[Feature]: time-slicing GPUs

[aviallon]

Suggestion Description

For some workloads, it can be very beneficial to over-subscribe GPUs. The nvidia device plugin allows being configured to time-slice GPUs¹, meaning a single GPU effectively acts as several resources.
I would like to have such functionality for ROCm's device plugin.

Operating System

No response

GPU

No response

ROCm Component

No response

Footnotes

1. https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/gpu-sharing.html# ↩