Free DIY AI Tool Gets a Speed Boost on New AMD Graphics Cards

There's a free piece of software called llama.cpp that lets ordinary people run powerful AI chatbots directly on their own laptop or desktop, with no monthly subscription and no company watching what they type. On September 11, its volunteer developers released a small update labeled b10905. It doesn't add a shiny new button — it tunes how the software talks to your computer's graphics card.

Specifically, the update targets AMD's newest graphics cards (the Radeon RX 9000 series, known as RDNA4). It improves something called flash attention, which is a clever shortcut that lets AI read long conversations without running out of memory. Think of it like a librarian who can skim a whole bookshelf instead of carrying every book to their desk. The result is that long chats and big documents stay smoother and use less of your machine's memory.

Here's the honest part: this is a tiny, technical release aimed at hobbyists who already run AI at home. It won't change anything for you if you use ChatGPT, Claude or Gemini in a browser. But the bigger trend is worth noticing. Every few weeks, local AI gets a little faster and a little cheaper, meaning the free, private, do-it-yourself option keeps creeping closer to the paid ones.

One fun detail: the release notes credit two AI coding assistants, Claude and Codex, for helping write parts of this update. In other words, AI is being used to make AI better — a loop that's quietly speeding up how fast these tools improve. For now, though, this release mostly matters to a specific crowd: AMD graphics card owners who like tinkering.