GPU job submissions

With the recent update to the Discovery cluster, it has changed how you request GPU resources.

To submit to the Discovery cluster GPU nodes, you must specify the number of nodes, cores, and the number of GPU's you plan to use.

Example: To request a 4 core, and 2 GPU compute job. Here is how you would set it up inside your .pbs script:

#PBS -l nodes=1:ppn=4:gpus=2
#PBS -l feature=gpu

To submit for an interactive job on the GPU nodes for the same resources use:

mksub -I -l nodes=1:ppn=4:gpus=2 -l feature=gpu

If you have any questions, please feel free to reach out to research.computing@dartmouth.edu

Details

Article ID: 123557
Created
Mon 1/4/21 10:45 AM
Modified
Tue 1/19/21 12:01 AM