How busy the SeaWulf and NVwulf clusters are, who uses them and how Stony Brook compares, refreshed monthly.
Stony Brook's research clusters, in numbers. SeaWulf is the campus CPU cluster and NVwulf the GPU cluster for AI and simulation. The figures below come from the clusters' job records through October 8, 2026 and are refreshed monthly.
At a glance
NVwulf GPU time used in September89%of installed GPU-hours (H200 90%, B40 87%)
NVwulf GPU-hours, last 12 months384,870October 2025 to September 2026
People running NVwulf jobs in September96from 26 departments
SeaWulf core-hours, last 12 months149.7M468 people ran jobs in September
Cloud cost of the same work$8.6Mlast 12 months, both clusters, AWS on-demand prices
NVwulf share of campus GPU-hours58%last 12 months, vs SeaWulf A100s
Small lines show the trend from October 2025 to September 2026.
GPU work has moved to NVwulf
GPU-hours per month on each cluster, by the month a job started. NVwulf opened in June 2025 with 48 H200 GPUs and added 44 RTX PRO 6000 Blackwell (B40) GPUs in January 2026.
NVwulfSeaWulf A100 GPUs
How full the clusters are
September average use as a share of each pool's practical ceiling (the most GPUs, or the 99th percentile of cores, ever in use at once). Above about 90% jobs queue for longer.
NVwulf GPUsSeaWulf GPUsSeaWulf CPUs
Who uses the GPUs
NVwulf GPU-hours by department over the last 12 months. 77 research groups from 26 departments have used NVwulf since it opened.
New and returning users
People who ran at least one NVwulf job each month, split into first-time users and people who had used it before.
ReturningNew this month
How GPU jobs end
Share of NVwulf GPU-hours by how the job ended. 49% of GPU time ran to the job's time limit. If your training runs do that, checkpointing lets them continue in the next job instead of starting over; RCI can help set it up.
How Stony Brook compares
GPUs in the main research clusters at Stony Brook and at peer universities, from their public hardware pages (checked October 9, 2026). Models and ages differ, so this compares size, not speed.
Use the clusters
Any Stony Brook researcher can request an account on SeaWulf or NVwulf; classes can get accounts for students. See the High-Performance Computing pages for how to apply, the quick start guide, and Open OnDemand for a browser-based way in.
About these numbers
Source: Slurm job records exported October 8, 2026. GPU-hours and core-hours are what was allocated times run time, not measured activity.
Pool ceilings are inferred from the records (the most GPUs ever in use at once); installed counts are 48 H200 and 44 B40 GPUs on NVwulf and about 44 A100s on SeaWulf.
Cloud equivalent prices the same hours at AWS Linux on-demand list prices (p5en.48xlarge for H200, g7e.48xlarge for RTX PRO 6000, p4d.24xlarge for A100, hpc7a.96xlarge for CPU) and leaves out storage, networking and staff.
Peer counts come from each institution's public pages; Binghamton and Michigan do not publish totals.
Hover a chart for exact values, or use Show the numbers for a table. This page is refreshed monthly.
