The ggml‑org team published llama.cpp version b11715 on 2026‑10‑17. The update adds a HIP instruction (__builtin_amdgcn_perm) for AMD GPU unpacking, improving AMD GPU performance. It also ships pre‑compiled binaries for Apple Silicon, Intel macOS, iOS, and several Ubuntu configurations (CPU, Vulkan, CUDA 12.8). Users can download the appropriate package from the GitHub release page.
Why it matters
Developers can now run Llama models faster on AMD GPUs and get ready‑to‑run binaries for more platforms without building from source.
