​
Select an offer

Start in the Compute Marketplace (sidebar). Choose a GPU type, apply any filters, then click Launch on the offer you want. To start from an app instead, open Templates in the sidebar, choose the environment, then pick one of the suggested GPUs or browse all GPUs with the template kept. You will be taken to the deploy configuration page with the selected offer’s specs displayed at the top:
  • GPU model, VRAM, and architecture
  • Location
  • vCPU count, RAM, and disk specs

​
Configure your deployment

1

Choose a template

Templates determine what software is installed when your instance boots. QuantaCloud, Massed Compute, Verda and Hyperstack offers show the same four choices:
  • PyTorch + Jupyter, with JupyterLab
  • ComfyUI
  • Open WebUI + Ollama
  • A plain GPU VM (labelled Bare Metal), with drivers and Docker but no application container
For the GPU VM option, some offers also let you choose an OS image, a base Ubuntu and CUDA image from the provider’s own catalog, sorted with the newest CUDA version first.Vast.ai offers use a separate template catalog that Vast.ai maintains itself, with its own defaults. It is unrelated to the four templates above.
2

Set disk size

Some offers include a fixed amount of storage. For others, you can adjust the disk size using the slider or number input.Storage costs are included in your hourly rate breakdown.
3

Select a region

If multiple locations are available for your offer, choose your preferred region from the dropdown.
4

Choose an SSH key

Select which SSH key to use for the instance. The key you used most recently is preselected, and among keys you have not used yet, your own keys come before the signup key, which is labelled (from signup). One key is attached per launch, so pick the one you intend to use before you deploy rather than after.Keys are listed with their fingerprint for identification. If you need to add a new key, there is a link to the SSH keys management page.
QuantaCloud only accepts RSA or Ed25519 keys. A key that is ECDSA only fails there with an SSH key error, so keep at least one RSA or Ed25519 key if you deploy to QuantaCloud.
5

Attach a volume (optional)

You can optionally attach an existing volume. Only volumes that are Available and in the same region as your deployment are shown in the dropdown. You can also create a new volume from this page.
Volumes exist only on Verda and Hyperstack offers. If the volume option does not appear, the offer’s provider does not support them.
6

Review cost and deploy

The cost summary shows a full breakdown:Your current credit balance and estimated runtime (balance divided by hourly rate) are displayed.Click Deploy to launch. If your balance cannot cover the first hour, you will see Fund and Deploy instead, which takes you to add credits first.

​
After deployment

Once you click Deploy, your instance goes through these states:
  1. Initializing: setting up the deployment
  2. Provisioning: creating your instance at the provider
  3. Connecting: for a template with an application, the platform waits for that application to start answering inside the container. A plain GPU VM spends little time here.
  4. Running: your instance is ready to use
A plain GPU VM is typically Running in 1 to 3 minutes. A template with an application takes about 6 to 8 minutes. You can monitor progress on the Deployments page.
You are charged for the first hour when you deploy. After that, billing continues hourly. If your balance cannot cover the next hour, the instance switches to per-minute billing until the balance reaches zero, then it stops and its disk is deleted. See Billing for the full mechanics, including the warning emails.

​
Stopping a deployment

On the Deployments page, click Stop on any running deployment. The instance is terminated, its disk is deleted, and billing stops. Any attached volume is detached and returned to the Available state. The volume itself is not deleted. There is no pause that stops billing, only stop. On Verda and Hyperstack, the API can also power a VM off and on without deleting it, with billing continuing while it is off: see Shutdown and Start. See the stop instance guide for a checklist to run before you do.

​
What happens if a deployment fails?

If provisioning fails, the Deployments page shows a Failed status, a plain-language reason, and a suggested next step. The reserved first hour is refunded to your credit balance automatically, whatever the cause. Pick a different offer, region or SSH key depending on what the message says, or see the deployment failures reference for the full list of causes.