資源描述：
《GPUMLib An Efficient Open-Source GPU》由會員上傳分享，免費在線閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、InternationalJournalofComputerInformationSystemsandIndustrialManagementApplicationsISSN2150-7988Volume3(2011)pp.355-362?MIRLabs,www.mirlabs.net/ijcisim/index.htmlGPUMLib:AnEf?cientOpen-SourceGPUMachineLearningLibraryNoelLopes1andBernardeteRibeiro21UDI,PolytechnicInstituteofGuarda,P
2、ortugalCISUC,UniversityofCoimbra,Portugalnoel@ipg.pt2DepartmentofInformaticsEngineeringCISUC,UniversityofCoimbra,Portugalbribeiro@dei.uc.ptAbstract:GraphicsProcessingUnits(GPUs)placedatourdis-ciatedwithMLproblemsincreases,thetrendistohavemoreposalanunprecedentedcomputational-power,
3、largelysurpass-challengingandcomputationallydemandingproblemsthatingtheperformanceofcutting-edgeCPUs(CentralProcess-canbecomeintractablefortraditionalCPU(CentralProcess-ingUnits).Thehigh-parallelisminherenttotheGPUmakesingUnit)architectures.Therefore,thepressuretoshiftdevel-thisdev
4、iceespeciallywell-suitedtoaddressMachineLearn-opmenttowardsparallelarchitectureswithhigh-throughputing(ML)problemswithprohibitivelycomputationalintensivehasbeenaccentuated.Inthiscontext,theGraphicsProcess-tasks.Nevertheless,fewMLalgorithmshavebeenimplementedingUnit(GPU)representsac
5、ompellingsolutiontoaddressontheGPUandmostarenotopenlyshared,posingdif?cul-theincreasingneedsofcomputationalperformance,inpar-tiesforresearchersandengineersaimingtodevelopGPU-basedticularintheML?eld.systems.Tomitigatethisproblem,weproposethecreationofanInthelasteightyearstheperforma
6、nceandcapabilitiesofopensourceGPUMachineLearningLibrary(GPUMLib)thattheGPUshavebeensigni?cantlyaugmentedandtodaysaimstoprovidethebuildingblocksforthedevelopmentofef-GPUs,includedinmainstreamcomputingsystems,arepow-?cientGPUMLsoftware.Experimentalresultsonbenchmarkerful,highlyparall
7、elandprogrammabledevicesthatcanbedatasetsshowthatthealgorithmsalreadyimplementedyieldusedforgeneral-purposecomputingapplications[2].Sincesigni?canttimesavingsovertheCPUcounterparts.GPUsaredesignedforhigh-performancerenderingwhereKeywords:GPUComputing,machinelearningalgorithms.repea
8、tedoperationsarecommon,the
