Wink Pings

NVIDIA Open-Sources NVKE and NVKR: Run Managed Kubernetes On Your Own Bare-Metal GPUs

NVIDIA open-sourced NVKE and NVKR under the Apache 2.0 license, aimed at enabling managed Kubernetes to run on self-owned bare-metal GPU servers. Both projects are currently in public preview. For teams working on GPU infrastructure, they form a set of reference implementations that can be reviewed and disassembled component by component.

NVIDIA has open-sourced two projects: NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR), released under the Apache 2.0 license. They target a very specific scenario: letting AI factories run managed Kubernetes on their own bare-metal GPU servers. An AI factory can be understood as a GPU-intensive data center. In other words, all components and deployment recipes needed to build a full managed Kubernetes stack on GPU machines are now made public. Both projects are in public preview for now and have not reached General Availability (GA) yet.

For teams building AI infrastructure, the most valuable part of the release is its delivery model. Managed Kubernetes is already mature on CPU clusters, but deploying it on bare-metal GPU environments requires handling a long list of trivial but critical workflows including driver management, device plugins, scheduling, workload isolation, and upgrades. NVKE provides the core runtime engine, and NVKR provides the curated deployment recipes. Combined, the two projects act like a full deconstruction of NVIDIA's own AI factory solution. You don't have to adopt the entire full stack, you can pick only specific parts to reference as pre-built working samples, rather than having to build everything completely from scratch.

By Asif Razzaq, MarkTechPost, 2026/11/17. Link: NVIDIA Open-Sources NVKE and NVKR for Managed Kubernetes on GPU Infrastructure

发布时间: 2026-11-17 18:37