No fixed closing date.
Kaggle notebooks
Kaggle Notebooks GPU quota
Kaggle gives you a free notebook in the browser with an optional GPU. The weekly GPU quota is 30 hours, and sometimes more when Kaggle has spare machines. No university email is required. It is the most reliable way to start an AI or data job today.
Free. There is no credit card. Hours you waste on an idle session are still subtracted from the weekly quota.
Who it is for
Anyone who can create a Kaggle account. The same weekly quota applies to every user.
What you get
A hosted notebook with CPU, and a GPU when you turn the accelerator on. Official pages have described the free GPU as an NVIDIA Tesla P100, and other Kaggle pages also mention T4 GPUs. The weekly GPU quota resets around Saturday 00:00 UTC. Sessions do not run forever, so save your work.
How to get it
- Create a Kaggle account.
- Open a new notebook. Leave the accelerator off while you install libraries and check that the code runs.
- Turn the GPU on only when the code uses PyTorch, TensorFlow, or another GPU library. Pandas and ordinary R do not get faster on a GPU.
- Watch the quota in your account settings. Stop the session before you close the tab. An idle interactive session can keep counting for up to an hour.
- Download the notebook if you need a copy. Do not rely on the session disk as your only storage.
Deadlines
There is no published closing date. Apply when your one-page description is ready. An empty calendar file would only pretend otherwise, so this page does not invent a reminder.
Read this before you apply
- Thirty hours a week will not train a very large language model from scratch. It will fine-tune a small model and run a lot of analysis.
- The GPU model can change. Check the accelerator menu rather than trusting a blog.
- Kaggle is a public notebook site. Do not upload data you are not allowed to share.
Check before you rely on this
- One official page names Tesla P100 GPUs. Another Kaggle page mentions T4s and P100s. Treat the exact chip as variable.