NVIDIA Opens NVKE and NVKR for GPU-Based Kubernetes Services

NVKE manages clusters, while NVKR builds the software

NVIDIA Kubernetes Engine, or NVKE, provides the software for running a managed Kubernetes service on an AI factory’s infrastructure. It offers APIs to provision and manage clusters, allowing customers to consume GPU capacity through Kubernetes APIs, tools and ecosystem components they already use.

NVIDIA Kubernetes Recipes, or NVKR, handles reproducible software builds. It produces node images, binaries, container images and Helm charts from specific source versions, while recording the inputs used in each build. Its recipes define validated combinations of software versions and GPU generations.

The projects can be adopted together or independently. NVKE has no runtime dependency on NVKR, and NVKR can build artifacts for other Kubernetes management systems. NVIDIA says the division allows AI factories to retain control of their software, operations, service and customer experience.

A management cluster provisions customer workloads

The NVKE management plane includes a REST API and controllers that run on a management cluster. The command-line tool nvkectl starts a temporary seed cluster on a bootstrap host, then uses it to form the management cluster. Once that process is complete, the temporary cluster is deleted.

The management plane creates workload clusters for AI factory customers. Each workload cluster has its own Kubernetes control plane, etcd and node pools. Machines are supplied through NVIDIA Infra Controller, or NICo, which manages bare-metal lifecycle through CAPNICo, its Cluster API infrastructure provider for NICo.

NVKE also includes an OpenTelemetry-based Helm chart for collecting logs, metrics and traces and exporting them over OTLP to a monitoring backend. NVIDIA says a node must present an attested SPIFFE identity before receiving join credentials.

Release sets let customers select validated configurations

AI factories publish release sets that customers select by name. These sets include machine images, cluster releases and node pool releases, allowing customers to choose a validated Kubernetes version without handling operating system images or cloud-init directly.

In NVIDIA’s walkthrough, NVKR builds a node image from version-pinned sources. The build compiles the NVKE node agent and OVN-Kubernetes from source, adds checksum-pinned Kubernetes 1.36 packages and the SPIRE agent, and produces a QCOW2 disk image. The management cluster and each node pool can use different images, including images for different CPU architectures or GPU generations.

Inside workload clusters, teams can add the GPU-accelerated runtime with AI Cluster Runtime, node packages with NodeWright and break-fix automation with NVSentinel. NVKE is a Certified Kubernetes AI Platform and Certified Kubernetes Installer, and NVIDIA says it was conformance tested on Kubernetes 1.36.

Public preview limits production use for now

NVKE and NVKR are available on GitHub under the Apache 2.0 license. NVIDIA says the license allows teams to inspect the code, develop in parallel and contribute improvements. Both projects are in public preview, and their APIs, configuration and behavior may change between releases.

NVIDIA does not yet recommend either project for production workloads. NVIDIA Cloud Partners Firmus and SK Telecom are integrating NVKE and NVKR into their AI factories, while Mirantis, Rafay, Spectro Cloud, OpenNebula Systems, vCluster and Netris are incorporating components into managed Kubernetes and network automation offerings. Users can evaluate NVKE with a local quickstart on a single machine or consult the NVKE documentation to deploy and operate the management plane.