Nvidia GPU support in Confidential containers requires to have a podVM built with all the relevant bits in place.

Epic Goal

• Having a PodVM that can be used with Confidential GPU instance and use the HW from within the workload

Why is this important?

• This is required for AI/ML workload using CoCo (and peer-pods)

Scenarios

1. As an OpenShift administrator, I want to have a ready to use podVM image that can be configured with Confidential GPU instance
2. As an OpenShift user, I want to deploy my GPU workload in confidential instance and use it with my workload.
3. placeholder for remote attestation support

Acceptance Criteria 

(The Epic is complete when...)

1. RHEL 10 based PodVM works with Confidential GPU instance (e.g. in Azure- Standard_NCC40ads_H100_v5), workload can be used from within it (running nvidia-smi show it's in confidential mode)
2. The creation of the podVM image is embedded in the pipeline.
3. placeholder for using in BM also

Additional information

1. Driver has to be from version 570.172.08 or above
2. Base RHEL10 podvm to support the 6.9+ kernel requirement (as discussed)