Using GPU Passthrough

Using GPU Passthrough in Virtual Machines

Now that you've installed the NVIDIA GPU operator, and your nodes have been configured for GPU passthrough and VFIO, we can attach our GPU to a KubeVirt VM.

Info

If you haven't installed the GPU operator, or configured your nodes for GPU passthrough you can see our guide here

Attach GPU to a KubeVirt VM Workload

1. Navigate to Workloads in Juno
2. Select your VM template
3. In the GPU section, you'll see a True/False toggle
4. Enable GPU passthrough
5. Enter the exact device name from the allocatable resources:
6. Example: nvidia.com/TU104_GEFORCE_RTX_2080

GPU allocation rules

• Currently: Each VM can request exactly one GPU
• Multiple GPUs: Multiple GPU passthrough per VM is coming soon (TBD)
• Device name: Must match the exact string from kubectl get nodes output