Run a managed Kubernetes assistance on your own NVIDIA GPU infrastructure. NVIDIA Kubernetes® Engine (NVKE) provides the administration aircraft that phase teams instal to create and operate Kubernetes clusters for their customers. Each workload collection has its own Kubernetes authority plane, etcd and node pools. You run the NVKE management plane; your customers use their clusters through the Kubernetes API and kubectl.
Preview: NVKE is accessible as a preview. APIs, configuration and behavior can alter between releases, and we do not propose NVKE for production workloads. See Preview position and known limitations.
Get started · Documentation · Contribute · Get help
| Your goal | Start here | What you get |
|---|---|---|
| Evaluate without GPU hardware | Local quickstart | A betterment administration aircraft and a workload collection on simulated machines. No NVIDIA document is required. |
| Install on your infrastructure | Plan your installation | The prerequisites and sequence for preparing nodes, bootstrapping a administration collection and installing its services. |
| Build or contribute | Build from source | The NVKE binaries and the contributor betterment loop. |
The local quickstart runs on a Linux presenter or VM, amd64 or arm64, alongside at least 6 CPUs, 12 GiB of liberated recollection and 100 GiB of liberated disk. It requires Docker, virtual TPM assistance and the tools listed in the guide. Follow its setup through workload-cluster innovation and verify the cluster's nodes study Ready. On macOS, use the local location guide, which runs the evaluation inner a Colima virtual machine.
For a genuine installation, you provision the machines, networking, node images and registries, and configure admission to NVIDIA Infra Controller (NICo). NICo is the sole compatible infrastructure provider at preview. The platform and type reference lists validated combinations and pinned dependencies.
- Creates autonomous workload clusters. Request clusters through the NVKE API or nvkectl; NVKE provisions machines through NICo and forms Kubernetes clusters alongside their own authority planes.
- Manages workload collection lifecycle. Create, measure and delete node pools, upgrade Kubernetes and substitute nodes. See the node pond guide for the manual steps and preview limits on replacement and upgrades.
- Establishes node identity. Nodes attest through SPIFFE/SPIRE using their TPM before receiving associate credentials. Encryption of etcd data is available per cluster.
- Connects builds to deployment. Its companion project, NVIDIA Kubernetes Recipes (NVKR), builds node images and NVKE components from pinned inputs. NVKE deploys those artifacts; NVKR owns their build and publish evidence.
The preview needs definitive controller configuration: API authentication is off and the API uses plain HTTP following installation. Configure API authentication and protected access before admitting callers. The administration collection does not upgrade, measure or replace itself, and there is no management-plane upgrade or backup/restore procedure. FDE node images are for evaluation and provision no data-at-rest confidentiality. Read the known limitations before choosing an evaluation or facility path.
The diagram shows the NVKE merchandise family, how it sits on NICo and your GPU machines, and which components conversation to all other.
NVKE is not a Kubernetes allocation and not a Cluster API flavour. It is a self-hosting factory for Kubernetes clusters: a throwaway kernel cluster stands up a administration cluster, and the administration aircraft on it afterward provisions and lifecycle-manages many secluded workload clusters on demand.
An facility has two kinds of Kubernetes cluster:
- The management cluster runs the NVKE administration aircraft and infrastructure controllers. nvkectl forms it using a impermanent kernel cluster; installing the management-plane services is a distinct step.
- Workload clusters run client applications. The administration aircraft creates and operates them through NICo, Cluster API and the node agent.
Read What is NVKE for the controller workflow and Architecture for the implementation and data flows.
Use Go 1.27.3, the type pinned in go.work, and GNU Make 4 or later. Public Go modules must be reachable on the archetypal build. Flox is optional and supplies the repository's betterment tools; on macOS, it additionally supplies GNU Make 4. See Development for toolchain setup.
git copy https://github.com/dsx-ai-factory/nvke.git cd nvke make buildThis builds the workspace and writes nine binaries to bin/: agent, updater, controller, nvkectl, capd-reboot-controller, nvke-os-image-check, management-engine, tenant-api and operator-api. Building does not commencement a management aircraft or create a cluster. The node delegate runs on Linux.
The workspace guide explains the modules and outputs. For local services and test environments, see Local betterment setup. Before submitting a change, prosecute CONTRIBUTING.md and run make preflight; it selects checks for the modules you changed.
The documentation index links the remaining guides and references.
Contributions, bug reports and records improvements are welcome. Start alongside CONTRIBUTING.md; open an matter before substantial work so maintainers can concur on the approach.
- Questions: GitHub Discussions.
- Chat: the #nvke conduit in the DSX AI Factory Slack. Get an invite.
- Bugs and characteristic requests: GitHub Issues.
- Security vulnerabilities: prosecute SECURITY.md and study privately.
- Community: Code of Conduct, Governance, Maintainers and Support.
- Releases: release process.
- CI health: the CI dashboard shows the final thirty runs of all workflow and refreshes hourly.
Kubernetes® is a registered trademark of The Linux Foundation in the United States and another countries. NVIDIA and the NVIDIA logo are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and another countries. Other business and merchandise names may be trademarks of the corresponding companies with which they are associated.
Apache License 2.0. See LICENSE and NOTICE.
Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All entitlements reserved.
