Models & Labs
Llama.cpp b9095 Release Enhances CUDA AllReduce
llama.cpp Releaseshigh confidence
Why it matters
- →Provides a NCCL-free alternative for CUDA AllReduce, reducing dependencies.
- →Enhances system reliability with improved error logging and hang detection.
- →Targets specific GPU configurations, optimizing performance for certain use cases.
Llama.cpp's b9095 release introduces an internal AllReduce kernel for CUDA, offering a NCCL-free implementation for tensor parallelism. This update supports configurations with two GPUs and FP32 tensors up to 256 KB, providing an alternative to NCCL and enhancing flexibility. The release also improves error logging and introduces a watchdog feature to detect hangs, aiming to increase system reliability. These changes are designed to streamline operations and reduce dependencies for developers using llama.cpp in tensor-parallel modes.
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.
What happened next
- 1llama.cpp b9105 Release Enhances CUDA Integration
llama.cpp Releases · May 12, 2026 · Same story
- 2llama.cpp b9788 release enhances dual-GPU support
llama.cpp Releases · June 26, 2026 · Same story
Follow this story
Open the full 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.
