We have a heterogenous cluster with 3 nodes that have GPUs in them. Two of those nodes have 2 cards each and the third node has 3 cards. They are all setup with proper GRES definitions and all work individually. We have a user that would like to utilize all of the resources from a single job, but if we ask for 3 cards and 3 nodes Slurm tells us we don't have the resources and if we ask for 2 cards and 3 nodes then we're leaving one of the card un-utilized. Is there a way to tell Slurm that we want all 7 cards allocated and used?
Thanks, Mike Robbert
smime.p7s
Description: S/MIME cryptographic signature
