ggerganov/llama.cpp Released Build b12001 with ROCm RDNA4_M Support
llama.cpp release b12001 introduces ROCm/HIP support for the new RDNA4_M family of AMD graphics hardware. Specifically, it adds definitions and configurations for gfx1170 and gfx1171 hardware architectures. This release expands local AI execution capabilities to newer AMD GPU architectures, ensuring users with the latest hardware can efficiently run large language models. It helps maintain broad hardware compatibility across the rapidly evolving GPU landscape. The update adjusts architecture definitions to account for the unique characteristics of gfx117X chips, such as 128b WMMA instructions, separating them from standard RDNA4 implementations. It includes contributions from developers and AMD representatives like Mario Limonciello.
## BACKGROUND
llama.cpp is a popular open-source software library designed for efficient inference of large language models on various hardware platforms, co-developed with the GGML tensor library. ROCm is AMD's open-source software platform for GPU-accelerated high-performance computing and machine learning.