Optimizing Network Performance for Distributed DNN Training on GPU Clusters: ImageNet/AlexNet Training in 1.5 Minutes

TL;DR

This paper introduces GradientFlow, a network optimization framework that significantly accelerates distributed DNN training on GPU clusters, achieving unprecedented speedups for ImageNet/AlexNet training.

Contribution

The paper presents GradientFlow, a novel communication backend with innovative techniques like lazy allreduce and sparse communication, enabling faster distributed training.

Findings

Achieved 410.2x speedup on 512 GPUs for ImageNet/AlexNet.

Completed 95 epochs of training in 1.5 minutes.

Outperformed existing distributed training methods.

Abstract

It is important to scale out deep neural network (DNN) training for reducing model training time. The high communication overhead is one of the major performance bottlenecks for distributed DNN training across multiple GPUs. Our investigations have shown that popular open-source DNN systems could only achieve 2.5 speedup ratio on 64 GPUs connected by 56 Gbps network. To address this problem, we propose a communication backend named GradientFlow for distributed DNN training, and employ a set of network optimization techniques. First, we integrate ring-based allreduce, mixed-precision training, and computation/communication overlap into GradientFlow. Second, we propose lazy allreduce to improve network throughput by fusing multiple communication operations into a single one, and design coarse-grained sparse communication to reduce network traffic by only transmitting important gradient chunks. When training ImageNet/AlexNet on 512 GPUs, our approach achieves 410.2 speedup ratio and completes 95-epoch training in 1.5 minutes, which outperforms existing approaches.