Map gpu resources in pbs

Hello everybody!

I am new to the forum and not really a system engineer, so forgive me if I am asking something silly/not precise.

I have a multi-nodes cluster with some gpus nodes in it. Is it possible to map gpu resoruces with PBS?

Ideally, I would like to send jobs assign max gpu memory, with pbs swaping node/gpu on the node if it saturates. Do you think this thing possible?

Thank you very much for the help!

1 Like

Hi @alie89

Please follow this discussion GPU Access Limited by CGroup - #6 by Phhere
In a nutshell - you would need cgroups , also with the help of it you can map the topology to a bit.
I am not sure whether there are controlling features available related to gpu memory , gpu cores etc.