Pick your cluster's portal
Each cluster has its own OnDemand portal and its own apps. Sign in with your NetID, then approve the Duo request. You need an account on that cluster first.
SeaWulf
CPU · 18 appsGeneral-purpose computing: notebooks, R, MATLAB, chemistry and structural biology.
- Nodes
- 40-core Intel and 96-core AMD
- Sessions
- Up to 4 hours, or 3.5 days on extended queues
- Who
- Anyone with a SeaWulf account
NVwulf
GPU · 19 appsGPU work: AI and language models, cryo-EM, protein prediction and imaging.
- GPUs
- RTX PRO 6000 Blackwell, 96 GB each, up to 4
- Sessions
- Up to 8 hours, or 48 hours on the long queue
- Who
- NVwulf account needed
ClinWulf
PHI · 8 appsResearch with protected health information, on CPU or GPU, kept inside ClinWulf.
- Queues
- CPU and GPU, short, long and extra
- Sessions
- Up to 8 hours, or 10 days on extra queues
- Who
- IRB-approved studies in Health Sciences
- Access
- From within Health Sciences and Stony Brook University Hospital and Medical Center, or via Citrix Desktop or the Stony Brook Hospital & Medical Center VPN
No account yet? Request one. Not sure which cluster fits? Try the cluster picker.
How a session works
- Sign inOpen your cluster's portal and sign in with your NetID and Duo.
- Pick an appChoose it from Interactive Apps in the top menu, or from the dashboard.
- Fill in the formQueue, hours, cores, memory and, on NVwulf and ClinWulf, GPUs. Small requests start sooner.
- Connect, then DeleteWhen the card says Running, click connect. Click Delete on the card when you're done.
Closing the browser tab doesn't end a session. It keeps running, and you can reconnect from My Interactive Sessions until its hours run out.
Find an app
30 apps across the three portals, plus Job Composer for batch jobs. Search or filter, then open the guide or the cluster's app page.
Browse every app on each cluster: SeaWulf apps, NVwulf apps, ClinWulf apps.
Choosing resources
Every app runs as a Slurm job and counts against your allocation like any other job. Start with a short queue and 1 to 4 cores; ask for more only if your program uses them.
| Cluster | Quick sessions | Longer work | GPUs |
|---|---|---|---|
| SeaWulf | Short shared queues, up to 4 hours | Long up to 24 hours, extended up to 3.5 days | None. Use NVwulf for GPU work |
| NVwulf | b40x4 (Regular), up to 8 hours | b40x4-long, up to 48 hours | 1 is enough for most apps, up to 4 |
| ClinWulf | CPU or GPU short (GPU short up to 2 hours) | Long up to 8 hours, extra up to 10 days | GPU queues, 1 to 4 |
More detail in the queue selection guide.
Good habits
If something goes wrong
My session stays Queued
The queue is busy. Delete the session and try a shorter queue, fewer cores or GPUs, less memory or fewer hours. On NVwulf and ClinWulf, Clusters then System Status shows how busy the nodes are.
The card says Completed right after it started
The app didn't start, often because your home folder is full or the settings didn't fit the queue. Open the session's log from the card, and check your space with myquota.
The app is blank or keeps disconnecting
Refresh the page or reconnect from the session card. If the session has ended, launch a new one.
The portal itself is acting strangely
Use Restart Web Server in the Help menu at the top right. It restarts your own portal process and leaves your running jobs alone.
I can't sign in
Check your NetID password and Duo. Too many failed attempts can lock Duo; see Duo troubleshooting.
More answers in the OnDemand overview. Still stuck? Open a ticket with the cluster, the app and the time it happened.
Missed the October 1 training?
Arshad Mehmood and David Carlson walked through the dashboard, files and jobs, and live demos of RStudio on SeaWulf and AlphaFold on NVwulf. The slides are online.