A GPU box for your code.

Develop and test GPU code on a cloud Linux machine with a dedicated NVIDIA GPU. Work from any computer through JupyterLab in your browser, or over SSH.

Start a dev machine — $0.60/hr

Connect from any computer

All you need is a web browser — or an SSH client if you prefer your own terminal.

Studio (JupyterLab)

Opens in your browser from the machine page — notebooks, a terminal, a file browser

SSH

Connect from your own terminal as root, after adding your public key from a Studio terminal

Built for GPU development

Root in a Linux container

You are root inside your container: install system packages with apt-get and Python libraries with pip.

PyTorch and CUDA libraries included

Test GPU code without setting up drivers. The CUDA compiler (nvcc) is not in the image; install the CUDA toolkit if you need to build CUDA code.

A GPU that's yours while it runs

Each machine gets one whole NVIDIA GPU, from an RTX 3060 (12 GB) to an RTX A6000 (48 GB). Check it with nvidia-smi.

Stop between sessions

Charges run only while the machine runs. A stopped machine keeps its disk for 24 hours, then it is deleted — keep your code in git or download it.

Pricing

Start at $0.60/hr on Standard, or $0.40/hr on Starter for lighter work. Charged by the second while the machine runs.

Standard
GPURTX 3090
VRAM24 GB
RAM24 GB
$0.60/hr

24 GB of VRAM for larger models

Starter
GPURTX 3060
VRAM12 GB
RAM16 GB
$0.40/hr

Lower cost for lighter work

View all plans →

Start a dev machine — $0.60/hr