Configuring GPU usage

CML supports heterogeneous GPU clusters’ configurations with homogenous GPUs within the nodes.

See the following example on a cluster with two nodes dedicated for GPU workloads:

  • GPU Node A - 4xnVidia A10G
  • GPU Node B - 2xnvidida A100
Set a default GPU accelerator label at: Site Administration > Runtime > Workload Accelerators.
The default GPU accelerator label is suggested in your workload selection screen.
For backwards compatibility, the default GPU accelerator label is used for those APIv2 requests that do not specify a GPU accelerator label but specify the number of GPUs for the workload.
You can limit the number of GPUs that can be used by a single workload (or a single session, single job and so on).