Slurm on Carina

Carina has three Slurm queues. Choose your queue based on how much time you need. All three queues run on the same hardware.

Queues and Partitions are the same thing and the terms are used interchangeably.

The dev partition is for active development; there is a two-hour time limit, which keeps the resources available and wait times short.

The normal partition is for running your jobs. The time limit is two days. Your jobs may sit in the queue waiting for resources.

Jobs in long need more than two days, and can run up to five days.

GPUs can be requested on any of the partitions. For example, including this in an sbatch script will request 1 gpu.

#SBATCH --gres=gpu:1

To understand these commands, read the Slurm Primer

This will give you a compute node in the dev partition with one core for one hour:

srun --pty bash

This command requests 2 GPUs on a compute node in the dev partition:

srun --pty -p dev --gres=gpu:2 bash

See your current activity:

squeue -u <sunetid>

Stop all of your jobs

scancel -u <sunetid>