> ## Documentation Index
> Fetch the complete documentation index at: https://docs.usechamber.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Cluster Deployment

> Deploy GPU-ready Kubernetes clusters with Terraform

Chamber provides Terraform modules that deploy complete GPU-ready Kubernetes clusters on AWS and Google Cloud. A single `terraform apply` creates the entire infrastructure stack — from networking to GPU workload orchestration.

## What Gets Deployed

Each module provisions the full stack your team needs to run GPU workloads with Chamber:

```mermaid theme={null}
flowchart TD
    VPC["VPC / Network"] --> K8s["Kubernetes Cluster\n(EKS or GKE)"]
    K8s --> System["System Node Pool"]
    K8s --> Karpenter["Karpenter\n(GPU Autoscaling)"]
    K8s --> NVIDIA["NVIDIA GPU Operator\n(Drivers & Metrics)"]
    K8s --> KAI["KAI Scheduler\n(Fractional GPU)"]
    K8s --> Agent["Chamber Agent"]
    Agent -->|Secure connection| Chamber["Chamber Control Plane"]
```

## Terraform Modules vs. Helm Agent

Chamber offers two ways to connect a cluster. Choose based on your starting point:

|                     | Terraform Modules                    | Helm Agent Only                       |
| ------------------- | ------------------------------------ | ------------------------------------- |
| **What it creates** | Full infrastructure stack            | Chamber agent only                    |
| **Prerequisites**   | Cloud provider account + Terraform   | Existing Kubernetes cluster with GPUs |
| **Time to deploy**  | \~15-20 minutes                      | \~30 seconds                          |
| **GPU autoscaling** | Included (Karpenter)                 | Not included                          |
| **GPU drivers**     | Included (NVIDIA GPU Operator)       | Must be pre-installed                 |
| **Best for**        | New clusters, greenfield deployments | Existing clusters                     |

<Info>
  Already have a Kubernetes cluster with GPU nodes? Skip the Terraform modules and [install the Chamber agent directly via Helm](/agent/installation).
</Info>

## Choose Your Cloud

<CardGroup cols={2}>
  <Card title="AWS EKS" icon="aws" href="/cluster-deployment/aws-eks">
    Deploy a GPU-ready EKS cluster on Amazon Web Services
  </Card>

  <Card title="Google Cloud GKE" icon="google" href="/cluster-deployment/google-gke">
    Deploy a GPU-ready GKE cluster on Google Cloud Platform
  </Card>
</CardGroup>

## Components

Both modules deploy the same core components:

<AccordionGroup>
  <Accordion title="Karpenter — GPU autoscaling">
    [Karpenter](https://karpenter.sh/) automatically provisions and deprovisions GPU nodes based on workload demand. When a GPU job is submitted, Karpenter launches the right instance type. When nodes sit idle, Karpenter consolidates or terminates them to reduce costs.
  </Accordion>

  <Accordion title="NVIDIA GPU Operator — drivers and metrics">
    The [NVIDIA GPU Operator](https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/index.html) installs GPU drivers, the device plugin (enabling `nvidia.com/gpu` resource requests), DCGM-Exporter (GPU metrics for Chamber dashboards), and GPU Feature Discovery (node labeling by GPU type).
  </Accordion>

  <Accordion title="KAI Scheduler — fractional GPU support">
    The KAI Scheduler enables GPU time-sharing, allowing multiple containers to share a single GPU. This improves utilization and reduces costs for workloads that don't need a full GPU.
  </Accordion>

  <Accordion title="Chamber Agent — workload orchestration">
    The [Chamber Agent](/agent/overview) connects your cluster to the Chamber control plane, reporting capacity, managing workloads, and collecting GPU metrics.
  </Accordion>
</AccordionGroup>