NVIDIA Launches NVKE and NVKR for GPU Kubernetes Management

NVIDIA has announced two open-source projects, NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR), aimed at simplifying the management of Kubernetes clusters on GPU infrastructure. Both projects are now available in public preview and are part of the NVIDIA DSX AI Factory Platform, which supports AI factories in designing, building, and operating GPU infrastructure.

Key Features of NVKE and NVKR

NVKE provides the software necessary to run a managed Kubernetes service on AI factory infrastructure. It includes APIs for provisioning and managing clusters, enabling AI factories to retain control over their software, operations, and customer experience. NVKR complements NVKE by building reproducible operating system images, container images, and software packages tailored to specific GPU generations and Kubernetes versions.

The tools can be used together or independently. NVKE has no runtime dependency on NVKR, and NVKR can produce artifacts for other Kubernetes management systems. Both projects are licensed under Apache 2.0 and are available on GitHub, allowing teams to inspect, develop, and contribute to the codebase.

Operational Workflow and Technical Details

NVKR builds node images by compiling software configurations validated by NVIDIA. These images include checksum-pinned Kubernetes 1.36 packages and GPU-specific runtime components. The build process also records all inputs for reproducibility. NVKE, on the other hand, manages Kubernetes clusters by deploying a management plane that provisions workload clusters for customers. The management plane uses NVIDIA Infra Controller (NICo) for bare-metal lifecycle management and integrates OpenTelemetry-based telemetry pipelines for monitoring.

To deploy NVKE, teams use the nvkectl tool to bootstrap a management cluster and provision workload clusters. Each workload cluster includes its own Kubernetes control plane, etcd, and node pools. Additional components, such as AI Cluster Runtime (AICR) and NVSentinel, enhance GPU-accelerated runtime and automation within the clusters.

Public Preview and Limitations

During the public preview phase, APIs, configurations, and behaviors may change between releases. NVIDIA has advised against using NVKE and NVKR for production workloads at this stage. Detailed documentation, including prerequisites and installation steps, is available to guide teams through building node images, deploying the management plane, and provisioning clusters.

With NVKE and NVKR, NVIDIA aims to streamline Kubernetes-based GPU infrastructure management, catering to the growing demand for scalable and efficient AI factory operations.