Skip to content

Add support for GPUs in Slurm

André Breda requested to merge ist189409/nixrnl:labs-nvidia-slurm into 6-labs-add-hosts

Description of changes

Slurm currently does not detect or track GPU usage. Combined with ConstrainDevices=yes, I believe this makes it impossible to utilize GPUs within Slurm, and the reason for jobs started with --exclusive to not be currently scheduled in these nodes.

This MR changes Slurm configuration to recognize GPUs as a type of resource, and to autodetect them in machines with NVIDIA GPUs. It may be interesting to further investigate this later to also detect the Intel iGPU.

Things done

  • Tested
  • Updated documentation (Wiki/NetBox)
  • Breaking change

Merge request reports