Introduction to BYOK

Bring Your Own Kubernetes (BYOK) connects your Kubernetes cluster to DGX Cloud Lepton. You add GPU nodes from your cluster to Lepton node groups. Then you use Lepton to run and manage workloads on these nodes.

This guide is for the administrators who prepare, connect, and maintain the cluster.

How BYOK Works

You install the cluster agent, lepton-cluster-agent, in your cluster. The agent opens an outbound, authenticated tunnel to Lepton. Lepton uses this tunnel to access your Kubernetes API and to install its platform components. You do not need to expose your Kubernetes API publicly.

Your cluster supplies:

• Platform capacity: CPU nodes for the cluster agent and the platform components.
• GPU nodes that you add to Lepton node groups. You run workloads on these nodes through Lepton.

Lepton hosts its control plane. You do not install the control plane in your cluster.

BYOK and BYOC

Bring Your Own Compute (BYOC) connects your machines to a Kubernetes environment that Lepton manages. With BYOK, your machines are already in your Kubernetes cluster. You continue to operate the infrastructure, Kubernetes, networking, storage, and GPU software.

Lepton manages its platform components and the workloads on the nodes that you add. When you release a node, Lepton stops managing it. The Kubernetes Node and the machine stay in your cluster.

Get Started

1. Review the requirements. Check the cluster capabilities, platform capacity, GPU nodes, and the access that you grant to Lepton.
2. Prepare your cluster. Install the prerequisites, prepare the GPU nodes, and do the checks.
3. Connect your cluster. Register the cluster, install the cluster agent, and add the GPU nodes.
4. Manage your cluster. Maintain nodes, upgrade components, and release resources.