NVIDIA container on Kubernetes

Create a RuntimeClass for nvidia

kubectl apply -f nvidia-runtimeclass.yaml

Deploy the NVIDIA Device Plugin (v0.17.1) as a DaemonSet

kubectl apply -f nvidia-device-plugin.yml

verify

kubectl describe node $(hostname) | sed -n '/Capacity:/,/Allocatable:/p'

should see

  nvidia.com/gpu: 1

run test

kubectl apply -f gpu-test.yaml.yml

kubectl apply -f nvidia-runtimeclass.yaml

kubectl apply -f nvidia-device-plugin.yml

kubectl apply -f gpu-test.yaml.yml