://en.wikipedia.org/wiki/Regent_University"
    ],
    "notes": "ADDED Oct 2026. Regent's Engineering & Computer Science department offers teaching programs, many of them online; no research-computing service or cluster was found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Virginia",
    "school": "Liberty University School of Law",
    "university": "Liberty University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://rescale.com/resources/liberty-university-engineering-invests-in-cloud-native-hpc-to-enable-new-rd-capabilities/",
      "https://www.liberty.edu/engineering/research/creationeering/",
      "https://www.liberty.edu/residential-at-liberty/bs-computer-science-general/"
    ],
    "notes": "ADDED Oct 2026. No campus-wide research-computing service or on-premises cluster was found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "caveat": "Liberty's School of Engineering runs its simulation work on Rescale's commercial cloud HPC platform, an engineering arrangement for its classes and research rather than a campus-wide service a law faculty member could join.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Virginia",
    "school": "The Judge Advocate General's Legal Center and School (U.S. Army JAG School)",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "None — U.S. Army school, not part of a university",
    "specs": "",
    "notes": "ADDED Oct 2026. Listed for completeness: the ABA's approved list includes the Army's JAG School alongside the 198 J.D. schools. It is a special case. It awards only an LL.M. in Military Law, and the ABA says it reviews the program against the relevant standards rather than formally accrediting it. It sits on the University of Virginia's North Grounds and its students often share courses with UVA Law, but it is run by the U.S. Army, not by UVA, so UVA's research-computing resources (see the UVA row) are not its own. Its faculty are serving judge advocates, mostly Army lieutenant colonels and majors.",
    "caveat": "The NSF ACCESS route that applies to every other standalone row is less clear here. ACCESS treats a named list of military service academies as academic institutions, and the JAG School is not on that list. Federal agency research staff can qualify only if their agency doesn't usually give them access to other computing resources. Treat ACCESS eligibility as unconfirmed for this school.",
    "networks": [],
    "sources": [
      "https://www.americanbar.org/groups/legal_education/accreditation/approved-law-schools/",
      "https://www.americanbar.org/groups/legal_education/accreditation/approved-law-schools/alphabetical/",
      "https://en.wikipedia.org/wiki/The_Judge_Advocate_General%27s_Legal_Center_and_School",
      "https://allocations.access-ci.org/allocations-policy"
    ]
  },
  {
    "state": "New York",
    "school": "University at Buffalo School of Law",
    "university": "University at Buffalo (SUNY)",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "UB-HPC, open to all UB researchers: NVIDIA GH200 Grace Hopper (arm64 partition, separate allocation), H100, A100 and V100. Industry partitions add A100, L40S and HGX H100 nodes that faculty reach only through the preemptible scavenger partition. Plus Empire AI (alpha+ and beta) allocations under the SUNY umbrella.",
    "slurm": "Yes",
    "hpcName": "Center for Computational Research (CCR): UB-HPC + Empire AI",
    "specs": "UB-HPC: 20,000–26,000+ CPU cores on InfiniBand; a separate faculty cluster of 50+ owner partitions whose idle nodes are open via scavenger; 1TB free storage per UB faculty group; Open OnDemand web access.",
    "sources": [
      "https://www.buffalo.edu/ccr/support/research_facilities.html",
      "https://docs.ccr.buffalo.edu/en/latest/getting-started/",
      "https://docs.ccr.buffalo.edu/en/latest/available-resources/",
      "https://www.buffalo.edu/ccr/support/research_facilities/external-resources.html",
      "https://docs.ccr.buffalo.edu/en/latest/howto/empireai/",
      "https://www.buffalo.edu/ccr/support/research_facilities/ub-hpc/academic-hardware-specs.html"
    ],
    "notes": "ADDED Oct 2026. Free: CCR says there is no cost for UB faculty groups to use UB-HPC (students need a faculty sponsor), and UB describes its HPC resources as provided at no cost to faculty in every discipline. Empire AI's data center is being built at UB; faculty apply for Empire AI time through SUNY proposal calls or an off-cycle request form, and CCR staff onboard approved groups.",
    "networks": [
      "Empire AI (New York State)"
    ]
  },
  {
    "state": "New York",
    "school": "Fordham University School of Law",
    "university": "Fordham University",
    "standalone": false,
    "tier": "unspecified",
    "tierLabel": "GPUs present, specific model not named in source",
    "gpu": "Magis cluster: separate GPU nodes totaling 64,000 CUDA cores (model not named)",
    "slurm": "Not confirmed",
    "hpcName": "Magis — Fordham IT Research Computing",
    "specs": "56 TFLOPS cluster built by Fordham IT with the Office of Research to support research and sponsored projects.",
    "sources": [
      "https://www.fordham.edu/information-technology/it-services/academic-technology/research-computing/",
      "https://www.fordham.edu/academics/research/office-of-research/initiatives-and-infrastructure/internal-funding-opportunities/high-performance-computing/"
    ],
