Giant Swarm offers a GPU Operator app which can be installed in workload clusters when using GPU based instance types.
Here, we define the GPU Operator upstream version with additional configuration and network policies.
Installation can be done via Happa or by running helm install --namespace kube-system gpu-operator helm/gpu-operator.
Toolkit and Driver must kept disabled in the values.yaml file. Flatcar already deals with the driver and the toolkit.
This app is based on the images and manifests provided by NVIDIA in their GPU Operator project.