NVIDIA Open‑Sources NVKE and NVKR to Enable Managed GPU Kubernetes
NVIDIA released its Kubernetes Engine and Recipes under Apache 2.0, letting AI teams run managed Kubernetes on bare‑metal GPU servers.
Source: MarkTechPostNVIDIA has placed its NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR) in the public domain under an Apache 2.0 license, giving AI developers a ready‑made path to run managed Kubernetes directly on bare‑metal GPU hardware. The announcement, reported by MarkTechPost, signals a shift toward more self‑hosted, high‑performance AI compute clusters.
What are NVKE and NVKR?
NVKE is a purpose‑built Kubernetes distribution that integrates NVIDIA’s GPU‑specific drivers, device plugins, and scheduling logic. NVKR is a collection of curated Helm charts and configuration templates that automate the deployment of common AI workloads—such as model serving, training pipelines, and data preprocessing—on top of NVKE. Together they form a stack that abstracts the complexities of configuring GPUs inside a Kubernetes cluster.
Why does open‑sourcing matter?
Releasing the code under the permissive Apache 2.0 license removes barriers for organizations that need full control over their compute environment. Companies can audit, modify, and extend the engine without vendor lock‑in, while also benefiting from community contributions. Open source also encourages broader adoption, as teams can align the stack with internal security policies and integrate it with existing CI/CD pipelines.
How does the stack enable a managed Kubernetes service on bare‑metal GPU servers?
Traditional managed Kubernetes offerings run on cloud VMs, which abstract away the hardware layer. NVKE brings the management plane to on‑premise servers that are equipped with NVIDIA GPUs, allowing the same declarative APIs (kubectl, Helm, etc.) to schedule GPU‑intensive pods directly on the hardware. NVKR’s recipes pre‑configure device plugins, GPU‑aware autoscalers, and monitoring stacks, so an operator can launch a "managed" service that feels like a cloud offering while retaining the performance and cost advantages of owning the hardware.
What practical steps are needed to adopt NVKE/NVKR?
- Provision bare‑metal servers with supported NVIDIA GPUs and a compatible Linux distribution.
- Install the NVKE binaries and bootstrap the control plane using the provided installer script.
- Apply NVKR Helm charts to set up GPU device plugins, metrics exporters, and example AI workloads.
- Configure networking and storage according to the organization’s policies.
- Validate the deployment by running a sample training job or model server and confirming GPU utilization through the built‑in dashboards. The public preview documentation includes step‑by‑step guides and sample configurations, making the process approachable for teams familiar with Kubernetes but new to GPU orchestration.
How will this affect AI factories and research labs?
AI factories—large‑scale production environments that train and serve models continuously—often struggle with the latency and cost overhead of moving workloads to public clouds. By running NVKE on their own GPU farms, they can keep data on‑premise, reduce network hops, and achieve tighter security compliance. Research labs gain a reproducible environment that mirrors production clusters, simplifying the transition from prototype to deployment. The open‑source nature also means that cutting‑edge features, such as new driver optimizations, can be merged upstream faster than with proprietary solutions.
How to explore NVKE/NVKR with Grafy tools?
While Grafy does not host a dedicated Kubernetes deployment module, developers can leverage Grafy Code to script the installation and configuration steps described above. The AI‑assisted coding studio can generate installer scripts, Helm values files, and CI pipelines, reducing manual boilerplate. For teams that need to read the newly released documentation, Grafy Reader offers a searchable interface to ingest PDF or markdown guides, allowing users to ask natural‑language questions about specific configuration flags. Both apps are accessible from the Grafy platform and can accelerate the onboarding process.
Frequently asked questions
Does open‑sourcing NVKE mean that anyone can run a fully managed GPU Kubernetes cluster on their own hardware?
Yes. The Apache 2.0 license permits anyone to download, modify, and operate NVKE on bare‑metal servers, provided they follow the installation steps and use NVKR’s recipes to automate the setup.
Are NVKE and NVKR ready for production workloads?
Both components are currently in public preview, which means they are functional and documented but may still receive updates and bug fixes. Early adopters can run production‑grade workloads but should monitor release notes for stability improvements.
Can Grafy Code directly deploy NVKE to a cluster?
Grafy Code can generate the necessary scripts and Helm commands, but the actual execution still requires a target environment (e.g., a set of GPU servers). Users can run the generated code locally or through a CI/CD pipeline to provision the cluster.
Source: MarkTechPost — NVIDIA Open-Sources NVKE and NVKR for Managed Kubernetes on GPU Infrastructure


