Running Interactive GPU Job in SLURM

Hello all - I am adjusting to the new SLURM situation. Does anyone have any ideas about how to request a GPU in an interactive job? Running an interactive job is straightforward:

srun --pty /bin/bash,

and here I am told:

GPUs can be requested in both Batch and Interactive jobs.

So, I know it’s possible, I just don’t know how to request the GPU from the interactive job. Any help would be much appreciated!

Hi Landry, we are still getting used to the new scheduler ourselves, but does this article answer your question?

1 Like

Hi Dr. Finn - yes, thank you for the link! Maybe I’m impatient, but after running

sinfo -O gres -p gpuq,

I can see what is available, but when I request the one of the resources that is listed, I receive a message saying that the requested partition configuration not available. The shell then remains unchanged while displaying job queued and waiting for resources. I guess my confusion is as to why the resources are listed as available when they actually aren’t.

I was able to figure it out after having ITC help me out. The command that I ran to allow me run an interactive job on a GPU was:

srun -p debug --gres=gpu:1 --time=02:00:00 --pty /bin/bash

So my understanding of this is that I am submitting an interactive just by choosing the debug partition, choosing one gpu (without specifying which specific gpu), choosing my walltime of two hours, and finally --pty /bin/bash to interact with the terminal (thanks @paxton for the details on what each of these flags mean).

1 Like