How to use Charmed Kubeflow with NVIDIA GPUs

If you want to use Kubeflow backed with NVIDIA GPUs, your nodes must have runtimeClassName: nvidia parameter specified.

Contents:

MicroK8s

In case of microk8s deployment, this parameter is added by default as you enable the gpu addon.

microk8s enable gpu

Other Kubernetes clusters

In case of other kubernetes deployment this parameter may not be specified. Please refer to this documentation in that case with guide on how to setup NVIDIA GPU Operator.