Sets up mig on a k3s cluster (using forked images by default), lots of this work heavily used a CERN blog

helm install -f ~/gpu-operator-mig-values.yaml --wait --generate-name \
    -n gpu-operator --create-namespace \
    nvidia/gpu-operator

kubectl apply -f ./configs/nvidia-mig-config.yaml

now label nodes to apply mig presets, e.g

kubectl label node <NAME> nvidia.com/mig.config=1g.10gb

If working, the node should have mig gpus listed under Allocatable (note pods will have to allocate mig gpus, rather than just nvidia.com/gpu, using strategy: single will use MIG across all GPUs and allow for scheduling as nvidia.com/gpu)