.png){kind=icon}
.png){kind=icon}
Return
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}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
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.
Keywords:
Training
Graphics processing units
Computational modeling
Servers
Data models
Computer architecture
Bandwidth
Distributed computing
deep learning
computer network
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
I
IF:
5.7
Papers:
834
Citations:
3.0K