GradientFlow: Optimizing Network Performance for Large-Scale Distributed DNN Training

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.

Automated summary

Scaling out deep neural network (DNN) training is crucial for reducing model training time, but high communication overhead in distributed DNN training is a major performance bottleneck. In this study, we propose GradientFlow, a communication backend, and employ various network optimization techniques to tackle this problem. By integrating methods such as ring-based allreduce, mixed-precision training, and computation/communication overlap, as well as introducing lazy allreduce and coarse-grained sparse communication, we were able to achieve impressive speedup ratios when training AlexNet and ResNet-50 on the ImageNet dataset using multiple GPUs.