SLURM Features

From Wiki
Revision as of 12:47, 24 February 2024 by Administrator (talk | contribs) (Created page with "As the cluster is heterogeneous, we employ "node features", i.e. tags which describe the capabilities of each node and can be employed at job submission to request appropriate nodes. A list of the currently supported features is as follows: * gpu_P100_16G * gpu_1080_8G * gpu_2080Ti_11G * gpu_RTX5000_16G * gpu_RTX6000_24G * gpu_RTXA5000_24G * gpu_A40_48G where both the exact GPU model name and the available VRAM are reported in the name of each node feature. These can...")
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

As the cluster is heterogeneous, we employ "node features", i.e. tags which describe the capabilities of each node and can be employed at job submission to request appropriate nodes.

A list of the currently supported features is as follows:

  • gpu_P100_16G
  • gpu_1080_8G
  • gpu_2080Ti_11G
  • gpu_RTX5000_16G
  • gpu_RTX6000_24G
  • gpu_RTXA5000_24G
  • gpu_A40_48G

where both the exact GPU model name and the available VRAM are reported in the name of each node feature. These can be employed at job submission with the "--constraint" directive, which supports expressing a boolean combination of features. For instance, one might request a node with 24 GB of VRAM using the following expression:

#SBATCH --constraint="gpu_RTX6000_24G|gpu_RTXA5000_24G"