Analysis of Hardware Prefetchers Suitable for CNN Applications

Hyeong Gi Seong, Hyokeun Lee, Hyun Kim, Hyuk Jae Lee

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

Abstract

Employing CPUs for convolutional neural networks (CNNs) is of interest to many users due to its availability. However, to make CPUs more competitive, it is necessary to bridge the performance gap between CPUs and other accelerators such as GPUs and TPUs. Date prefetching is the promising optimization technique because CNN operations tend to have predictable data access patterns. This paper examines and analyzes the usefulness of various hardware prefetchers in CNN applications.

Original languageEnglish
Title of host publication2021 IEEE International Conference on Consumer Electronics-Asia, ICCE-Asia 2021
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781665408578
DOIs
StatePublished - 2021
Event2021 IEEE International Conference on Consumer Electronics-Asia, ICCE-Asia 2021 - Gangwon, Korea, Republic of
Duration: 1 Nov 20213 Nov 2021

Publication series

Name2021 IEEE International Conference on Consumer Electronics-Asia, ICCE-Asia 2021

Conference

Conference2021 IEEE International Conference on Consumer Electronics-Asia, ICCE-Asia 2021
Country/TerritoryKorea, Republic of
CityGangwon
Period1/11/213/11/21

Keywords

  • Convolutional Neural Network
  • Hardware Prefetcher
  • McSimA+

Fingerprint

Dive into the research topics of 'Analysis of Hardware Prefetchers Suitable for CNN Applications'. Together they form a unique fingerprint.

Cite this