gZCCL: Compression-Accelerated Collective Communication Framework for GPU Clusters

TL;DR

gZCCL is a novel framework that enhances GPU cluster communication efficiency by integrating accuracy-aware compression, significantly outperforming existing solutions while maintaining high data quality.

Contribution

This paper introduces gZCCL, the first general framework for GPU-aware, compression-enabled collectives with error control, improving performance and data accuracy.

Findings

Up to 4.5X faster collective computation

Up to 28.7X faster data movement

High reconstructed data quality in applications

Abstract

GPU-aware collective communication has become a major bottleneck for modern computing platforms as GPU computing power rapidly rises. A traditional approach is to directly integrate lossy compression into GPU-aware collectives, which can lead to serious performance issues such as underutilized GPU devices and uncontrolled data distortion. In order to address these issues, in this paper, we propose gZCCL, a first-ever general framework that designs and optimizes GPU-aware, compression-enabled collectives with an accuracy-aware design to control error propagation. To validate our framework, we evaluate the performance on up to 512 NVIDIA A100 GPUs with real-world applications and datasets. Experimental results demonstrate that our gZCCL-accelerated collectives, including both collective computation (Allreduce) and collective data movement (Scatter), can outperform NCCL as well as Cray MPI by up to 4.5X and 28.7X, respectively. Furthermore, our accuracy evaluation with an image-stacking application confirms the high reconstructed data quality of our accuracy-aware framework.