The Pulse
NVIDIA open-sources two Kubernetes projects for GPU infrastructure
Two projects for building and operating managed Kubernetes services on GPU infrastructure are available in public preview. The projects separate cluster management from reproducible builds of the software artifacts those clusters use.
AI.info Team · Written with AI · fact-checked ·
Two projects for managed GPU clusters
NVIDIA has made two open-source projects for building and operating managed Kubernetes services on GPU infrastructure available in public preview. NVIDIA Kubernetes Engine (NVKE) handles cluster management, while NVIDIA Kubernetes Recipes (NVKR) builds the software artifacts those clusters use.
NVIDIA says AI factory customers want to use GPU capacity through familiar Kubernetes APIs and tools while the factory operates the clusters and describes the projects as dividing the work: NVKE supplies APIs to provision and manage clusters, and NVKR produces operating system images, container images and packages from defined software versions.
Factories can adopt the projects together or separately, NVIDIA says. NVKE does not depend on NVKR at runtime, and NVKR can build artifacts for other Kubernetes management systems.
How the system is assembled
The walkthrough begins with building a node image using NVKR and deploying an NVKE management plane. The management plane runs on a Kubernetes management cluster, which the nvkectl command-line tool sets up by starting a temporary seed cluster on a bootstrap host.
From there, NVKE’s controller requests servers from NVIDIA Infra Controller, or NICo, through CAPNICo, the infrastructure provider for NICo. Once the management cluster is in place, it can create customer workload clusters, each with its own Kubernetes control plane, etcd and node pools. A node must present an attested SPIFFE identity before it receives credentials to join, NVIDIA says.
NVKR builds node images from version-pinned sources and records the inputs. In the walkthrough, the build compiles the NVKE node agent and OVN-Kubernetes, adds Kubernetes 1.36 packages and the SPIRE agent, then produces a QCOW2 disk image.
Factories publish named release sets for customers to choose from. The release sets include a machine image, a cluster release that records the Kubernetes version and control-plane configuration, and a node-pool release for worker nodes. Customers submit cluster requests through the NVKE API and access their Kubernetes API through a kubeconfig.
Preview status and partners
NVIDIA says its cloud partners Firmus and SK Telecom are integrating NVKE and NVKR into their AI factories. NVIDIA also names Mirantis, Rafay, Spectro Cloud, OpenNebula Systems and vCluster as partners integrating components into managed Kubernetes offerings, while Netris is enabling NVKE for network automation.
Both projects are available on GitHub under the Apache 2.0 license. They are in public preview, and NVIDIA says their APIs, configuration and behavior may change between releases; it does not recommend them for production workloads yet.