# GPU

> GPU machines for AI, inference or compute. Where to find them, how to choose them, and how to have a service use them.

Source: https://docs.fransys.cloud/en/clusters/gpu/ Checked against the console on 2026-10-23.

## Where to find GPUs

Only some hosting providers and some regions offer them, and their stock changes often. When you create a cluster, answer **Yes** to **Need GPUs (AI, compute)?**: only regions offering GPU machines remain, with the price of the cheapest GPU machine.

If no region appears, no GPU machine is available right now at these providers. Try again later, try **Anywhere** for where your data may go, or answer **No**.

## Choose the machine

At the **How many machines?** step, pick the GPU machine in a machine group. GPU machine types are listed under **With GPU**, with their CPUs, memory and monthly price.

## Have a service use the GPU

In the application's panel, **Resources** tab, set **GPU** to **NVIDIA**. The service will be placed on one of the cluster's GPU machines.

To serve an AI model, an image such as Ollama or vLLM is added as an [Application](https://docs.fransys.cloud/en/projets/application/).

## What it costs

The machine at the hosting provider's price, plus €100 per GPU per month for Fransys. See [Pricing](https://docs.fransys.cloud/en/budget/grille-tarifaire/). A GPU machine is expensive: take the cluster offline when you're not using it.