The Casper cluster now has 64 new high-throughput computing (HTC) nodes, which are for batch jobs that require the use of one full node or less. As the first step in
migrating the Casper system’s scheduler from Slurm to the PBS Pro workload manager, the HTC nodes are accessible only through PBS Pro.
This new documentation provides details for preparing PBS scripts and submitting jobs on Casper:
Migrating Casper jobs from Slurm to PBS. Other Casper nodes will be transitioned to PBS Pro over the next few weeks. About half of the Casper nodes in use previously will move from Slurm to PBS on
Wednesday, March 24. Those will be:
- 7 CPU-only nodes
- 4 GP100 GPU nodes
- 2 4xV100 GPU nodes
- 3 8xV100 GPU nodes
The rest of the Casper CPU-only and GPU nodes will move from Slurm to PBS on
Wednesday, April 7. At that point, all Casper jobs will be submitted via PBS. Please
contact CISL if you have questions or need assistance.
