How can I make a pod use a specific GPU from a k8s node?

[FLM210]

How can I configure the pod to use a specific model of GPU from the two different GPU models available on my node?

[FLM210]

Ex: my machine has both RTX 3080 and RTX 3080 Ti. I want the pod to use the RTX 3080 Ti.

[evopen]

maybe https://gitlab.com/nvidia/kubernetes/device-plugin/-/merge_requests/127 will help

[shivamerla]

@FLM210 mixed GPU types are currently not supported by the device-plugin and gpu-feature-discovery on the same node. Above linked PR will allow resource-renaming of GPUs, but that is not officially supported (disabled in the code). Those features will be available with the upcoming Dynamic Resource Allocation plugin later this year.

[FLM210]

@shivamerla Okay ~~~

[PotapIlya]

@shivamerla
hi, is it now supported?

i installed sandboxDevicePlugin and it works for vm (nvidia.com/GH100_H100L_94GB)

but i cant get the same result on default devicePlugin