TY - GEN
T1 - FlexFlow
T2 - 23rd IEEE Symposium on High Performance Computer Architecture, HPCA 2017
AU - Lu, Wenyan
AU - Yan, Guihai
AU - Li, Jiajun
AU - Gong, Shijun
AU - Han, Yinhe
AU - Li, Xiaowei
N1 - Publisher Copyright:
© 2017 IEEE.
PY - 2017/5/5
Y1 - 2017/5/5
N2 - 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.
AB - 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.
KW - Accelerator
KW - Complementary Effect
KW - Convolutional Neural Networks
KW - Flexible Dataflow
UR - https://www.scopus.com/pages/publications/85019607855
U2 - 10.1109/HPCA.2017.29
DO - 10.1109/HPCA.2017.29
M3 - 会议稿件
AN - SCOPUS:85019607855
T3 - Proceedings - International Symposium on High-Performance Computer Architecture
SP - 553
EP - 564
BT - Proceedings - 2017 IEEE 23rd Symposium on High Performance Computer Architecture, HPCA 2017
PB - IEEE Computer Society
Y2 - 4 February 2017 through 8 February 2017
ER -