NPU Operator tutorials¶
These tutorials describe step-by-step scenarios for using NPUs in a Kubernetes environment.
To install the NPU Operator, see its installation guide for Kubernetes or OpenShift.
| Tutorial | What you will do |
|---|---|
| Run an NPU Pod | Request one NPU and verify it inside a Pod |
| Schedule a Pod by NPU family | Inspect NPU information and schedule a Pod on a node with the selected family |
| Allocate two NPUs to one Pod | Allocate two NPUs to a container and verify that they share an RSD group |
| Request NPUs with DRA | Request one NPU with a Claim and verify it inside a Pod |
| Specify NPU allocation conditions with DRA | Set allocation conditions by count, product name, PCIe topology, or UUID and verify the result |
| Verify RDS devices in an NPU Pod | Verify NPU and RDS devices inside a Pod on a Rebel100™ node |
| Upgrade NPU drivers automatically | Apply a target driver version and observe its automatic rollout |
| Run different driver versions on separate nodes | Apply different driver versions across families or nodes of the same family and verify the instances |
| Attach an NPU to a KubeVirt VM | Attach a VFIO NPU resource and verify it inside the guest |
| Separate NPU nodes for containers and VMs | Separate container and VM passthrough nodes and verify their resources |
| Collect NPU metrics with Prometheus | Configure scraping and query an NPU metric in Prometheus |