NVIDIA Support

In order to consume GPU resources on the cluster, the NVIDIA GPU support must be deployed. NVIDIA support requires that all nodes in the cluster are entitled to pull RHEL content. In order to set this up, the bastion node must be running RHEL. These entitlements will be copied to the cluster.

To setup NVIDIA Support:

farosctl install-plan nvidia-support

This install plan is the equivalent to running all of the following commands.

# deploy red hat entitlements to the cluster
farosctl deploy redhat-entitlements

# install nvidia drivers on the cluster nodes
farosctl deploy nvidia-drivers

After the install is completed, it will take quite some time for all of the changes to be applied. Each node will need to have their configuration updated and be restarted. Watch the cluster console and wait for the operators to finish.

To ensure the GPUs are available, list all the resources on each node.

oc get nodes -o jsonpath="{range .items[*]} {@.metadata.name} {@.status.capacity} {'\n'}"

The nodes with GPU’s should include a resources called nvidia.com/gpu.