A template determines what software runs on your instance from the moment it starts. Pick one on the deploy page and DeployGPU builds the container or image for you, instead of you installing CUDA, PyTorch or a web app by hand.

​
The four templates

App templates take longer than a plain GPU VM because the platform waits for the application itself to answer inside the container, not just for the machine to boot.
The GPU VM template is a virtual machine with drivers and Docker preinstalled, not a bare metal host. The dashboard keeps the label Bare Metal for this option. This page calls it a GPU VM because that is what it is.

​
What each template does not include

None of the three app templates come with models preloaded. Pull or download what you need from inside the application after it starts: ollama pull in Open WebUI, or the model manager in ComfyUI. PyTorch and the CUDA toolkit are ready to import in the Jupyter template, but no datasets or checkpoints ship with it. There is no vLLM template, no standalone Ollama template, no custom Docker image template and no TensorFlow template. If your workload needs one of those, run it yourself on the GPU VM template over SSH, the same way you would on any Linux box with a GPU. The Jupyter template has no separate login of its own either: the private link is the only gate, so anyone who cannot open that link cannot reach JupyterLab.

​
Opening a template

Templates do not expose a public URL. Each one gets a private HTTPS link at a random subdomain of app.deploygpu.ai, and only your DeployGPU account can open it. On the Deployments page, click Open Application to launch it in a new tab. There is no separate Jupyter URL shown next to the SSH details. For the three app templates, Open Application is how you reach the software, not SSH. The GPU VM template has no application link at all. Connect to it with the SSH command shown on the deployment, exactly as described in Connect via SSH.

​
Choosing a template

Templates run on QuantaCloud, Massed Compute, Verda and Hyperstack instances. They do not run through the API, which launches plain VMs only, and they are not available on Vast.ai. Vast.ai offers use a completely separate, dashboard-only template catalog maintained by Vast.ai itself, with its own categories and its own default. It has no connection to the four templates on this page, and switching between a Vast.ai offer and a QuantaCloud, Massed Compute, Verda or Hyperstack offer changes which template list you see. No template is restricted to a particular GPU. The dashboard lets you pick any of the four templates on any compatible offer, so check the GPU a workload actually needs (VRAM in particular) before you pick the offer, not after. For the GPU VM template only, some offers also let you choose an OS image: a base Ubuntu and CUDA image from the provider’s own catalog, independent of the four templates above. The list favors images with both CUDA and Docker preinstalled, sorted with the newest CUDA version first.

Deploy a GPU

See where template selection fits in the full deploy flow, alongside disk size, region and SSH keys.