Models & Labs
Llama.cpp b9387 Release Enhances AMD MFMA Performance
llama.cpp Releaseshigh confidence
Why it matters
- →Optimizes performance for AMD MFMA hardware, enhancing efficiency.
- →Provides significant throughput gains, crucial for high-performance computing.
- →Maintains stability across non-AMD hardware, ensuring broad compatibility.
Llama.cpp's b9387 release focuses on optimizing performance for AMD MFMA hardware, particularly in quantized matrix multiplication tasks. The update adjusts batch threshold logic, resulting in throughput improvements of up to 76% on AMD's MI250X hardware. This release is tailored for users utilizing AMD GPUs, enhancing efficiency without introducing new models. The changes are byte-identical for non-AMD paths, ensuring stability across different hardware configurations.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
How we got here
- 1Llama.cpp b9158 Release Enhances AMD Support
llama.cpp Releases · May 15, 2026 · Same story
- 2Llama.cpp b9255 Release Enhances HMX Matmul
llama.cpp Releases · May 21, 2026 · Same story
More from llama.cpp Releases
Coding Toolscoding
llama.cpp b12022 adds ROCm 10 and Snapdragon support
This release quietly cements llama.cpp as the universal inference runtime by finally bringing ROCm 10.0 to Linux and Windows alongside CUDA 13.4, effectively closing the hardware gap for AMD users who were previously left behind. The inclusion of native Snapdragon support on Linux marks a significant step toward running local LLMs on ARM-based mobile chips without emulation overhead. While KleidiAI on Apple Silicon is disabled in this build, the broader expansion to diverse accelerators ensures that developers aren't locked into NVIDIA ecosystems anymore.
llama.cpp Releases
Coding Toolscoding
llama.cpp b12023 fixes shared prefix bug
This release addresses a specific inference artifact where decision prefixes were incorrectly bleeding into output embeddings. By capping shared prefixes at the first output position, llama.cpp ensures that option tokens remain distinct for parent and child tasks. This is a targeted fix for multi-step reasoning or voting workflows rather than a performance boost. It stabilizes complex generation patterns without changing the underlying model architecture.
