Free GPU without a credit card: what actually exists
“Free GPU” describes three different arrangements. They fail in different ways, and only one of them is usually available to someone who will not enter a card.
A notebook quota
Google Colab is what most people mean. It gives you a hosted notebook with GPU and TPU access at no charge, and it promises nothing about what you get on a given day: runtime length, GPU type and availability all vary with demand. It is built for interactive work, not unattended jobs.
- No card and no committed capacity.
- Sessions end when you idle out or hit the current limit.
- SSH shells, remote desktops, crypto mining and long unattended jobs are disallowed on free runtimes and can be terminated without warning.
Kaggle notebooks are the other common free option. Its quota is set out in Kaggle's own documentation and does change, so read that page rather than a blog post from last year.
A cloud credit
Oracle, Azure and similar programmes give you credit that can be spent on GPU instances. The credit is money, not capacity: it expires, and the instance you want may cost more per hour than the credit is worth. Oracle and Azure both require a card for identity verification, which rules them out for plenty of people even though the money is free.
- Check the expiry before planning anything.
- Check whether the GPU class you need is covered by the credit.
- Check what happens when it ends: stop, or bill?
Hosted inference
The third option never gives you a GPU. Cloudflare Workers AI, the NVIDIA API catalog and similar services run the model for you and meter the work in their own units — Neurons, rate limits, credits. You get no shell and no custom CUDA, but you also spend nothing and wait for nothing.
What to decide first
- Do you need a shell, or an answer? If it is an answer, stop shopping for GPU credits.
- How long does one run take? Anything measured in hours is a poor fit for a free notebook.
- What happens on the first day of next month? That is the real cost.
Good finds. No digging.
Follow new listings and documented changes in your RSS reader.