OpenShift Virtualization 4.22 earns NVIDIA’s HGX hypervisor certification

NVIDIA now lists Red Hat OpenShift Container Platform 4.22 with OpenShift Virtualization 4.22 as a certified hypervisor platform for Hopper SXM systems. The accessible NVIDIA certification record replaces the Red Hat announcement that later disappeared, and it narrows the claim to a specific platform version, hardware track and validation scope.

What the certification covers

NVIDIA’s x86 certification track is scoped to NVSwitch-based, eight-GPU Hopper SXM systems, covering NVIDIA HGX and DGX platforms. It requires two phases: a single-node test with one eight-GPU virtual machine using full GPU and NIC passthrough, followed by a multi-node test with two eight-GPU virtual machines on two identical Hopper nodes.

The certification table names the covered platform precisely: Red Hat OpenShift Container Platform 4.22 with OpenShift Virtualization 4.22. It does not state that all OpenShift versions, GPU models or virtualization configurations inherit the result. NVIDIA also directs readers to its AI Enterprise Infrastructure Support Matrix for currently supported component versions.

That specificity matters for platform teams. The listing validates a demanding passthrough topology, but it is not a blanket performance guarantee for every virtualized AI workload.

How GPU workloads fit into OpenShift Virtualization

NVIDIA’s OpenShift documentation describes OpenShift Virtualization as KubeVirt-based VM management inside Kubernetes. It allows virtual machines and containers to coexist in one cluster, while NVIDIA’s GPU Operator can provision worker nodes for either GPU-accelerated containers or GPU-accelerated virtual machines.

Those modes are not interchangeable on a single worker. NVIDIA documents three node workload labels—container, vm-passthrough and vm-vgpu—and says one worker runs one of those GPU workload types, not a mixture. The required software also differs: containers use the data-center driver and container toolkit; passthrough VMs use vfio-pci; and virtual-GPU deployments require the NVIDIA vGPU Manager.

For passthrough and vGPU, administrators can choose GPU Operator-managed components or OpenShift Virtualization’s native procedures. NVIDIA warns that GPU Operator operands must be disabled on nodes that use OpenShift Virtualization’s native PCI passthrough path to avoid conflicts.

What platform teams should test

Organizations evaluating the certified configuration should first confirm that their target matches the listing: OpenShift Container Platform and OpenShift Virtualization 4.22 on Hopper SXM infrastructure. They should then reproduce the relevant topology—single- or multi-node, full GPU and NIC passthrough—and validate their own storage, networking and model-serving behavior.

Operational tests matter as much as throughput. Teams should verify node labeling, driver ownership, device discovery, upgrades and failure recovery before consolidating containerized AI services and GPU-backed VMs. The certification is useful evidence for the named stack; the production decision still depends on whether that stack’s explicit constraints fit the workload.