How GPU rentals work

A dedicated GPU machine with full SSH access, two ways: On-Demand (start one that is free now and pay while it runs) or Scheduled (book the hours ahead and have them for sure).

1. On-Demand: start now, pay while it runs

• Top up your AxForge balance (€10 at least), pick a machine that is free, press Start.
• It is paid from your balance by the started hour. The first hour counts from the start.
• It ends when you press End, when your balance runs out (after a 10-minute grace), at the hours you chose (at most 48 hours), or when a Scheduled booking on that machine begins.
• ⚠️ Closing SSH or your browser does not end it. Only End does.

2. Scheduled: book the hours ahead

• Book whole hours (1 hour at least): continuous, working hours, or your own pattern.
• Pay when you book, by card or from your balance. The machine is prepared before your time starts.
• Cancel 24 hours or more before it starts: the time is released and the value stays on your balance. Later than that, the time is yours whether you use it or not.
• 🔴 End early (type END): access ends at once, the machine is wiped, and the hours still to come are released. They are not refunded.

3. Connect and run

• Add your SSH key on the machine's page in the console; it is installed on the machine.
• Then connect. The exact line, with your host and port, is on that page:

# HOST, PORT and BOOKING are your machine's: copy the line from its page in the console
ssh -o HostKeyAlias=axforge-BOOKING -o StrictHostKeyChecking=accept-new dev@HOST -p PORT

The two -o options give every booking its own entry in your known_hosts: each booking has its own host key, and a machine you rented before connects without a warning.

On the machine, the tools are installed, one command each:

start-comfyui                # ComfyUI      → port 8188
start-jupyter                # JupyterLab   → port 8888 (token: axforge)
vllm serve Qwen/Qwen3-8B     # an OpenAI-compatible API → port 8000
start-ollama                 # Ollama       → port 11434
hf download <repo> --local-dir /workspace/models/<name>

Models go on your own disk, /workspace: ComfyUI loads them from /workspace/models/<folder> (checkpoints, diffusion_models, loras, vae …). Every image, video and music model's page in the catalogue has its steps.

Which models fit the machine you rent: RTX 3060 (one 12 GB card) · RTX 5090 (32 GB) · DGX Spark (about 110 GB for the model). Each model page shows the smallest card that runs it.

The apps run on the machine, not on the internet. To use one in your browser, forward its port:

# a second terminal: the app's port on your own computer (here ComfyUI, 8188)
ssh -o HostKeyAlias=axforge-BOOKING -o StrictHostKeyChecking=accept-new -L 8188:localhost:8188 dev@HOST -p PORT
# then open http://localhost:8188

4. If the machine does not work

• On-Demand: while the service is failing nothing is charged, and whole hours it was down are not charged at all.

5. Nothing is saved

• ⚠️ Nothing is saved. When a rental ends, the machine and everything on it (including /workspace) are deleted and cannot be recovered.
• Copy out anything you want to keep before it ends.