Overview

NVIDIA Kubernetes Engine (NVKE) empowers platform teams to deliver a managed Kubernetes service directly on their own bare-metal NVIDIA GPU infrastructure. It provides a robust management plane that enables the creation and operation of independent Kubernetes workload clusters, each with its own control plane, etcd, and node pools. While NVKE handles the management plane, your customers interact with their dedicated clusters via the standard Kubernetes API and kubectl. Currently in preview, NVKE is designed for evaluation and exploration, with APIs and behavior subject to change.

Key Features

• Self-Hosting Kubernetes Factory: NVKE acts as a self-hosting factory, enabling the lifecycle management of numerous isolated workload clusters from a single management plane.
• Independent Workload Clusters: Provision and manage distinct Kubernetes clusters, each tailored for specific customer workloads with their own control planes and resources.
• Comprehensive Lifecycle Management: Seamlessly create, scale, and delete node pools, manage cluster upgrades, and handle node replacements.
• Secure Node Identity: Enforces node identity through SPIFFE/SPIRE and TPM attestation before granting join credentials, with options for etcd data encryption.
• Integrated Build-to-Deployment Pipeline: Works in conjunction with NVIDIA Kubernetes Recipes (NVKR) for building and deploying hardened node images and NVKE components.
• GPU-Optimized Infrastructure: Designed to leverage NVIDIA GPU hardware for accelerated computing workloads within Kubernetes.

Typical Use Cases

• Internal Platform-as-a-Service (PaaS): Enterprises can build and offer their own GPU-accelerated Kubernetes service to internal development teams, abstracting away the underlying infrastructure complexity.
• AI/ML Workload Orchestration: Deploy and manage large-scale AI and machine learning training and inference workloads on dedicated, bare-metal GPU clusters.
• Research and Development Clusters: Provide researchers with on-demand, isolated Kubernetes environments equipped with powerful GPUs for experimentation and development.
• Hybrid Cloud GPU Management: Extend on-premises GPU capabilities into a managed Kubernetes offering, complementing cloud-based solutions.