Adding a GPU to a host VM manually¶

It is possible to assign a GPU to your virtual machine by editing the VM to add a hostDevices section to the VM's YAML configuration:

kubectl -n arc-gpu-test-ns get vm
NAME             AGE   STATUS    READY
gpu-test-iac-0   32d   Running   True
gpu-test-iac-1   32d   Running   True

kubectl -n arc-gpu-test-ns edit vm gpu-test-iac-0

Add the following to the devices section of the YAML configuration (at the level of the other devices in this section):

devices:
  disks:
  - bootOrder: 1
    disk:
      bus: virtio
    name: disk-0
  hostDevices:
  - deviceName: nvidia.com/GA100_A100_PCIE_80GB # Type of your device (via pcidevice)
    name: sl-g02-08-000031000                   # Name of your device
  interfaces:
  - bridge: {}

In order to get the deviceName above, you can use the pcidevice resource:

kubectl get pcidevice sl-g02-08-000031000 -o jsonpath='{.status.resourceName}'
nvidia.com/GA100_A100_PCIE_80GB

Once this is saved, you will need to restart the VMI for this VM so that these changes can be picked up. If your VMI was not running on the same node that the GPU is attached to, it will be migrated automatically and restarted.