NVIDIA has open-sourced two projects for running managed Kubernetes on GPU infrastructure, MarkTechPost reports: NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR). Both are parts of the NVIDIA DSX AI Factory Platform, ship under Apache 2.0, and are in public preview. NVKE is the control plane for a managed Kubernetes service that you install on your own NVIDIA GPU bare metal. Customers get ordinary clusters and use kubectl. NVKR is the build system that produces the OS images, container images, Helm charts, and packages those clusters depend on. NVIDIA’s own blog says they are not yet recommended for production, and the preview still has lifecycle gaps, including no management-plane upgrade or backup procedure in the write-up I saw.
I have not installed this, and you should not point a preview control plane at a cluster you care about.
Review before you install: read the release notes yourself, back up your config, and try new versions somewhere you can afford to break first.
Isolate it: try new software in a disposable VM or on a spare machine first, with a snapshot you can roll back to. Don’t point it at a machine you care about.
My take
Open, attested, upstream Kubernetes for GPU clouds is a serious idea if the preview hardens. Until there is a real upgrade path, and until the default management API is locked down, I would treat NVKE and NVKR as something to evaluate in a lab, not as a new production plane.
God Bless, and Happy hacking!