Optimizing Allreduce Operations for Modern Heterogeneous Architectures with Multiple Processes per GPU

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Adams, Michael, Bienz, Amanda
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908851504152576
author Adams, Michael
Bienz, Amanda
author_facet Adams, Michael
Bienz, Amanda
contents Large inter-GPU all-reduce operations, prevalent throughout deep learning, are bottlenecked by communication costs. Emerging heterogeneous architectures are comprised of complex nodes, often containing $4$ GPUs and dozens to hundreds of CPU cores per node. Parallel applications are typically accelerated on the available GPUs, using only a single CPU core per GPU while the remaining cores sit idle. This paper presents novel optimizations to large GPU-aware all-reduce operations by extending the lane-aware algorithm to heterogeneous architectures and notably using multiple CPU cores per GPU to accelerate these operations. Using GPUDirect RDMA and host copy communications respectively, these multi-CPU-accelerated GPU-aware all-reduces yield speedups over system MPI of up to $3$x on LLNL's Tuolumne supercomputer and up to $2.45$x for large MPI all-reduces across the NVIDIA A100 GPUs of NCSA's Delta supercomputer.
format Preprint
id arxiv_https___arxiv_org_abs_2508_13397
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Optimizing Allreduce Operations for Modern Heterogeneous Architectures with Multiple Processes per GPU
Adams, Michael
Bienz, Amanda
Distributed, Parallel, and Cluster Computing
Large inter-GPU all-reduce operations, prevalent throughout deep learning, are bottlenecked by communication costs. Emerging heterogeneous architectures are comprised of complex nodes, often containing $4$ GPUs and dozens to hundreds of CPU cores per node. Parallel applications are typically accelerated on the available GPUs, using only a single CPU core per GPU while the remaining cores sit idle. This paper presents novel optimizations to large GPU-aware all-reduce operations by extending the lane-aware algorithm to heterogeneous architectures and notably using multiple CPU cores per GPU to accelerate these operations. Using GPUDirect RDMA and host copy communications respectively, these multi-CPU-accelerated GPU-aware all-reduces yield speedups over system MPI of up to $3$x on LLNL's Tuolumne supercomputer and up to $2.45$x for large MPI all-reduces across the NVIDIA A100 GPUs of NCSA's Delta supercomputer.
title Optimizing Allreduce Operations for Modern Heterogeneous Architectures with Multiple Processes per GPU
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2508.13397