Allocate device memory to container

Allocate a certain size of device memory by specifying resources such as nvidia.com/gpumem. This resource is optional. With the default memoryFactor: 1, each unit of nvidia.com/gpumem equals 1 MiB.

Use a non-negative integer without suffixes such as Mi, Gi, or G. For example, request 16384 for 16 GiB with the default memory factor. Kubernetes quantity suffixes do not express HAMi memory units.

resources:
  limits:
    nvidia.com/gpu: 1 # requesting 1 GPU
    nvidia.com/gpumem: 3000 # Each GPU contains 3000 MiB device memory

Allocate a percentage of device memory by specifying resource nvidia.com/gpumem-percentage. Optional, each unit of nvidia.com/gpumem-percentage equals 1% of device memory. Use a plain integer such as 50 for 50%, without a % suffix.

resources:
  limits:
    nvidia.com/gpu: 1 # requesting 1 GPU
    nvidia.com/gpumem-percentage: 50 # Each GPU contains 50% device memory

note

Prefer specifying memory using one of these resources. If both are set and nvidia.com/gpumem is a positive integer, it takes precedence when determining the memory request. If nvidia.com/gpumem is omitted or 0, a nvidia.com/gpumem-percentage value from 1 to 100 is used to calculate the memory request.

If nvidia.com/gpumem-percentage is 100 and nvidia.com/gpucores is not specified, HAMi automatically sets nvidia.com/gpucores: 100, even when nvidia.com/gpumem is also set to a positive integer.

nvidia.com/gpumem-percentage must be between 0 and 100, inclusive; 0 is treated as unset. Requests with values outside this range are rejected, even when nvidia.com/gpumem is also set to a positive integer.