Get started
Getting capacity out of vMetal takes two choices. First, which driver provisions the server. Second, what that server becomes once it's provisioned.
1. Choose a provisioning driver​
Start with the workflow that matches the infrastructure you provide. Each driver uses the same NodeClaim provisioning API and lifecycle.
| If you want to | Start with |
|---|---|
| Provision a bare metal instance | Configure the Metal3 driver |
| Provision a virtual instance | Configure the KubeVirt driver |
| Use an existing NVIDIA Infra Controller site | Configure the NICo driver, then provision with NICo |
| Provision through Terraform or OpenTofu | Configure the Terraform driver |
The infrastructure operator configures the driver and makes capacity available. Platform and Tenant administrators define which resources Projects can consume.
2. Choose what the server becomes​
Once servers are registered and a node type describes them, decide what to claim them for. All three paths create a NodeClaim and draw from the same pool. They differ by a field or two on the request.
| If you want to | Go to |
|---|---|
| Hand out a server over SSH, with no Kubernetes on it | Independently provisioned instances |
| Run a cluster's control plane on your own hardware | Control plane instances |
| Give an existing cluster dedicated worker capacity | Cluster nodes |
Configure supporting resources​
| If you want to | Go to |
|---|---|
| Understand how an instance becomes a private node | Cluster attachment |
| Configure reusable images, SSH keys, and startup configuration | Images and reusable configuration |
| Understand the instance and server model | Instance and server concepts |
| Claim a specific server from inventory, not just a node type | Machine inventory |