Tag
llama.cpp build b9095 introduces NCCL-free tensor parallelism for dual Blackwell PCIe GPUs, enabling efficient multi-GPU inference without relying on NCCL.