Pay per minute, no subscription
Train your model on a cloud GPU, from your browser
Write or upload your training code, pick a GPU, and see the price and the most it can cost before you start. GridPup rents the machine, runs your code, saves the results and shuts it down.
Try the editor without an account · Sign in and add credit to run
Training without looking after a machine
No cloud account
One GridPup balance. We rent the machine for you on the cloud with the best price for your GPU.
The cost before you start
The price per hour and the most a run can cost, shown before anything is rented. It stops at your limit.
No idle hours
The machine is shut down the moment training ends, fails or is stopped. You pay for the minutes it ran.
Everything in one place
Edit your code, watch the logs, stop the run and download the results from the same page.
How it works
Write or upload your code
Edit in the browser, or upload files, a folder or a .zip. GridPup suggests what to install and run.
Choose a GPU and limits
Pick a GPU with today's price, and the most the run may cost and how long it may take.
Review and start
See exactly what will run, on which machine, and the worst-case cost. Then start.
Watch and download
Follow the logs live, stop when you like, and download your model from /outputs when it ends.
Pay only for the machine
Per minute, from your credit, at the cloud's price shown before you start. Prices change with the market; these are live.
Loading today's prices…
Add credit from $10 in Billing. Up to five runs at once.
Questions
What can I run?
Any training code that runs on Linux with an NVIDIA GPU: PyTorch, TensorFlow, JAX, Hugging Face, Lightning, scikit-learn and more. You choose a ready environment (PyTorch, TensorFlow, CUDA or plain Ubuntu) and add your own packages in Setup, for example pip install -r requirements.txt.
How much does it cost?
You pay only for the machine while it runs, per minute, from your GridPup credit. The price per hour is shown before you start, together with the most the run can cost. There is no subscription and no fee on top.
Can I stop a run?
Yes, at any time. Stop ends your code, saves what is in /outputs, and shuts the machine down. Stop now shuts it down at once without saving. Every run also stops by itself at the cost or time limit you set.
Where do my results go?
Everything your code writes to /outputs (the GRIDPUP_OUTPUT_DIR folder) is saved when the run ends: models, checkpoints, metrics. Download it from the run for 7 days.
Do I need an account to try it?
No. Open the editor and write or upload code without an account; your files stay in your browser. Sign in and add credit when you want to run it on a GPU.
How big can my project be?
Up to 2,000 files and 100 MB of code per project, each file up to 10 MB. Download larger datasets from your code during the run, for example from Hugging Face; uploading large data to GridPup is coming next.
What if the machine fails?
GridPup picks the cheapest verified machine with your GPU. If one does not start, it is shut down and the next is tried, and you are not charged for it. If a machine fails during training, the run ends and you are charged only for the time it ran.
How many runs can I have at once?
Up to five runs at the same time; more wait in a queue and start as soon as one finishes.
Start in the editor
No account needed to write or upload your code. Sign in and add credit when you are ready to run.