Free AI Software llama.cpp Now Runs on Older Nvidia Graphics Cards

There are two ways to use an AI chatbot today. You can pay a subscription and let a company's servers do the thinking, or you can download a free program and run the AI on your own machine. llama.cpp is the most popular option for the second path. It's free, it works offline, and nothing you type leaves your computer. That privacy and zero monthly cost are why nearly 129,000 people have starred it on GitHub, a code-sharing site.

This release, labeled b11090, is a small behind-the-scenes repair rather than a new feature. Graphics cards from Nvidia need a translator called CUDA to do AI math, and a recent change accidentally broke that translator for older 'Volta' cards — think 2017-era chips like the Tesla V100, common in secondhand workstations. The fix widens a technical rule so it fits both new and old cards. Think of it as a key that suddenly stopped fitting older locks, now filed back into shape.

One detail worth noticing: the patch was written with help from DeepSeek V4.1 Flash, an AI model. In other words, an AI helped repair the free software people use to run AI. That loop — machines maintaining the tools that run machines — is becoming ordinary in open-source projects, where volunteers and hobbyists do the work.

What should you actually do? Probably nothing. This is a pre-release build aimed at developers, not a polished app with a friendly installer. But if you're eyeing a cheap used graphics card or an older PC for local AI, this is quietly good news: the door isn't closing on your hardware yet. Free, private AI keeps working on machines that big companies might consider obsolete.