Requesting GPU Models

There are currently four nodes appointed for gpu-accelerated programming available for researchers to use on Monsoon. These nodes all contain several GPUs each of varying models ranging from the K80s, P100s, A100s, and V100s)

When running a job, users are able to request specific GPU models that their jobs will utilize. To specify what model a user would like to use while using srun, the -C flag can be used.

[abc123@wind ~]$ srun -C a100 -G 2 --pty /bin/bash

• The -C or –constraint flag is used to specify constraints for the resources required by the job. These constraints can include a specific node name or a node count
• The -G or –gpus flag is used to specify the number of GPUs that is required for the job. Users are able to specify up to four GPUs with the p100, a100, and v100 models and eight GPUs with the k80 model.

NOTE: It is recommended to request only the number of GPUs that align with the program’s actual needs.

Users are also able to submit an sbatch job with a specific constraint that designates the GPU model they would like to use also specify how many GPUs they would like for their job.

#SBATCH --job-name=NAMESCRIPT
#SBATCH --output=/scratch/NAUID/NAME.txt
#SBATCH --constraint=a100
#SBATCH --gpus=2
#SBATCH --chdir=/scratch/NAUID/DESIREDPATH
#SBATCH --time=3:00:00
#SBATCH --mem=1000

NOTE: There are different models of GPUs available, each with its own set of features and capabilities. Newer GPU models often offer improved performance and speed compared to older ones. For example, The A100 is a newer and more powerful GPU model that may fit more research needs then using the k80 model which is an older and comparatively slower model. It is important to assess your specific needs, requirements, and budget to determine which GPU model is the best fit for your use case.