{"id":3680,"date":"2024-10-16T12:07:29","date_gmt":"2024-10-16T19:07:29","guid":{"rendered":"https:\/\/in.nau.edu\/arc\/?page_id=3680"},"modified":"2025-01-03T15:10:32","modified_gmt":"2025-01-03T22:10:32","slug":"gpus","status":"publish","type":"page","link":"https:\/\/in.nau.edu\/arc\/gpus\/","title":{"rendered":"GPUs on Monsoon"},"content":{"rendered":"\n
\n
<\/p>\n

\n

\n
\n

Contact Advanced Research Computing<\/h3>\n <\/div>\n
\n \n
\n <\/i>\n Email:<\/span>\n <\/div>\n
ask-arc​@nau.edu<\/div>\n <\/a>\n <\/div>\n<\/div>\n\n<\/p>\n

\n

\n View Microsoft Teams Meeting Information<\/a>\n <\/div>\n<\/p>\n

<\/div>\n<\/div>\n\n

GPUs on Monsoon<\/h1>\n

For some of your jobs on Monsoon, you might require the use of one or more Graphics Processing Units (GPUs) in order to accelerate your jobs. By using Slurm on Monsoon, you can easily request a GPU for your job.<\/p>\n

\n

Note: At the time of writing this article, Monsoon only contains NVIDIA GPUs.<\/em><\/p>\n

Checking Available GPUs<\/h2>\n

To get a list of all GPUs on Monsoon along with what GPUs are available for use, you can use the gpu_status<\/span> command:<\/p>\n

$ gpu_status\r\nAvailable GPUs: 10<\/span>\/24\r\nk80: 7<\/span>\/12\r\np100: 0<\/span>\/4\r\nv100: 3<\/span>\/4\r\na100: 0<\/span>\/4\r\nPending GPU jobs: 57<\/span>\r\nRunning GPU jobs: 5<\/span>\r\n<\/code><\/pre>\n
Note: If you are running deep learning software, we recommend using the a100<\/span> GPUs. If those are unavailable, try using either the p100<\/span> or the v100<\/span> GPUs instead.<\/em><\/p>\n
Note: Some GPUs may be shown as available, but cannot be requested. This is because some nodes are dedicated to specific research groups.<\/em><\/p>\n
Submitting a GPU Job<\/h2>\n
To quickly request a GPU of any model, you can use either the -G<\/span>\u00a0or \u2010\u2010gpus=<\/span> flags in either your srun<\/span> or salloc<\/span> commands:<\/p>\n
# Both of these do the same thing!<\/span>\r\n$ srun -G 1<\/span> nvidia-smi\r\n$ srun --gpus=<\/span>1<\/span> nvidia-smi\r\n<\/code><\/pre>\n
$ salloc --gpus=<\/span>1<\/span>\r\n$ srun nvidia-smi\r\n<\/code><\/pre>\n
Or, if you want to add it to your SBATCH script, you can add the following line where the rest of your SBATCH parameters are:<\/p>\n
#SBATCH --gpus=1<\/span>\r\nnvidia-smi\r\n<\/code><\/pre>\n
If you want to request a specific GPU model, you can specify that in the same argument:<\/p>\n
$ srun -G k80 nvidia-smi\r\n$ srun -G k80:4 nvidia-smi    # requests four k80 GPUs<\/span>\r\n<\/code><\/pre>\n
#SBATCH --gpus=a100<\/span>\r\nnvidia-smi                    # the program you want to run\r\n<\/code><\/pre>\n
Another way to request a specific GPU model would be to use the -C<\/span> flag:<\/p>\n
$ srun -C k80 -G 4 nvidia-smi<\/code><\/pre>\n
Relevant Flags<\/h3>\n
There are additional flags you can provide Slurm to fine-tune your jobs, but they may not be useful for everyone.<\/p>\n