Keuka Cluster
Keuka cluster offers 2 Nvidia A100 GPUs and 2 Nvidia H100 GPUs in an HPC cluster environment. This cluster is currently in early user testing by invitations only. If you are interested in participate in testing, email CAC Help.
Hardware
| Node Name | CPU Cores | RAM (GB) | GPU |
|---|---|---|---|
| keuka.cac.cornell.edu | 8 | 64 | N/A |
| c0001 | 64 | 471 | 2 x H100 |
| c0002 | 32 | 113 | 2 x A100 |
Home Directories
- Path: ~
- User home directories are hosted in a CephFS share. Data in user home directories are NOT backed up.
- Access via the Keuka Cluster Globus collection.
Scheduler
- Slurm Quick Start
- Slurm documentation
- All jobs must specify which project to charge against using the
-Aoption. For example,-A shm7_0003. You can only specify the project of which you are a user. - There is only 1 partition/queue:
cloud - Hyperthreading is not enabled in this cluster.
Request GPUs
Use the --gres:<GPU Type>:<Number of GPUs> option in your sbatch, srun, or job script to request GPUs. For example, the srun --pty -c 4 --gres=gpu:a100:1 -A shm7_0003 /bin/bash command will start an interactive session of 4 CPU cores and 1 Nvidia A100 GPU.