Problem you have encountered:

I found 2 docs about nvidia GPU driver on auto-provisioned nodes:

• this one does not mention what happens is the label is omitted. https://cloud.google.com/kubernetes-engine/docs/how-to/gpus#installing_node_auto_provisioning
• this one does mention that the value "default" is the default when label is omitted. But from the tests i performed, this is wrong. On an auto-provisioned GPU node without the label and on control-plane 1.31.4-gke.1372000, the driver wasn't installed. https://cloud.google.com/kubernetes-engine/docs/how-to/node-auto-provisioning#choose-driver-version

What you expected to happen:

• Install the nvidia driver "default" on auto-provisioned GPU nodes
• Update the first link docs to match the second link docs

Steps to reproduce:

• create an 1.31.4-gke.1372000 GKE cluster
• setup auto-provisioner to auto provision GPU nodes (t4 for example)
• spawn a pod requesting the gpu
• wait and see that driver isn't installed an device plugin never add the device to the node because of that.