Historique de la page

Now that you have a handle on basic and more advanced jobs, it's time to get even more advanced. Some researchers require GPUs for their work, and luckily, that is really easy to do with Slurm.

The SPORC cluster currently has:

• 100 A100 cards
• 16 Nvidia V100 cards
• 48 Nvidia P4 cards

CUDA is available through either Spack or module if it is needed.

What's Available?

If you are using GPUs through SPORC, you will have up to 4 A100, 4 V100 code or 2 P4 cards per node.

Formatting the Command

The general form for --gres is as follows:

--gres=gpu:type:count

where type is the name of the kind of GPU you want and count is the number of GPUs of that type you want to request. For example:

--gres=gpu:v100:1

With this command, your job will run on one V100 GPU. The first available GPU will be used if you do not specify a type . Everything else about the SBATCH file is the same

If you want either a v100 or a100 you can use a constraint:  --gres=gpu --constraint="vram40|vram32"This wiki page is deprecated. You can find this documentation on our new documentation site: https://research-computing.git-pages.rit.edu/docs/using_gpus_with_gres.html