Every Microsecond Matters: Achieving Near Speed-of-Light Latency in GPU Collectives
Distributed, Parallel, and Cluster Computing
Summary
The authors looked at how GPUs communicate when working together, especially for tasks like decoding language models that need many small, quick exchanges between GPUs. They found ways to make this communication much faster by removing some waiting steps and using smarter memory techniques. By improving a library called NCCL, they created new communication methods that almost reach the fastest possible speeds set by hardware limits. Their improvements help AI models work faster and also benefit traditional scientific computing tasks.
Authors
Siyuan Shen, Anton Korzh, John Bachan, Tiancheng Chen, Arnav Goel, Ludwig Schneider, Pouya Kousha, Zhenhao He, Sylvain Jeaugey, Kamil Iskra, Nishank Chandawala, Jeff R. Hammond, Torsten Hoefler
Abstract
GPU collective communication is typically optimized for bandwidth, yet many emerging workloads are increasingly limited by latency. Long-context decode-heavy large language model (LLM) inference is a prime example, where serving large models requires multiple GPUs, and many small collectives lie directly on the critical path of token generation. Therefore, even microsecond of overhead can impact performance and cost. In this work, we study how to approach the hardware Speed-of-Light (SoL) lower bound for GPU collectives within a scale-up network. We identify key principles for near-optimal designs, including barrier-free synchronization and efficient use of symmetric memory and multicast. Building on NCCL's device-side API, we develop low-latency interfaces for constructing custom collective kernels and use them to implement new symmetric collectives in NCCL. Microbenchmarks show substantial latency reductions for small and medium messages, reducing overhead to within 7% of the absolute SoL lower bound. When integrated into real applications, these kernels improve inter-token latency and throughput in LLM inference and accelerate cuSOLVERMp, demonstrating benefits for both AI inference and traditional HPC workloads.