Nvidia GPU support
You can assign the graphical processing unit (GPU) resources within your IBM® Cloud Private cluster to applications and jobs.
GPU is the processing power behind new workloads that are paving the way in fields such as machine learning and high performance computing systems. Starting with Kubernetes 1.6.1, you can now manage GPU in a similar way to that of other resources such as CPU and memory.
The following restrictions apply to GPU use in your IBM Cloud Private cluster.
- You must represent GPU resources with positive integer values that indicate the number of GPU physical cores. Partial GPU core allocation is not supported.
- You must use GPU driver version 352 or later. Older versions might not be compatible with IBM Cloud Private.
- You can assign GPU resources to specific containers in a Pod. You cannot share the GPU resources with other containers in the Pod.
- You can declare only GPU resource limits, not requests.
For more information about deploying an application with attached GPU resources, see Creating a deployment with attached GPU resources.