Submitting GPU jobs
The HPC supports GPU jobs. The HPC includes an increasing number of compute nodes equipped with GPU hardware.
GPU resources available on the HPC#
GPU resources on the HPC are available in certain access Slurm accounts (partitions):
|NVIDIA GeForce GTX 1080 Ti
backfill2 (4 hour time limit)
|NVIDIA RTX A4000
|Owner accounts only
|NVIDIA H100 Tensor Core GPU
|Limited access for pilot programs; Request access
In addition, if your department, lab, or group has purchased GPU resources, they will be available on your owner-based Slurm account.
Free GPU jobs longer than four hours#
If you want to run jobs that are longer than four hours in our general access queues, we are accepting requests on a case-by-case basis. Contact us to request access.
If your research group has purchased dedicated GPU resources on our system, then there is no need to request access. If you submit your job to your owner-based queue, you will be able to run GPU jobs for longer than four hours.
Submitting GPU Jobs#
Via Open OnDemand#
You can submit graphical jobs via Open OnDemand to run on GPU nodes. Note that this may increase your job wait time in the queue. Follow the instructions for submitting a job, and specify 1 or more GPUs in the "GPUs" field:
Some interactive apps may not have this field. If you encounter this, please let us know.
Via the CLI#
If you have not yet submitted a job to the HPC cluster, please read our tutorial first.
If you wish to submit a job to node(s) that have GPUs, add the following line to your submit script:
Nodes contain two to four GPU cards. Specify the number of GPU cards per node you wish to use after the
For example, if your job requires four GPU cards, specify
Full Example Submit Script#
The following HPC job will run on a GPU node and print information about the available GPU cards:
Your job output should look something like this:
For more information and examples, refer to our CUDA software documentation