Available GPU servers
Price per hour for the whole machine, billed by the minute. Launch usually takes 2–5 minutes.
How to choose a GPU server
The GPU
The catalogue shows servers that are free right now. The price is per hour for the whole machine, billed by the minute, in euros. Start with the graphics card: for running LLMs and generating images an RTX 3090 or RTX 4090 with 24 GB is usually enough; training large models needs an A100 or H100 with 80 GB.
The rest of the hardware
Then look at the rest of the machine. RAM should be at least as large as VRAM, otherwise the model loads slowly. A fast disk and a link of 100 Mbit/s or more save time downloading weights and datasets. Reliability shows how steadily the machine has worked before: for long training runs, choose servers with a high score.
After launch
Once it starts, SSH and Jupyter access appears in your dashboard. The server disk is wiped when you stop. If you need your models and results, click Save and stop: we keep the /workspace folder for €0.03 per GB per month, and your next server gets it automatically. Saving does not work yet for custom Docker images. Not sure which card your model needs? See the AI apps and model guides.
How it works
Click Rent, log in or sign up, and top up enough for one hour of the machine. The server starts and SSH and Jupyter access appears in your dashboard. You can save your models and results when you stop and get them back on your next server.