NVIDIA Kubernetes Engine (NVKE) is a management plane for running Kubernetes on NVIDIA GPU bare metal infrastructure. Platform teams install NVKE to create and operate independent workload clusters, each with its own control plane, etcd, and node pools. Customers access clusters via standard Kubernetes API and kubectl.
Currently in preview, NVKE is not recommended for production. A local quickstart runs on Linux (amd64/arm64) with 6 CPUs, 12 GiB RAM, and 100 GiB disk using Docker and virtual TPM. macOS users evaluate via Colima VM. Real installations require machines, networking, node images, registries, and NVIDIA Infra Controller (NICo) as the sole compatible provider.
NVKE provisions clusters through NICo and Cluster API, manages lifecycle (scaling, upgrades, node replacement), and establishes node identity via SPIFFE/SPIRE with TPM attestation. Encryption of etcd data is available per cluster. The companion project NVKR builds node images and components from pinned inputs.
Preview limitations include plain HTTP API without authentication, no management-plane upgrade or backup/restore, and FDE node images lacking data-at-rest confidentiality. NVKE is a self-hosting factory: a throwaway seed cluster stands up the management cluster, which then provisions isolated workload clusters on demand.
Source: Hacker News · Summarized by HeadlinesBriefing