Computing resources report¶

For the College of Information Science Computing Resources Working Group. This report connects available services, changes in access and a practical response. The resource inventory contains service-specific access, cost and contact details.

Review basis: dated provider and campus documentation, mostly checked October 8–19, 2026, with campus HPC allocation and teaching policies rechecked October 20, plus the explicitly attributed ORAI update. This is a working assessment, not a completed college-wide needs study.

Key findings¶

• Resources exist, but access and support differ. College and campus services, national allocations and bounded free platforms serve different work.
• Continuity is the gap illustrated by the provider history. Closed services, narrower eligibility, shorter windows and tighter limits can break a course or research workflow.
• The opportunity is a supported route, not only more hardware. Match activities to existing services, identify the remaining gap, and test shared capacity, expertise and portable environments.

Available resources¶

The inventory contains 32 entries: 6 college, 20 university and 6 national/external. It covers hardware, software and models, data, RSE and technical support, training and collaboration. Unknown local equipment, licenses and available support hours are recorded as unknown, rather than assumed absent.

Campus starting points include Puma CPU/GPU computing, software licensing, library data and digital-scholarship services, and research software expertise.

Campus HPC: Puma classroom access¶

Verified October 20, 2026: the current HPC teaching policy explicitly permits Puma class groups. Earlier references to an Ocelote-only policy are superseded.

Check the PI's remaining allocation with va and actual group limits/usage with job-limits <group_name>, confirm class-group setup and device requirements, and test assignment fit before required use. No signed-in balance, peak-class rehearsal or reserved capacity is established here. Allocation and charging policy · Job limits · HPC Consult.

The hardware page lists Lynx/H200 as not available yet. Its Puma MIG memory table and narrative differ; confirm the requested device rather than assuming a memory quota. ACCESS remains the proposed large CPU/GPU allocation route for the separate instructional pilot.

Briefing cross-check: a Jeremy Frumkin (UITS) hardware slide in supplied October 14–15 AI AWS Days notes lists 40 GB A100 MIG slices on Puma, marks Ocelote retired and lists Lynx for 2026. These are cluster hardware descriptions, not a PI/classroom allocation or proof of current availability. The current public policies above govern the documented access route; confirm the GPU/MIG shape before sizing a class. The private notes export is not published here.

Browse the inventory · Campus AI transition and supplied prices

External routes and their limits¶

Open a resource group for its role, status and evidence. These are dated comparisons; availability does not establish account eligibility or reserved capacity.

Commercial offers and free credits require particular care with payment methods, account scope and paid conversion. Full commercial comparison.

The growing gap¶

The selected history shows losses across compute, classroom workflows, repositories, hosting and databases. It is not a census or a measured decline in total free compute, nor does it establish the size of the college’s unmet demand. GitHub Free repositories and base free Colab remain available.

Documented changes by year¶

What the college still needs to establish¶

Which activities are blocked; who is affected; whether the barrier is capacity, eligibility, awareness, affordability or support; and what existing services can realistically absorb. The resource tool prepares an individual brief. An optional private pilot queue lets Barney and Greg review submitted briefs. Participation and verification are still needed before making college-wide claims.

The opportunity¶

Instructional-computing case study: the proposed DS/MLIS pilot uses ACCESS for CPU and GPU. Campus AWS is only for component testing; free GPU notebooks support preliminary benchmarks, followed by target-resource replay. CloudBank remains conditional. The pilot’s $0 cash and Greg’s four total hours/week are its own constraints, not college-wide budget or staffing findings. TA/course-lead effort is separate; paid needs require confirmed eligible grant funding.

No award, capacity reservation, provider benchmark or classroom deployment is established here. All 13 pilot readiness checks remain planned. Pilot details and request draft.

Planned next step: ORAI AI Innovation Pilot¶

Greg Chism will draft a separate proposal to showcase ORAI tools alongside his infrastructure components in an AI-native classroom, through ORAI AI Innovation Pilots. His capacity is four hours/week total, shared with other duties. The earlier two-course illustration requires 230 hours over 23 weeks; the corrected time ceiling is at most 92 hours, leaving a 138-hour gap before other commitments. It requires rescoping or confirmed distinct support.

The ORAI program targets one working proof of concept within eight weeks of execution after assessment, with one pilot per faculty/staff member per year and essential data/access ready at entry. Fit, acceptance, staff assignments and infrastructure are unconfirmed. Drafting the proposal is next; it has not been submitted. An ORAI prototype does not provide semester-long operations or an ACCESS award.

Key considerations¶

The working-group primers turn these into discussion questions. The approaches and recommendations remain candidates for group review.

Working-group next steps¶

By the end of Fall 2026, aim for preliminary drafts of the inventory, needs assessment, portfolio of approaches and recommendations. Refine and circulate within the college early in Spring 2027. Identify owner roles, a university channel for collecting needs, resource-owner checks and participation across faculty, students and technical staff.

Prioritize practical college actions while separating capacity, agreements and shared staffing questions that need university participation. Working materials.

Evidence and limitations¶

The historical ledger retains 181 source and verification records, including unavailable sources and their limitations. Campus sources and review dates are attached to inventory entries. Dates of observation, policy effect and recheck are distinct; a page response does not test a signed-in account or available capacity.

The Arizona AI Assistant transition, planned $200/month Bedrock credit and supplied model prices come from Wolfgang Jetner at ORAI, relayed by Greg Chism. Recipient scope, renewal, overage rules, migration timing and campus API access still need confirmation. The screenshot was received October 19; its capture/effective date is unknown.

No college-wide survey totals, confirmed spare staffing, institutional endorsement or approved funding are inferred. The request tool generates a local brief; optional Luna drafting requires consent for public/synthetic information and does not submit a request to the working group.

Source evidence · Chronological research archive · Original audit text · Grouped supplements