    "notes": "ADDED Oct 2026. Access is by application: the Office of Research convenes a three-member committee to review project requests, with priority for tenured, tenure-track and emeriti faculty, so law faculty apply like anyone else. Students and outside collaborators are added under a faculty PI.",
    "networks": []
  },
  {
    "state": "New York",
    "school": "Benjamin N. Cardozo School of Law",
    "university": "Yeshiva University",
    "standalone": false,
    "tier": "unspecified",
    "tierLabel": "GPUs present, specific model not named in source",
    "gpu": "Katz School of Science and Health: an AI lab and GPU computing hub, including a high-performance GPU server from MIT Cambridge Research (models not named)",
    "slurm": "Not confirmed",
    "hpcName": "Katz School AI lab / GPU computing hub",
    "specs": "Katz's computer science department also runs an IoT lab and a security operations center; a 30,000 sq ft expansion at the Beren campus opened in fall 2025.",
    "sources": [
      "https://www.yu.edu/news/featured/yeshiva-universitys-katz-school-expands-stem-footprint-midtown-manhattan",
      "https://www.yu.edu/news/yu/yeshiva-university-launches-two-masters-programs-computer-science",
      "https://www.yu.edu/katz/computer-science-engineering",
      "https://en.wikipedia.org/wiki/Benjamin_N._Cardozo_School_of_Law"
    ],
    "notes": "ADDED Oct 2026. No central Yeshiva University research-computing service or shared cluster was found. Cardozo itself runs a Data Law Initiative.",
    "caveat": "This GPU capacity belongs to the Katz School's computer science programs, not a university-wide research-computing service, and nothing found says Cardozo faculty can use it; a collaboration with Katz faculty would be the likely route.",
    "networks": []
  },
  {
    "state": "New York",
    "school": "St. John's University School of Law",
    "university": "St. John's University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://www.stjohns.edu/academics/programs/computing-and-machine-learning-bachelor-science",
      "https://www.stjohns.edu/academics/programs/computer-science-master-science"
    ],
    "notes": "ADDED Oct 2026. St. John's offers computing and machine learning degrees and a Machine Learning Institute for seminars and student research, but no shared research-computing cluster or service was found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "New York",
    "school": "Elisabeth Haub School of Law",
    "university": "Pace University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://www.pace.edu/news/seidenbergs-deep-learning-afternoon-thunder-compute",
      "https://www.pace.edu/seidenberg/faculty-and-research/faculty-and-student-research",
      "https://www.pace.edu/seidenberg/academics/departments/computer-science/welcome-chair"
    ],
    "notes": "ADDED Oct 2026. Pace's Seidenberg School of Computer Science and Information Systems does AI and data-science teaching and research, and its Data Science Club has worked with a commercial GPU cloud provider, but no shared research-computing cluster or service was found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "New York",
    "school": "Touro Law Center",
    "university": "Touro University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "",
    "hpcName": "None found",
    "specs": "",
    "sources": [
      "https://www.touro.edu/about/research/",
      "https://rmi.touro.edu/ihe/",
      "https://las.touro.edu/academics/computer-science/"
    ],
    "notes": "ADDED Oct 2026. Touro's university-wide AI initiative, led by an associate provost for AI, focuses on teaching, policy and operations, and its Research Council includes the law school's associate dean for research; no shared research-computing cluster or service was found. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Florida",
    "school": "Ave Maria School of Law",
    "university": "",
    "standalone": true,
    "tier": "standalone",
    "tierLabel": "N/A — standalone (no parent university)",
    "gpu": "",
    "slurm": "",
    "hpcName": "None — independent law school, no parent institution",
    "specs": "",
    "sources": [
      "https://www.avemarialaw.edu/about-us/",
      "https://www.floridabar.org/the-florida-bar-news/ave-maria-school-of-law-headed-to-collier-county/",
      "https://en.wikipedia.org/wiki/Ave_Maria_School_of_Law"
    ],
    "notes": "ADDED Oct 2026. Independent: the school describes itself as an independent law school established in 1999, and the Florida Bar News reported at the time of its move to Florida that it is not affiliated with Ave Maria University, despite the shared founder and name. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ]
  },
  {
    "state": "Florida",
    "school": "Jacksonville University College of Law",
    "university": "Jacksonville University",
    "standalone": false,
    "tier": "none",
    "tierLabel": "No HPC resource identified",
    "gpu": "",
    "slurm": "No",
    "hpcName": "None found",
    "specs": "",
