An Integrated GPU Power and Performance Mode.pdf

An Integrated GPU Power and Performance Mode.pdf

ID:34406308

大小:795.15 KB

页数:10页

时间:2019-03-05

An Integrated GPU Power and Performance Mode.pdf_第1页
An Integrated GPU Power and Performance Mode.pdf_第2页
An Integrated GPU Power and Performance Mode.pdf_第3页
An Integrated GPU Power and Performance Mode.pdf_第4页
An Integrated GPU Power and Performance Mode.pdf_第5页
资源描述:

《An Integrated GPU Power and Performance Mode.pdf》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库

1、AnIntegratedGPUPowerandPerformanceModelSunpyoHongHyesoonKimElectricalandComputerEngineeringSchoolofComputerScienceGeorgiaInstituteofTechnologyGeorgiaInstituteofTechnologyshong9@gatech.eduhyesoon@cc.gatech.eduABSTRACT1.INTRODUCTIONGPUarchitecturesareincreasin

2、glyimportantinthemulti-coreeraTheincreasingcomputingpowerofGPUsgivesthemaconsid-duetotheirhighnumberofparallelprocessors.Performanceop-erablyhigherthroughputthanthatofCPUs.Asaresult,manypro-timizationformulti-coreprocessorshasbeenachallengeforpro-grammerstry

3、touseGPUsformorethanjustgraphicsapplications.grammers.Furthermore,optimizingforpowerconsumptionisevenHowever,optimizingGPUkernelstoachievehighperformanceismoredifficult.Unfortunately,asaresultofthehighnumberofpro-stillachallenge.Furthermore,optimizinganapplic

4、ationtoachievecessors,thepowerconsumptionofmany-coreprocessorssuchasabetterpowerefficiencyisevenmoredifficult.GPUshasincreasedsignificantly.Thenumberofcoresinsideachip,especiallyinGPUs,isin-Hence,inthispaper,weproposeanintegratedpowerandperfor-creasingdramatica

5、lly.Forexample,NVIDIA’sGTX280[2]has30mance(IPP)predictionmodelforaGPUarchitecturetopredictthestreamingmultiprocessors(SMs)with240CUDAcores,andtheoptimalnumberofactiveprocessorsforagivenapplication.ThenextgenerationGPUwillhave512CUDAcores[3].Eventhoughbasicin

6、tuitionisthatwhenanapplicationreachesthepeakmem-GPUapplicationsarehighlythroughput-oriented,notallapplica-orybandwidth,usingmorecoresdoesnotresultinperformancetionsrequireallavailablecorestoachievethebestperformance.improvement.Inthisstudy,weaimtoanswerthefo

7、llowingimportantques-WedevelopanempiricalpowermodelfortheGPU.Unlikemosttions:Doweneedallcorestoachievethehighestperformance?previousmodels,whichrequiremeasuredexecutiontimes,hard-Canwesavepowerandenergybyusingfewercores?wareperformancecounters,orarchitectura

8、lsimulations,IPPpre-Figure1showsperformance,powerconsumption,andefficiencydictsexecutiontimestocalculatedynamicpowerevents.Wethen(performanceperwatt)aswevarythenumberofactivecores.1Theusetheoutco

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。