EnterpriseNVIDIAAMD

Red Hat OpenShift

Red Hat

Enterprise Kubernetes with a curated operator model. GPUs are enabled via the Node Feature Discovery and NVIDIA GPU Operators from OperatorHub; AMD provides an AMD GPU Operator.

Bootstrap with Red Hat OpenShift
Enablement

NFD + NVIDIA GPU Operator (OperatorHub) / AMD GPU Operator

Driver install

GPU Operator builds/loads drivers against the RHCOS kernel via the Driver Toolkit (entitlement-free on current releases).

Container runtime

CRI-O with the NVIDIA container runtime.

Scheduling

Default scheduler; MachineSets for GPU nodes; NFD labels drive placement.

Partitioning

MIG (single and mixed strategies) and time-slicing via the operator's ClusterPolicy.

Networking

OVN-Kubernetes; SR-IOV and RDMA via the SR-IOV Network Operator for training fabrics.

Upgrades

Operator-managed; the GPU Operator rebuilds drivers on cluster upgrades.

GPU enablement
  1. 1

    Install Node Feature Discovery

    From OperatorHub; labels nodes with PCI/GPU features.

  2. 2

    Install the NVIDIA GPU Operator

    From OperatorHub, then create a ClusterPolicy.

    bash
    oc get csv -n nvidia-gpu-operator
oc apply -f clusterpolicy.yaml
  3. 3

    AMD path

    Install the AMD GPU Operator for Instinct nodes instead of the NVIDIA operator.

Validate the enablement
verify GPUs
oc get nodes -o json | jq '.items[].status.allocatable' | grep -i gpu
oc -n nvidia-gpu-operator get pods
oc debug node/<gpu-node> -- chroot /host nvidia-smi
Known limitations
  • ·RHCOS immutability means drivers must go through the operator, not manual installs.
  • ·SR-IOV/RDMA adds its own operator and node config.
Best for

Regulated enterprises wanting a supported, operator-driven platform with strong policy controls.