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

# Clusters

> Deployable allocations of GPU, CPU, RAM, and storage carved from the machines you hold.

![Acasia Developer Portal Clusters page showing an allocated NVIDIA H200 machine and two active clusters](https://docs.acasia.com/assets/portal-clusters-_gOrzbMl.png)

Clusters are the primary compute environment developers interact with — used for:

* Direct SSH access and interactive workloads
* Jupyter and IDE-based development
* Inference endpoint hosting
* Batch and long-running GPU workloads
* Experimentation and model fine-tuning

## Allocated Machines section

| Field          | Description                            |
| -------------- | -------------------------------------- |
| Machine name   | Identifier for the underlying hardware |
| GPU model      | GPU type available for allocation      |
| Available GPUs | GPUs not yet allocated to clusters     |
| Total GPUs     | Total GPUs on the machine              |

Use this section to understand how much capacity is available before creating a new cluster.

## Cluster list fields

| Field     | Description                        |
| --------- | ---------------------------------- |
| Name      | Cluster identifier                 |
| Status    | Current lifecycle state            |
| Machine   | Underlying machine                 |
| GPU       | GPU model allocated to the cluster |
| GPU Count | Number of GPUs allocated           |
| vCPU      | CPU allocation                     |
| RAM       | Memory allocation                  |
| Storage   | Storage allocation                 |
| Network   | Network speed                      |
| SSH       | SSH connection status and access   |

## Cluster lifecycle states

| State        | Meaning                                                                    |
| ------------ | -------------------------------------------------------------------------- |
| Active       | The cluster is available for use — SSH, workloads, and endpoint deployment |
| Provisioning | The cluster is being created or prepared                                   |
| Terminated   | The cluster has been stopped or removed                                    |
| Failed       | The requested operation did not complete successfully                      |
| Out of Sync  | Acasia Cloud and the runtime environment do not currently match            |

## Sold Out state

When all GPU capacity on a machine has been allocated, the machine shows as **Sold Out** in the Allocated Machines section. No additional clusters can be created from that machine until capacity is freed.

If your organization needs additional GPU capacity, contact Acasia.

## Creating a cluster

<Steps>
  <Step title="Review available capacity">
    Check the Allocated Machines section to confirm GPUs are available before proceeding.
  </Step>

  <Step title="Click Create Cluster">
    Select the machine to allocate from and configure the cluster resource allocation — GPU count, CPU, RAM, and storage.
  </Step>

  <Step title="Wait for Active status">
    New clusters enter Provisioning state. Refresh the page and wait for the status badge to show Active.
  </Step>

  <Step title="Attach an SSH key">
    Before connecting, ensure a public SSH key is attached to the cluster. Navigate to Settings → SSH Keys to add a reusable key.
  </Step>
</Steps>

## Cluster actions

From the cluster list, you can:

* **Connect via SSH** — Open the SSH connection panel for the cluster
* **Deploy inference endpoint** — Navigate directly to endpoint deployment for the cluster
* **View details** — Inspect full hardware specifications and connection information
* **Terminate** — Stop the cluster and release its resources

<Warning>
  Terminating a cluster stops all workloads running on it, including active inference endpoints. This action cannot be undone. Confirm there are no critical workloads before terminating.
</Warning>

## Common issues

**Cannot create cluster** — No available GPU capacity on any allocated machine. Check the Allocated Machines section. If all machines are Sold Out, contact Acasia for additional capacity.

**Cluster stuck in Provisioning** — Provisioning may take several minutes. If the cluster remains in Provisioning for an extended period, check the logs or contact support.

**Cluster shows Out of Sync** — Acasia Cloud requested a state that the runtime has not yet confirmed. Review logs and wait for the control plane to reconcile. If the issue persists, contact support.

**SSH connection fails** — Confirm the cluster is Active, an SSH key is attached, and the correct private key is available locally. See [Connect to a cluster with SSH](/get-started/connect-cluster-ssh).