M. Sawyer
{"title":"Advancing","authors":"M. Sawyer","doi":"10.4324/9780429493386-60","DOIUrl":null,"url":null,"abstract":"— Theslow-downoftechnologyscalingcombinedwiththeexponential growthofmodernmachinelearningandartificialintelligencemodelshascreateda demandforspecializedaccelerators,suchasGPUs,ASICs,andfield-programmable gatearrays(FPGAs).FPGAscanbereconfiguredandhavethepotentialto outperformotheraccelerators,whilealsobeingmoreenergy-efficient,butare cumbersometousewithtoday’sfracturedlandscapeoftoolflows.Weproposethe conceptofanoperationsetarchitecturetoovercomethecurrentincompatibilitiesand hurdlesinusingDNN-to-FPGAcompilersbycombiningexistingspecialized frameworksintooneorganiccompiler thatalsoallowstheefficientandautomaticre- useofexistingcommunitytools.Furthermore,wedemonstratethatmixingdifferent existingframeworkscanincreasetheefficiencybymorethananorderofmagnitude.","PeriodicalId":132093,"journal":{"name":"Unorthodox Strategies for the Everyday Warrior","volume":"7 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2020-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"93","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Unorthodox Strategies for the Everyday Warrior","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.4324/9780429493386-60","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 93
推进
-技术规模的放缓与现代机器学习和人工智能模型的指数级增长相结合,创造了对专用加速器的需求,如gpu, asic和现场可编程网关(fpga)。fpga可以配置并具有超越加速器的潜力,同时也更节能,但在当今压裂工具流的环境下,它显得过于笨重。我们提出了fanoperationsearchitecture的概念,通过将现有的专门框架组合成一个有机编译器,从而克服当前使用dnn到fpga编译器的兼容性和障碍,该编译器还允许有效和自动地使用现有的社区工具。此外,wedemonstratethatmixingdifferent existingframeworkscanincreasetheefficiencybymorethananorderofmagnitude。
本文章由计算机程序翻译,如有差异,请以英文原文为准。