International Journal of Green Computing Volume 9 • Issue 1 • January-June 2018 Ecological Impact of Green Computing Using Graphical Processing Units in Molecular Dynamics Simulations Izabele Marquetti, North Carolina A&T State University, USA Jhonatam Rodrigues, North Carolina A&T State University, USA Salil S. Desai, North Carolina A&T State University, USA ABSTRACT Moleculardynamics(MD)modelsrequirecomprehensivecomputationalpowertosimulatenanoscale phenomena.Traditionally,centralprocessingunit(CPU)clustershavebeenthestandardmethodof performingthesenumericallyintensivecomputations.Thisarticleinvestigatestheuseofgraphical processingunits(GPUs)toimplementlarge-scaleMDmodelsforexploringnanofluidic-substrate interactions.MDmodelsofwaternanodropletsoverflatsiliconsubstratearetrackedwhereinthe simulationattainsasteadystatecomputationalperformance.DifferentclassesofGPUunitsfrom NVIDIA(C2050,K20,andK40)areevaluatedforenergyefficiencyperformancewithrespecttothree .greencomputingmeasures:simulationcompletiontime,powerconsumption,andCO2emissions .TheCPU+K40configurationdisplayedthelowestenergyconsumptionprofileforallthemeasures ThisresearchdemonstratestheuseofenergyefficientgraphicalcomputingversustraditionalCPU .computingforhigh-performancemoleculardynamicssimulations K EywoRDS Graphical Processing Units, Green Computing, High Performance Computation, Hybrid 3D Printing, Industry 4.0, Molecular Dynamics, Nanomanufacturing INTR oDUCTIoN Green Computing Computingtechnologiesareusuallyassociatedwithhighenergyconsumption.Thus,theenergy efficiencyofhardwareandoperatingsystemshasbecomeafinancialandenvironmentalconcern Zhu,Sun,&Hu,2012).IntheUnitedStates,desktopcomputerscanrepresentaround10%ofthe) consumptionofthecommercialelectricity.Also,high-performancecomputersrequireapowerful ,coolingsystemforheatdissipation,therebyexacerbatingtheenergyconsumption.Consequently computerscanproducecarbondioxideequivalenttomillionsofcars(Li,2012).Thechoiceofsoftware alsoinfluencestheenergyconsumptionprofileandneedstobeevaluatedforpoweroptimization purposes.Therefore,greenpracticesarenecessarytobeincorporatedintothedesignandoperation DOI: 10.4018/IJGC.2018010103  .Copyright©2018,IGIGlobal.CopyingordistributinginprintorelectronicformswithoutwrittenpermissionofIGIGlobalisprohibited  35 International Journal of Green Computing Volume 9 • Issue 1 • January-June 2018 of systems.Greencomputingisarecentapproachwhichallowscomputersystemstobeenergy efficientbyimplementingnewertechnologiestoreducepowerconsumptionwithoutcompromising ,.theirperformance(Binder&Suri,2009;Singh,2015;Zhou,Xiao,Yang,&Gong,2016;Zhuetal Inaddition,toimprovetheefficiencyandeconomyofcomputersystems,greencomputingalso.(2012 considersthereductionofcomputersystemsimpactontheenvironmentandpeople.Theseinclude theachievementofenergysavingsandenvironmentalprotection,whichcanimpactthelifecycleof computersystems(Zhang,Gong,&Li,2011).HarmonandAuseklis(2009)definegreencomputing ,asimprovingtheefficiencyofcomputingresourceswhileminimizingtheenvironmentalimpact consideringtheentirelifecycleofaproduct.Thiscanbeachievedbyreducingtheconsumptionof water,energy,limitedresourcesandavoidingtheuseofhazardousmaterials,therebyminimizingall .wasteintheentiresupplychain Thepracticeofgreencomputinginvolvesthedesign,manufacture,useandproperdisposalof computersandallsubsystems,suchasprinters,monitors,andstoragedevices,inanefficientand .(effectivewaythatcausesminimalimpacttotheenvironment(Velte,Velte,&Elsenpeter,2008 Thesecanbeachievedinourdailyroutinewithsimpleactions,suchasswitchingoffthecomputer ,(whennotbeingused,choosingcomputermonitorsthataremoreenergyefficient(LEDbacklighting keepingthecomputerwell-maintainedandoptimized,disablingunnecessaryprogramsthatstart ,automatically,recyclingcomputerparts,batteriesandprintercartridge(Appasami&SureshJoseph .(Chakraborty,Bhattacharyya,Nargiza,&Bedajna,2009;2011 Greencomputingcanbeconsideredintermsofbothsoftwareandhardware.Insoftware,its objectiveistoobtainmethodsthatincreasetheefficiencyofprograms,decreasingstoragespace -andsavingenergy.SomeexamplesincludeCloudComputing,DistributedComputing,andHigh PerformanceComputing.Forhardware,greencomputingincludescertaintechnologiesthatenable asignificantreductionintheenergyconsumptionandtheemissionsfootprint,whileincreasing .