Looks like we need a device plugin to expose GPU classes to containers. It's much easier if we automatically add the device plugin when RKM detects a NVIDIA or AMD GPU is present on a Windows node.
The k8s-directx-device-plugin implementation is fairly trivial, though not maintained so we will probably have to fork it.
Looks like we need a device plugin to expose GPU classes to containers. It's much easier if we automatically add the device plugin when RKM detects a NVIDIA or AMD GPU is present on a Windows node.
The k8s-directx-device-plugin implementation is fairly trivial, though not maintained so we will probably have to fork it.