Licensing

Provides insights that can help you identify and resolve potential challenges with HPE Ezmeral Unified Analytics Software licensing.

GPU Workloads Not Running

HPE Ezmeral Unified Analytics Software will not run GPU workloads unless there is a valid GPU license applied. Currently, there is no specific alert for this issue in the the UI; however, you can use kubectl commands to troubleshoot licensing issues.
Viewing GPU licensing labels
Run the following command to get the GPU-related licensing labels:
(reverse-i-search)'jq': kubectl get node <node-ip-address> -o json | jq '.metadata.labels'
The following table lists the GPU-related licensing labels to identify in the output:
Label Description
nvidia.com/gpu.count This label provides the GPU count exposed by NVIDIA. This is the number of virtual GPU devices. For example, if you have your A100 split into two MIG devices, they are reflected in the GPU count. The number reflects the MIG partitions.
hpe.com@.nvidia.com/gpucount HPE Ezmeral Unified Analytics Software introduced this label. The label provides the actual physical GPU count on that node.
Viewing the number of GPUs in use
Run the following command to view the number of GPUs currently in use. You can also see which applications and notebooks are running when you run this command:
kubectl describe <node-ip-address>

The output shows the number of GPUs currently in use. You can also see which applications and notebooks are running on the node.