Outcome Overview

Enable Intel's Infrastructure Processing Unit (IPU, Intel's class of SmartNICs/DPUs) support within OpenShift to offload networking and infrastructure functions from the host CPU, enhancing performance, reducing latency, and improving workload isolation. This outcome ensures OpenShift can recognize and manage Intel IPUs across the cluster, including integration with the node lifecycle, networking stack (e.g., SR-IOV, OVN-K), and MachineConfig workflows. It includes enabling IPU-aware scheduling, validating end-to-end functionality, and providing documentation and supportability guidance for customers deploying IPUs in OpenShift environments.

Success Criteria

• OpenShift nodes with Intel IPUs are automatically recognized and labeled correctly during provisioning.

• Intel IPU firmware and drivers are loaded successfully on RHCOS without manual intervention.

• SR-IOV workloads are able to offload traffic via Intel IPUs without packet loss or degradation.

• Nodes with Intel IPUs are schedulable using standard Kubernetes labels and taints.

• IPU-hosted workloads are observable via OpenShift Console and Prometheus/Grafana metrics.

• Networking functions (e.g., pod-to-pod, ingress/egress) are validated to work with IPU offload in hybrid clusters.

• Full documentation is published covering prerequisites, setup, supported use cases, and limitations.

• CI pipelines validate IPU functionality across at least one supported OpenShift version.

• Red Hat support teams are equipped with troubleshooting documentation and escalation procedures for IPU issues.

Expected Results (what, how, when)

Measurable OpenShift customer production use of Intel IPUs via telemetry and/or Insights, starting with the conclusion of the first IPU hardware enabled as a result of this effort. 

Post Completion Review – Actual Results

TBD