    "notes": "ADDED Oct 2026. Missing from the directory until now because it is new: the law school opened in downtown Jacksonville in August 2022 and received ABA provisional approval in February 2024. Jacksonville University is a private, teaching-focused university. Its central IT site is a help desk (accounts, VPN, Blackboard) with no research-computing service. The Computing Science department teaches in computer classrooms, and the STEAM Institute's CyberRange, forensics and FinTech labs are teaching spaces, not shared research clusters. Nothing found on the Marine Science Research Institute's pages either. Statewide routes don't reach it today. UF's HiPerGator policy for other Florida universities is written for the public State University System, research use there must be paid for, and outside accounts go through an institutional liaison; JU isn't on UF's liaison list. JU is also absent from SSERCA's member list. || National floor: this tier describes the PARENT INSTITUTION's own resource, not the ceiling on what a faculty member here can reach — see the NSF ACCESS note in the site header and README. That national route is tagged in this row's shared-networks field; it is not evidence about this school specifically, and the sources listed below are only those used to verify this row.",
    "networks": [
      "NSF ACCESS (national — any US academic institution)"
    ],
    "sources": [
      "https://www.ju.edu/it/",
      "https://www.ju.edu/computingscience/",
      "https://www.ju.edu/news/2022-08-01-ju-constructing-innovative-steam-institute-center.php",
      "https://www.ju.edu/msri/",
      "https://it.ufl.edu/rc/documentation/policies/other-universities/",
      "https://docs.rc.ufl.edu/access/federated_liaisons/",
      "https://sserca.flrnet.org/",
      "https://www.abajournal.com/news/article/newer-law-school-in-florida-gets-provisional-accreditation-from-the-aba"
    ]
  },
  {
    "state": "Michigan",
    "school": "Wayne State University Law School",
    "university": "Wayne State University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "WSU HPC Grid: nodes with 4x NVIDIA H200 (141GB), plus V100 nodes, requested through SLURM's gpu QoS (or the preemptible requeue QoS)",
    "slurm": "Yes",
    "hpcName": "Wayne State HPC Grid — Computing & Information Technology (C&IT)",
    "specs": "Heterogeneous SLURM cluster with dozens of node classes (Intel and AMD, up to 755GB RAM per node, 25–100Gb Ethernet and InfiniBand); about 1.5PB Panasas storage, up to 10TB per user and 50TB per group on request.",
    "sources": [
      "https://services.wayne.edu/TDClient/277/Portal/KB/Article/20229/Grid-FAQs",
      "https://tech.wayne.edu/hpc/nodes",
      "https://tech.wayne.edu/kb/high-performance-computing/hpc-tutorials/500113",
      "https://services.wayne.edu/TDClient/277/Portal/Requests/Service/10662/High-performance-computing-HPC",
      "https://tech.wayne.edu/kb/high-performance-computing/hpc-tutorials/500133"
    ],
    "notes": "ADDED Oct 2026. Free: Wayne State says there is currently no cost to use its Grid computing service, which is open to any WSU student, staff or faculty member with a research-related computational task and an AccessID. Not to be confused with Washington State University (also \"WSU\"), whose Kamiak cluster appears in the same searches. This school was missing from both the dataset and the backlog; a per-state count check against the ABA list caught it.",
    "caveat": "Many Grid nodes are owned by research groups, and the published node table doesn't say which GPU nodes are shared, so H200 time may come through the preemptible requeue QoS. Grid priority is first come, first served.",
    "networks": []
  },
  {
    "state": "Ohio",
    "school": "Ohio Northern University Pettit College of Law",
    "university": "Ohio Northern University",
    "standalone": false,
    "tier": "frontier",
    "tierLabel": "Frontier — H100/H200/B200 confirmed",
    "gpu": "128 NVIDIA H100 (Hopper) GPUs on Cardinal — 94GB HBM2e each with NVLink, four per node across 32 Dell PowerEdge XE9640 nodes. 776 NVIDIA A100s on Ascend (four each on 24 nodes, three each on 84 nodes, two each on 214 nodes). NVIDIA V100 32GB on Pitzer.",
    "slurm": "Yes — Slurm on Cardinal (RHEL 9); PBS compatibility disabled on Cardinal",
    "hpcName": "Ohio Supercomputer Center (OSC) — Cardinal, Ascend, Pitzer",
    "specs": "Cardinal: 378 Dell nodes, 39,312 cores, Intel Xeon CPU Max 9470 (Sapphire Rapids) with 128GB HBM2e and 512GB DDR5 per dense node, NDR200 InfiniBand (400Gb/s on GPU nodes), RHEL 9. Ascend: 322-node Dell AMD EPYC machine, 40,448 cores, 16.0 PFLOPS theoretical peak. Roughly 20PB storage plus 23PB expandable backup. Open OnDemand originated at OSC.",
    "sources": [
      "https://www.osc.edu/supercomputing/support/account",
      "https://www.osc.edu/services/cluster_computing",
      "https://www.osc.edu/resources/technical_support/supercomputers/cardinal",
      "https://www.osc.edu/content/available_service_categories",
      "https://en.wikipedia.org/wiki/Ohio_Supercomputer_Center"
    ],
    "notes": "ADDED Oct 2026. Ohio Northern, a private university in Ada, reaches OSC the same way Ohio's other private law schools do: OSC is the state's shared HPC center for Ohio higher education, funded through the Ohio Department of Higher Education and open to public and private institutions. This school was missing from both the dataset and the backlog; a per-state count check against the ABA list caught it.",
    "caveat": "Rated through the Ohio Supercomputer Center, not an Ohio Northern cluster; no campus research-computing service was found at ONU, and no ONU-specific OSC project page turned up, so confirm onboarding with OSC.",
    "networks": [
      "OSC (Ohio statewide)"
    ]
  }
]