Configure ClusterPolicy kernelModuleType for GPU Operator

💡 Quick Answer: Set driver.kernelModuleType: open to enable DMA-BUF and GPUDirect Storage. The auto default selects the recommended type based on driver branch and GPU model, but open guarantees compatibility with modern features.

The driver.kernelModuleType field in the ClusterPolicy controls which NVIDIA kernel module variant the GPU Operator builds and loads on each node.

Available Options

When to Use Each Option

• auto — Safe default for most clusters. Newer driver versions automatically select open.
• open — Required when you need DMA-BUF GPUDirect RDMA, GPUDirect Storage (GDS v2.17.5+), or want to ensure forward compatibility.
• proprietary — Only needed for legacy GPU architectures that do not support open modules.

Check Current Setting

oc get clusterpolicy gpu-cluster-policy -o jsonpath='{.spec.driver.kernelModuleType}'

Change the Setting

oc edit clusterpolicy gpu-cluster-policy

spec:
  driver:
    kernelModuleType: open

Restart driver pods to apply:

oc delete pod -n gpu-operator -l app=nvidia-driver-daemonset

Verify the Active Module Type

Check the driver container logs for the resolution:

oc logs -n gpu-operator ds/nvidia-driver-daemonset -c nvidia-driver-ctr | grep -i kernel

With auto, look for the line:

nvidia-installer --print-recommended-kernel-module-type
kernel_module_type=open

This confirms the Operator resolved auto → open for your hardware.

Verify on the Host

oc debug node/<node-name>
chroot /host
modinfo nvidia | grep license

Open kernel modules show Dual MIT/GPL licensing. Proprietary modules show NVIDIA only.

Feature Dependency Matrix

Why This Matters

Choosing the correct kernel module type determines which GPU features are available. Setting open unlocks DMA-BUF and GDS while maintaining full compute compatibility on Turing+ GPUs.