Allocate device core to container

Allocate a percentage of device core resources by specifying resource nvidia.com/gpucores. Optional, each unit of nvidia.com/gpucores equals 1% of device cores. Use a plain integer such as 50 for 50%. Do not use 50% or Kubernetes CPU quantities such as 500m.

resources:
  limits:
    nvidia.com/gpu: 1 # requesting 1 GPU
    nvidia.com/gpucores: 50 # Each GPU allocates 50% device cores.

note

HAMi-core uses time-slice to limit device core usage. Therefore, there will be fluctuations when looking at GPU utilization through nvidia-smi