Free AI Tool Llama.cpp Gets Faster on AMD Graphics Cards
This update makes running AI on your own computer quicker and cheaper.
llama.cpp's latest build notes list one change: "ggml-cuda: hip: tune GCN MMQ configs (#30021)." That's it — the entry names a tuning of GCN MMQ configs in the ggml-cuda HIP path. The same page lists attestations plus build artifacts across macOS/iOS, Linux, Android, Windows, and openEuler, covering CPU, Vulkan, CUDA 12/13, ROCm 10.0, SYCL, OpenVINO, OpenCL (Adreno), and Snapdragon CPU/Adreno GPU/Hexagon NPU targets, along with a UI build. The notes do not describe performance gains, hardware requirements, setup difficulty, or cost savings — according to the article, it's a config tuning commit alongside a build matrix.
- Llama.cpp now runs AI faster on AMD graphics cards, closing the gap with NVIDIA.
- This means you can use AI locally for free, with more privacy and no monthly fees.
- Setup still requires technical know-how, so it's not for everyone yet.
Why It Matters
Faster local AI on AMD means more people can run private, free AI without cloud costs.