# Use a dedicated GPU pool

> Keep a GPU Workspace on your assigned private capacity and reconnect through Nodus.

Source: https://www.nodus-compute.ai/docs/guides/workspaces/private-capacity/
Build revision: 53bfc045e800bf7bb44fa18623013c7aa9c9971e

**Beta:** this feature may change.

Sign in to Nodus and select the project containing your assigned pool. Private GPU Workspaces are in Beta. Ask your project administrator for the pool name and the GPU configuration that was assigned to you.

Create a Workspace manifest with `spec.placement.pool` set to that pool, a GPU request matching the assigned machine, and a home `volume`. The pool must use `routing.mode: Only` and `burstToMarket: Deny`. For a single RTX PRO 6000 Blackwell Server Edition GPU:

```yaml
apiVersion: nodus.dev/v1
kind: Workspace
metadata:
  name: dedicated
spec:
  image: nodus/workspace-pytorch-cuda
  volume: dedicated-home
  idleTimeout: 0s
  resources:
    gpu:
      type: [RTX-PRO-6000-BLACKWELL-SERVER]
      count: 1
      exact: true
  placement:
    pool: reserved-node
```

Save this as `workspace.yaml`, then run `nodus apply -f workspace.yaml`. Wait for the Workspace to show **Ready**. Open it in the console to use VS Code or Jupyter, or connect with `nodus ssh workspace/dedicated` after adding your SSH key to Nodus. Provider credentials are not needed.

This SSH connection opens your Workspace container. It does not provide administrative access to the physical server’s operating system. For an assigned bare-metal reservation, use [reserved server access](https://www.nodus-compute.ai/docs/guides/workspaces/reserved-server/).

`idleTimeout: 0s` keeps the session running while idle. Budget limits, available credit, maintenance and host health still apply. The estimate shows the private routing fee; any reserved-capacity contract is separate.

Use `nodus stop workspace/dedicated` and `nodus start workspace/dedicated` to end and restart a session. Save files in `/home/nodus` to keep them on the home Volume. Scratch files and running process memory are temporary. A sudden host failure can recover only the last saved snapshot.

Stopping a Workspace releases its use of the host and leaves the host enrolled in the pool. If the host is unavailable, the Workspace waits for private capacity. A pool with one assigned host returns to that host; a pool with several hosts can place the next session on another eligible host in the pool.