b9099

b9099 is a llama.cpp release that introduces a NCCL-free AllReduce implementation for LLAMA_SPLIT_MODE_TENSOR using a single-phase CUDA kernel that pipelines D2H copy, cross-GPU handshake via pinned-memory volatile flags, and the reduction in one kernel launch per GPU. The release also improves NCCL detection for source-tree builds and adds a static/dynamic switch for FindNCCL.cmake.

Source: llama.cpp Releases | 2026-05-10