KB005: Ursa Major Service Tiers & Allocation Guide

Scope: Ursa Major (Google Cloud) Audience: PIs, Researchers, Lab Managers Last Updated: Feb 17, 2026 (Operational Alignment)

1. Overview

Ursa Major provides a tiered resource model to ensure sustainable access to cloud computing for the UCR research community.

2. Tier 1: Strategic Allocation (The “Pool”)

Objective: Democratize access to strategic cloud tools (AI & Archive) that cannot be delivered on-premise. Cost: Covered by the Campus Shared Pool (Zero cost to PI).

Included Resources (Strict Whitelist)

| Resource | Specification | Notes | | :— | :— | :— | | Compute | e2-standard-2 | Standard general-purpose VMs. | | GKE | e2-standard-4 | Container orchestration clusters. | | Storage (Object) | Coldline Class | Long-term compliance archiving/backups. | | Storage (Object) | Standard Class | Working datasets (small scale). | | Persistent Disk | Zonal Balanced | VM Boot and local data disks. | | Database | Cloud SQL | Postgres HA. | | Analytics | BigQuery | Enterprise Edition. | | AI | Gemini API | Google Models Only (1.5 Pro/Flash). | | Agent Builder | Vertex AI Search | Subsidized if within standard vector limits. |

Excluded (Billable Overages)

Warning: The following are NOT covered by the subsidy and will trigger a recharge invoice:

• Marketplace Offerings: 3rd Party Models (Claude, Llama, Mistral) via Vertex AI.
• High-Performance VMs: n1, c2, m1, c3 families.
• Cloud GPUs: a2, t4, l4, v100.
• Bare Metal Solution.

3. Tier 2: Commercial Recharge (The “Override”)

Objective: Support cloud-native workloads that exceed the pool limits or require specialized infrastructure. Cost: Full pass-through billing to Grant COA at PSSA List Price via ITS.

Common Tier 2 Resources:

• Google Filestore (NFS): Managed high-performance file shares.
• Cloud GPUs: All NVIDIA GPU types (A100, T4, L4).
• High-Perf VMs: n1, c2, m1, c3 families.
• Marketplace Models: Claude (Anthropic), Llama (Meta) via Vertex AI.
• Instructional Use: Large-scale courses (unless using Consumer/Free tiers).

4. Tier 3: Dedicated Subscription (The “Enterprise”)

Objective: Large-scale, multi-year projects requiring their own contractual terms. Cost: Direct Contract between the Lab and Google (with ITS administrative oversight).

5. Strategic GPU & Compute Offload Hierarchy

When workloads exceed the Tier 1 Ursa Major limits, we enforce a strict 4-tier hierarchy to maximize local ROI before utilizing national resources:

1. UCR HPCC (Primary): The main and preferred destination for displaced GPU and advanced compute workloads.
   • Standard Access: $1,000/year Flat Rate per Lab.
   • Hardware Buy-In (CAPEX): Purchase hardware; we host it at zero ongoing cost.
2. NRP Nautilus Cluster (Secondary): A regional/national Kubernetes grid leveraging UCR hardware contributions. Best for containerized GPU tasks that overflow HPCC.
3. NAIRR Pilot (Tertiary): National AI Research Resource. Used for specialized, high-end AI allocations (e.g., massive H100 arrays, exotic accelerators) when local/regional capacity is insufficient.
4. NSF ACCESS (Quaternary): Traditional national supercomputing grids reserved for massive-scale jobs requiring thousands of nodes.