Lemonade 11.9 Local AI Server Released With Super Exciting AMD ROCm HRX Backend

Lemonade 11.9, an AMD-backed open-source local AI server for Linux, Windows, and macOS, adds an experimental llamacpp-hrx backend built on AMD's new HRX runtime, targeting Radeon RX 7900 series (RDNA3) and Strix Halo APUs on Linux (GFX1100 and GFX1151). HRX is a lighter, client-focused subset of ROCm developed with AMD's Loom/Hyperloom compiler stack, offering a native alternative to Vulkan and HIP backends. AMD engineer Stella Laurenzo detailed the effort in a llama.cpp GitHub discussion, citing 30-50% prefill token/s uplift and up to 10% decode uplift versus existing backends, though the work remains early and limited to a minimal kernel set for the Qwen3-30B-A3B model. Lemonade 11.9 also adds Qwen3-Next support to its pinned llama.cpp backend.

Nguồn: https://www.phoronix.com/news/Lemonade-11.9-ROCm-HRX. 8 Sync News chỉ tóm tắt và dẫn link; bản quyền nội dung thuộc tác giả và nguồn gốc.