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