nvke

nvke

Lets platform teams offer Kubernetes clusters to their customers, built on their own bare-metal NVIDIA GPU machines, with a separate control plane for each cluster. It is a preview, so not for production.

Made by dsx-ai-factory on GitHub. These notes are Mindket's, written from its README; the project's own files and license are in its repository.

About

What it is: NVIDIA Kubernetes Engine, or NVKE, is a management plane for running Kubernetes on bare-metal GPU machines. Teams that run a platform set it up so they can build and look after customer Kubernetes clusters on their own NVIDIA GPU servers. The project is written in Go.

Who it suits: Platform teams that own NVIDIA GPU servers and want to offer managed Kubernetes to their customers will find the closest fit.

Industry | Who uses it | How they use it | Example
Hosting and platform services | Platform teams | Operate the management plane and give each customer a cluster | A team offering GPU clusters to its customers
Infrastructure operations | Operators of GPU hardware | Run Kubernetes directly on bare-metal servers | A rack of GPU servers turned into separate clusters
Cluster users | Customers of a platform team | Work with their cluster through the Kubernetes API and kubectl | Deploying a workload to their own cluster with kubectl

You can skip it if you need something production-ready today or have no NVIDIA GPU hardware, since it is a preview.

Where to use it:

• Bare-metal servers with NVIDIA GPUs: the hardware the management plane runs on
• Kubernetes: each workload cluster has its own control plane, etcd and node pools
• The Kubernetes API and kubectl: how customers reach their clusters

Use case | Who | What they get back
Run a managed Kubernetes service | Platform teams | A service running on their own GPU hardware
Give each customer a cluster | Platform teams | A workload cluster with its own control plane, etcd and node pools
Use a cluster | Customers | Access through the Kubernetes API and kubectl

How to set it up:

• Bare-metal servers with NVIDIA GPUs
• A platform team that will install and run the management plane

1. Clone the repository with git.
2. Open the Get started guide that its README points to.
3. Read the preview status and known limitations before you install anything.

For newcomers and experts: Kubernetes runs and manages applications across many machines. NVKE adds a layer above it, so one team can create separate Kubernetes clusters for others on GPU hardware. A good first step is to read the preview limitations, then the Get started guide.

If you know the field, the design splits in two. You operate the management plane, and every workload cluster carries its own control plane, etcd and node pools. The topics include cluster-api. It is a preview, so the interfaces, settings and behavior may shift from one release to the next.

A platform team installs the management plane
It creates a workload cluster for a customer
Each cluster gets its own control plane and etcd
The customer uses it through kubectl

Good to know:

• Preview: interfaces, settings and behavior may shift from one release to the next
• Not for: production workloads, which the project does not recommend
• License: Apache-2.0
• Cost to run: none named, though you need your own NVIDIA GPU hardware
• What the kit holds: notes on the project and where its code is. Nothing in it runs by itself.

On GitHub, 8 November 2026

• Stars: 569 (123 that day)
• Forks: 90
• Watching: 10
• Open issues and pull requests: 182
• Language: Go
• License: Apache-2.0
• Topics: bare-metal, cluster-api, gpu, kubernetes, nvidia
• Trending: #15 repository of the day on GitHub trending (8 November 2026)

Links

• Code: https://github.com/dsx-ai-factory/nvke
• Website: https://docs.nvidia.com/nvke/
• Releases: https://github.com/dsx-ai-factory/nvke/releases
• Issues: https://github.com/dsx-ai-factory/nvke/issues
• Star history: https://www.star-history.com/#dsx-ai-factory/nvke&Date

Get the code

git clone https://github.com/dsx-ai-factory/nvke.git

Then follow the setup in its README. Nothing in this kit runs by itself.