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

# Your first instance

> A guided walkthrough of renting a GPU on the marketplace.

This guide assumes you've signed up and topped up your wallet. If not, start
with [Quickstart](/quickstart).

## Step 1 — pick a listing

Open [gpuoutlet.ai/app/create](https://gpuoutlet.ai/app/create). You browse the
marketplace by GPU, cheapest first. Each card shows the model, VRAM, how many
GPUs the machine has, and its live hourly rate — plus, for seller listings, a
**Seller Rating** (0–100) and a **Verified** badge.

Pick a listing, and where the machine has more than one GPU, choose how many you
want. Renting more than one gives you the **whole machine** — the GPUs are pooled
in a single node with their fast interconnect, ready for multi-GPU work. The
footer updates with the live price as you go.

<Tip>
  A count a machine can't satisfy is shown unavailable, and availability shifts
  as sellers list and de-list — re-check shortly if you need a specific config
  right now.
</Tip>

<Note>
  Where a listing offers it, you can also pick a **service tier** at launch:
  **On-demand** (uninterrupted at the headline price) or **Spot** (cheaper, but
  can be reclaimed when a full-price on-demand buyer needs the GPU). See
  [Service tiers](/concepts/instances#service-tiers).
</Note>

## Step 2 — pick a template

Click **Continue**. The template grid loads:

* **PyTorch + Jupyter** — CUDA + PyTorch 2.x with Jupyter Lab (default)
* **Ubuntu + CUDA** — bare OS + CUDA drivers, bring your own stack
* **Ollama** — daemon ready to `ollama pull`
* **vLLM** — OpenAI-compatible inference server
* **Text Generation WebUI** — Oobabooga's text-gen UI
* **ComfyUI** — node-based Stable Diffusion workflows
* **Stable Diffusion (A1111)** — AUTOMATIC1111 WebUI

Pick one. The footer updates with the per-hour cost (templates with preloaded
models can cost slightly more due to disk image size).

## Step 3 — launch

Click **Launch**. A modal shows the instance provisioning; 20–60 seconds later
it flips from **provisioning** to **running**, and the SSH command becomes valid.

## Step 4 — connect

Copy the host and port shown, then connect with **your private key** via `-i`
(the modal shows only the address — it can't know your key's filename):

```bash theme={null}
ssh -i ~/.ssh/id_ed25519 root@123.45.67.89 -p 22001
root@gpu-01:~#
```

Use the key whose public half you added (e.g. `~/.ssh/gpuoutlet_ed25519`). See
[Connecting via SSH](/guides/connecting-via-ssh) for details and troubleshooting.

Verify the GPU:

```bash theme={null}
nvidia-smi
```

## Step 5 — work

The instance is yours until you stop it. Common patterns:

```bash Quick PyTorch sanity check theme={null}
python -c "import torch; print(torch.cuda.is_available(), torch.cuda.get_device_name(0))"
```

```bash Get your code in theme={null}
git clone https://github.com/you/your-repo.git
cd your-repo && pip install -e .
```

```bash Or rsync from local (from your laptop, another terminal) theme={null}
rsync -avz ./project root@123.45.67.89:/root/ -e "ssh -p 22001"
```

## Step 6 — stop

Back in the dashboard, **Instances** → click the instance → **Stop**. It's
unreachable within \~5 seconds, the final partial second is settled, and the
total cost shows in the ledger.

<Warning>
  Stopping wipes the instance's local disk. Push your work somewhere persistent
  (S3, HuggingFace, your laptop via `scp`) BEFORE stopping.
</Warning>

## What if I just close my laptop?

The instance keeps running and the meter keeps ticking. Either stop it from the
dashboard when you're done, or keep your wallet balance at roughly the budget
you're comfortable with — when it hits \$0, every running instance stops within 5
seconds. (A launch-time **auto-stop after N hours** is on the roadmap.)