YAAD GPU LAB8 DGX Sparks for your team and your agents
Sign in with your @uwm.edu email. Then start or join a team to get a private workspace on a DGX Spark (root, GPU, Docker), or send batch jobs to any Spark with room. Connect the CSI VPN (FortiClient) before you SSH in. New here? Read the student guide.
Your team workspace
Compute pool
Ask for what you need. YAAD runs your job on any DGX Spark with room (your team's Spark first), runs teams in parallel, and queues jobs until capacity frees up. Results land in /workspace/results/<job>/.
Agent access
Give your own agent (Claude Code or any MCP client) a personal token. It can then see capacity, submit and watch jobs, and fetch results for your team, as you. Tokens expire after 14 days. Revoke one any time.
Help
Stuck? Tell the CSI host what happened. Add screenshots, a screen recording or logs. Tip: run yaad-diag > /workspace/diag.txt in your workspace and attach it (download it with scp first). Your team, Spark and recent jobs are attached automatically.
gpu.smoke.v1 — DGX GPU Smoke Test
Runs a fixed, pre-approved container on a DGX Spark that reports GPU visibility (nvidia-smi). No inputs, no code upload.
operator approval≈1–5 s runtimeno charge
Your requests
| Requested | Status | Result | |
|---|---|---|---|
| Sign in to see your requests. | |||