ggml-org/llama.cpp is a project that performs LLM inference in C/C++. In GitHub release b12001 it adds RDNA4_M family definitions for gfx1170 and gfx1171 to ROCm/HIP, and adjusts the related settings to match those targets. Therefore the core change confirmed in this version is the definitions and settings for the two targets, and the given release information offers no basis for claiming that any other changes were included.
This change focuses on ROCm/HIP-related settings for specific targets. gfx1170 and gfx1171 are named explicitly, and RDNA4_M family definitions matching those two targets were added. The area the project covers is C/C++-based LLM inference. The release also includes executables for macOS, iOS, and Linux. However, the provided information does not state the composition or usage conditions of each platform executable. Alongside the target of this update, the fact that these executables are included is also recorded in release b12001.
The scope of judgment should be limited to the changes stated in the release information. The provided content contains no performance figures or benchmark results for gfx1170 and gfx1171, and does not describe the status of other AMD GPU targets. Therefore this change alone cannot be taken as proof of faster processing or of a broad expansion of support. What is confirmed is the addition of definitions for the two targets and the adjustment of settings; anything beyond that is difficult to judge from this information alone.