Local Ai
b8739
llama.cpp release **b8739** is a build of the open-source C/C++ LLM inference engine that introduces HIP backend support for the CDNA4 (gfx950) GPU architecture, enabling hardware acceleration on A...
llama.cpp release b8739 is a build of the open-source C/C++ LLM inference engine that introduces HIP backend support for the CDNA4 (gfx950) GPU architecture, enabling hardware acceleration on AMD Instinct MI350X and MI355X datacenter GPUs. The gfx950 architecture shares most MFMA matrix instructions with CDNA3 (gfx942), with the exception of mfma_f32_16x16x8_xf32, which is unavailable on gfx950 and is routed to a standard f32 fallback path. The change was validated on an AMD Instinct MI355X with ROCm 7.0.1, achieving approximately 40,013 tok/s prefill and 254 tok/s decode on a Qwen2.5-1.5B Q4_K_M model with Flash Attention enabled.
Related
Source: local-ai