Skip to main navigation Skip to search Skip to main content

SwDNN: A Library for Accelerating Deep Learning Applications on Sunway TaihuLight

  • Jiarui Fang
  • , Haohuan Fu
  • , Wenlai Zhao
  • , Bingwei Chen
  • , Weijie Zheng
  • , Guangwen Yang
  • Tsinghua University
  • National Supercomputing Center

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

To explore the potential of training complex deep neural networks (DNNs) on other commercial chips rather than GPUs, we report our work on swDNN, which is a highly-efficient library for accelerating deep learning applications on the newly announced world-leading supercomputer, Sunway TaihuLight. Targeting SW26010 processor, we derive a performance model that guides us in the process of identifying the most suitable approach for mapping the convolutional neural networks (CNNs) onto the 260 cores within the chip. By performing a systematic optimization that explores major factors, such as organization of convolution loops, blocking techniques, register data communication schemes, as well as reordering strategies for the two pipelines of instructions, we manage to achieve a double-precision performance over 1.6 Tflops for the convolution kernel, achieving 54% of the theoretical peak. Compared with Tesla K40m with cuDNNv5, swDNN results in 1.91-9.75x performance speedup in an evaluation with over 100 parameter configurations.

Original languageEnglish
Title of host publicationProceedings - 2017 IEEE 31st International Parallel and Distributed Processing Symposium, IPDPS 2017
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages615-624
Number of pages10
ISBN (Electronic)9781538639146
DOIs
StatePublished - 30 Jun 2017
Externally publishedYes
Event31st IEEE International Parallel and Distributed Processing Symposium, IPDPS 2017 - Orlando, United States
Duration: 29 May 20172 Jun 2017

Publication series

NameProceedings - 2017 IEEE 31st International Parallel and Distributed Processing Symposium, IPDPS 2017

Conference

Conference31st IEEE International Parallel and Distributed Processing Symposium, IPDPS 2017
Country/TerritoryUnited States
CityOrlando
Period29/05/172/06/17

Keywords

  • Convolutional Neural Network
  • Deep Neural Network
  • Deep learning
  • Many-core Architecture

Fingerprint

Dive into the research topics of 'SwDNN: A Library for Accelerating Deep Learning Applications on Sunway TaihuLight'. Together they form a unique fingerprint.

Cite this