.png){kind=icon}
返回
GradientFlow: Optimizing Network Performance for Large-Scale Distributed DNN Training
.png){kind=icon}
PRE
AI
DOI：10.1109/TBDATA.2019.2957478.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
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 AlexNet and ResNet-50 on the ImageNet dataset using 512 GPUs, our approach could achieve 410.2 and 434.1 speedup ratio, respectively.
Keyword:
Training
Graphics processing units
Computational modeling
Servers
Data models
Computer architecture
Bandwidth
Distributed computing
deep learning
computer network
AI总结
对已上传原文的论文进行重点信息的提取，主要内容包括：简要概述、研究摘要、背景介绍、关键亮点、图文解析、展望与总结。
期刊
I
IF:
5.7
论文数:
887
被引数:
3.0K
机构
引用论文
Shape Control Synthesis of CuPt Alloys with Enhanced Hydrogen Evolution Reaction and Methanol Oxidation Reaction Activities.png)
ChemNanoMat
ChemNanoMatIF0
PREAI
分享收藏
查看更多内容