跳到主要导航 跳到搜索 跳到主要内容

FlexFlow: A Flexible Dataflow Accelerator Architecture for Convolutional Neural Networks

  • Wenyan Lu
  • , Guihai Yan*
  • , Jiajun Li
  • , Shijun Gong
  • , Yinhe Han
  • , Xiaowei Li
  • *此作品的通讯作者
  • University of Chinese Academy of Sciences
  • CAS - Institute of Computing Technology

科研成果: 书/报告/会议事项章节会议稿件同行评审

摘要

Convolutional Neural Networks (CNN) are verycomputation-intensive. Recently, a lot of CNN accelerators based on the CNN intrinsic parallelism are proposed. However, we observed that there is a big mismatch between the parallel types supported by computing engine and the dominant parallel types of CNN workloads. This mismatch seriously degrades resource utilization of existing accelerators. In this paper, we propose aflexible dataflow architecture (FlexFlow) that can leverage the complementary effects among feature map, neuron, and synapse parallelism to mitigate the mismatch. We evaluated our design with six typical practical workloads, it acquires 2-10x performance speedup and 2.5-10x power efficiency improvement compared with three state-of-the-art accelerator architectures. Meanwhile, FlexFlow is highly scalable with growing computing engine scale.

源语言英语
主期刊名Proceedings - 2017 IEEE 23rd Symposium on High Performance Computer Architecture, HPCA 2017
出版商IEEE Computer Society
553-564
页数12
ISBN(电子版)9781509049851
DOI
出版状态已出版 - 5 5月 2017
已对外发布
活动23rd IEEE Symposium on High Performance Computer Architecture, HPCA 2017 - Austin, 美国
期限: 4 2月 20178 2月 2017

出版系列

姓名Proceedings - International Symposium on High-Performance Computer Architecture
ISSN(印刷版)1530-0897

会议

会议23rd IEEE Symposium on High Performance Computer Architecture, HPCA 2017
国家/地区美国
Austin
时期4/02/178/02/17

指纹

探究 'FlexFlow: A Flexible Dataflow Accelerator Architecture for Convolutional Neural Networks' 的科研主题。它们共同构成独一无二的指纹。

引用此