移动CNN加速器计算更接近数据

Sumanth Gudaparthi, Surya Narayanan, R. Balasubramonian
{"title":"移动CNN加速器计算更接近数据","authors":"Sumanth Gudaparthi, Surya Narayanan, R. Balasubramonian","doi":"10.1109/EMC2.2018.00015","DOIUrl":null,"url":null,"abstract":"A significant fraction of energy in recent CNN accelerators is dissipated in moving operands between storage and compute units. In this work, we re-purpose the CPU's last level cache to perform in-situ dot-product computations, thus significantly reducing data movement. Since a last level cache has several subarrays, many such dot-products can be performed in parallel, thus boosting throughput as well. The in-situ operation does not require analog circuits; it is performed with a bit-wise AND of two subarray rows, followed by digital aggregation of partial sums. The proposed architecture yields a 2.74× improvement in throughput and a 6.31× improvement in energy, relative to a DaDianNao baseline. This is primarily because the proposed architecture eliminates a large fraction of data transfers over H-Tree interconnects in the cache.","PeriodicalId":377872,"journal":{"name":"2018 1st Workshop on Energy Efficient Machine Learning and Cognitive Computing for Embedded Applications (EMC2)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2018-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":"{\"title\":\"Moving CNN Accelerator Computations Closer to Data\",\"authors\":\"Sumanth Gudaparthi, Surya Narayanan, R. Balasubramonian\",\"doi\":\"10.1109/EMC2.2018.00015\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"A significant fraction of energy in recent CNN accelerators is dissipated in moving operands between storage and compute units. In this work, we re-purpose the CPU's last level cache to perform in-situ dot-product computations, thus significantly reducing data movement. Since a last level cache has several subarrays, many such dot-products can be performed in parallel, thus boosting throughput as well. The in-situ operation does not require analog circuits; it is performed with a bit-wise AND of two subarray rows, followed by digital aggregation of partial sums. The proposed architecture yields a 2.74× improvement in throughput and a 6.31× improvement in energy, relative to a DaDianNao baseline. This is primarily because the proposed architecture eliminates a large fraction of data transfers over H-Tree interconnects in the cache.\",\"PeriodicalId\":377872,\"journal\":{\"name\":\"2018 1st Workshop on Energy Efficient Machine Learning and Cognitive Computing for Embedded Applications (EMC2)\",\"volume\":\"1 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2018-03-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"1\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2018 1st Workshop on Energy Efficient Machine Learning and Cognitive Computing for Embedded Applications (EMC2)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/EMC2.2018.00015\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2018 1st Workshop on Energy Efficient Machine Learning and Cognitive Computing for Embedded Applications (EMC2)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/EMC2.2018.00015","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 1

摘要

在最近的CNN加速器中,很大一部分能量消耗在存储和计算单元之间的操作数移动上。在这项工作中,我们重新利用CPU的最后一级缓存来执行原位点积计算,从而显着减少数据移动。由于最后一级缓存有几个子数组,因此许多这样的点积可以并行执行,从而也提高了吞吐量。现场操作不需要模拟电路;它是通过对两个子数组行进行逐位与运算来执行的,然后是部分和的数字聚合。与DaDianNao基线相比,该架构的吞吐量提高了2.74倍,能耗提高了6.31倍。这主要是因为所提出的架构消除了高速缓存中H-Tree互连上的大部分数据传输。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
查看原文
分享 分享
微信好友 朋友圈 QQ好友 复制链接
本刊更多论文
Moving CNN Accelerator Computations Closer to Data
A significant fraction of energy in recent CNN accelerators is dissipated in moving operands between storage and compute units. In this work, we re-purpose the CPU's last level cache to perform in-situ dot-product computations, thus significantly reducing data movement. Since a last level cache has several subarrays, many such dot-products can be performed in parallel, thus boosting throughput as well. The in-situ operation does not require analog circuits; it is performed with a bit-wise AND of two subarray rows, followed by digital aggregation of partial sums. The proposed architecture yields a 2.74× improvement in throughput and a 6.31× improvement in energy, relative to a DaDianNao baseline. This is primarily because the proposed architecture eliminates a large fraction of data transfers over H-Tree interconnects in the cache.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
相关文献
二甲双胍通过HDAC6和FoxO3a转录调控肌肉生长抑制素诱导肌肉萎缩
IF 8.9 1区 医学Journal of Cachexia, Sarcopenia and MusclePub Date : 2021-11-02 DOI: 10.1002/jcsm.12833
Min Ju Kang, Ji Wook Moon, Jung Ok Lee, Ji Hae Kim, Eun Jeong Jung, Su Jin Kim, Joo Yeon Oh, Sang Woo Wu, Pu Reum Lee, Sun Hwa Park, Hyeon Soo Kim
具有疾病敏感单倍型的非亲属供体脐带血移植后的1型糖尿病
IF 3.2 3区 医学Journal of Diabetes InvestigationPub Date : 2022-11-02 DOI: 10.1111/jdi.13939
Kensuke Matsumoto, Taisuke Matsuyama, Ritsu Sumiyoshi, Matsuo Takuji, Tadashi Yamamoto, Ryosuke Shirasaki, Haruko Tashiro
封面:蛋白质组学分析确定IRSp53和fastin是PRV输出和直接细胞-细胞传播的关键
IF 3.4 4区 生物学ProteomicsPub Date : 2019-12-02 DOI: 10.1002/pmic.201970201
Fei-Long Yu, Huan Miao, Jinjin Xia, Fan Jia, Huadong Wang, Fuqiang Xu, Lin Guo
来源期刊
自引率
0.00%
发文量
0
期刊最新文献
Deep Learning Inference on Embedded Devices: Fixed-Point vs Posit Event Prediction in Processors Using Deep Temporal Models Invited Talk Abstract: Introducing ReQuEST: An Open Platform for Reproducible and Quality-Efficient Systems-ML Tournaments A High Efficiency Accelerator for Deep Neural Networks A Quantization-Friendly Separable Convolution for MobileNets
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
现在去查看 取消
×
提示
确定
0
微信
客服QQ
Book学术公众号 扫码关注我们
反馈
×
意见反馈
请填写您的意见或建议
请填写您的手机或邮箱
已复制链接
已复制链接
快去分享给好友吧!
我知道了
×
扫码分享
扫码分享
Book学术官方微信
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术
文献互助 智能选刊 最新文献 互助须知 联系我们:info@booksci.cn
Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。
Copyright © 2023 Book学术 All rights reserved.
ghs 京公网安备 11010802042870号 京ICP备2023020795号-1