This is installed by the NVIDIA AMD operators, but it's a common Kubernetes project, so we should manage this separately which allows centralized configuration, especially to do with scheduling tolerations so we can schedule it on a control-plane node and allow scale from zero.
Additional benefit: in a mixed vendor cluster these two aren't fighting over management in order to determine where to run the device drivers based on hardware.
This is installed by the NVIDIA AMD operators, but it's a common Kubernetes project, so we should manage this separately which allows centralized configuration, especially to do with scheduling tolerations so we can schedule it on a control-plane node and allow scale from zero.