(economicefficiencyandfacilitatingrecycling(Zhangetal.,2011 Graphical Processing Units Graphicsprocessingunits(GPUs)haveemergedinthepastfewyearsasanalternativetoCentral ProcessorUnits(CPUs)duetotheirpowerfulcapacityforapplicationsthatdemandhighcomputational (resources.GPUshaveadvancedfromafixed-functionprocessorwithgreatthree-dimensional(3D ,(graphicstoapowerfulprogrammableprocessorwithapplicationprogramminginterface(API resultinginprocessorswithhigharithmeticcapacity.Theseprocessorunitsarenowdesignedfor largeandparallelcomputationrequirementswithafocusonthroughputinsteadoflatency,having differentapplicationsthanaCPUarchitecture(Owensetal.,2008).Owensetal.(2008)emphasizethe importanceofGPUsforscientificcomputingapplicationstosolvelargeandcomplexcomputational problems,includingproteinsimulations,scalablemoleculardynamicssimulations,andcalculations .ofelectrostaticpotentialmaps Inhigh-performancecomputing,energyconsumption(Janetal.,2012)isanimportantfactor .becausesignificantcomputingpowerisrequiredtosolveintensivecalculationsandexecutesimulations Inadditiontothat,itispossibletoreducetheamountoftimetoperformanoperationandincreasethe productivity(Vijayaraghavan,Garg,Vijayaraghavan,&Gao,2015).Studiesinthegreencomputing fieldshowthatalargeamountofenergycanbesavedbyfocusingonenvironmentalprotectionand conservationofenergy(NRC,2010).Zhangetal.(2011)usedthreecomputingsystemstoevaluate parameterssuchasgreen-relatedindexestoimprovethegreencomputingsystems.Singh,Naik,and Mahinthan(2015)demonstratedthatthechoiceofenergy-efficientAPIwithoptimalparametershave asignificantimpactonenergysavingsfordifferentservers.Asanexample,itwaspossibletosaveup to76%energyduringafilereadingroutine.Cecilia(2013)presentedGPUimplementationsoftwo differentnature-inspiredoptimizationmethodstovalidatehardwareenhancementsusingNvidia’s Fermiarchitecture.Picariello,Rapuano,andVillano(2013)havedevelopedameasurementsystem andtest-bencharchitecturetomeasuretheinstantaneouspowerconsumptionfromthepowerlineof 36 International Journal of Green Computing Volume 9 • Issue 1 • January-June 2018 a workstationmotherboard.MarquettiandDesai(2016)investigatedtheinfluenceofGPUunitson theperformanceofdifferentbiomolecularmoleculardynamics(MD)modelsbenchmarkingdifferent .GPUswithCPUconfigurations Investigating Nanoscale Processes Through Molecular Simulations Severaldirect-writeprocessessuchaselectrosprayingandinkjetmethodsutilizedropletsrangingfrom .(nanotomicroscaledimensions(Jaworek,2007;Jaworek&Sobczyk,2008;Kullmannetal.,2012 Atomizeddropletsofliquidsatthenanoscalearedepositedondifferentsubstratematerialstodirectly writethree-dimensionalfeatures(Hon,Li,&Hutchings,2008).Inrecentyears,nanoscaledropletsare beingemployedtobuildcomplexgeometriesofmoleculesforbothbiomedical,sensorandelectronic applications(Buck,Xu,Brasuel,Philbert,&Kopelman,2004;DellaRocca,Liu,&Lin,2011;Ko etal.,2007,Marquetti&Desai,2018a,2018b).Animportantaspectofthisdepositionprocessis studyingtheinterfacialpropertiesbetweennanoscaleliquidswithavarietyofsubstratematerials Desai&Kaware,2012;Yang,2006).Continuummodelssuchascomputationalfluiddynamicshave) yieldedgoodresultsformicroscaledroplets.However,theyfailtogiveadetailedunderstandingof .(thecomplexfluid-structurephenomenonatnanoscaleregimes(Curtin&Miller,2003;Schatz,2007 Moleculardynamicsmodelingisapotentstrategytoexploreatomic-levelbehaviorofliquidsunder .(differentambientconditions(Ritos,Dongari,Borg,Zhang,&Reese,2013 Computational Complexity of Molecular Models Moleculardynamicsmodelsrequireenormouscomputationalpowertosimulatethousandsofatoms tomimicrealisticphenomenon.Inaddition,thetimestepsneededforthesesimulationsareextremely smallrangingupto1fs(10e-15h s)whic resultinlongerruntimes.Theselarge-scalecomputations .demandparallelcomputingarchitectureswhereeachtimestepcanbecompletedinfewmilliseconds Moleculardynamicsimulationsinvolvecomputingbothnon-bondedandbondedpotentialfunctions forpairwiseinteractionsamongatoms.Thenon-bondedinteractionsconsumethehigherfraction ofcomputationpoweroverthebondedenergycalculations.Non-bondedinteractionsaregeometry dependent which complicates