We have the following requirements w.r.to NVIDIA GPU operator for OpenShift to get NVIDIA gpu working with sandboxed containers (Kata) on OpenShift

• kataManager supported for NVIDIA GPU operator for OpenShift
  • kataManager only creating the cdi spec for both SNP and TDX, without adding crio configs and runtime classes (since CRIO config and runtimeclassses are created by the OSC operator)
• “Sandbox device plugin” supported for NVIDIA GPU operator for OpenShift. It does the GPU allocation
• ccManager supported for NVIDIA GPU operator for OpenShift. It’s responsible for enabling CC mode

This is an example clusterPolicy we have used https://github.com/cclaudio/osc-baremetal-gpu/blob/main/gpu-operator/nvidia-clusterpolicy.json