How to configure your browser for maximum performance #1

We need "Choose ANGLE graphics backend" setting that should be set to OpenGL (instead of ANGLE). After changing it, click on “Relaunch Now” ... Skiptocontent {{message}} light-tracer / issues Public Notifications Fork 0 Star 12 Code Issues 7 Pullrequests 0 Actions Projects 0 Wiki Security Insights More Code Issues Pullrequests Actions Projects Wiki Security Insights Newissue Haveaquestionaboutthisproject?SignupforafreeGitHubaccounttoopenanissueandcontactitsmaintainersandthecommunity. Pickausername EmailAddress Password SignupforGitHub Byclicking“SignupforGitHub”,youagreetoourtermsofserviceand privacystatement.We’lloccasionallysendyouaccountrelatedemails. AlreadyonGitHub? Signin toyouraccount Jumptobottom Howtoconfigureyourbrowserformaximumperformance #1 Open denisbogolopenedthisissue Jun14,2019 ·12comments Open Howtoconfigureyourbrowserformaximumperformance #1 denisbogolopenedthisissue Jun14,2019 ·12comments Labels hints Usefulinformationfromauthors Comments Copylink denisbogol commented Jun14,2019 • editedbyhb3p8 LightTracer'sperformancemostlydependsonthegraphicscardused,buttherearesomeotherimportantfactors.Below,wewillbrieflydescribehowtoconfigurethebrowserformaximumperformance.Pleasenote,thatcurrently,LightTracerworksondesktopversionsofGoogleChromeandMozillaFirefoxandrequiresWebGL2.Otherplatformsarenotofficiallysupported. Uselatestgraphicsdrivers Itisimportanttokeepyourgraphicsdriversup-to-date,becausebrowsersmaynotenableWebGL2forolddriverversions. Usediscretegraphicscard(ifavailable) IfyourPChasseveralGPUs,oryourlaptopisequippedwithNVIDIAOptimustechnology,pleaseensurethatWebbrowserisrunningonthemostpowerfulgraphicscard.ThesimplestwaytoidentifytheGPUusedforrenderingistocheckthehttps://webglreport.com/.Here,inthefield"UnmaskedRenderer"youmayseecurrentlyusinggraphicscard.Also,pleaseensure,thatyourbrowsersupportsWebGL2whichisrequiredtolaunchLightTracer.IfyourbrowserusesintegratedGPUinsteadofpowerfuldiscretegraphics,pleasetrytheseinstructions: HowtorunWebGLondiscreteNvidiaGPUfornotebookswithNvidiaOptimus UsenativeOpenGL Also,sinceLightTracerisimplementedinWebGL2,itishighlyreasonabletoswitchunderlying3DAPItoOpenGL.Thiswillnotonlyimprovetheperformancebutmakeshadercompilingprocessmuchfaster. InthecaseofGoogleChrome,youcanfindChromeflagsbyjusttyping“chrome://flags”or“about://flags”intheOmnibox(address/searchbar).Once"Chromeflags"isopen,youwillseealistoffeaturesthatyoucanenableordisable.Weneed"ChooseANGLEgraphicsbackend"settingthatshouldbesettoOpenGL(insteadofANGLE).Afterchangingit,clickon“RelaunchNow”buttonthatwillsaveyourchangesandrestartChromewithyourchangesinplace. ForusingOpenGLinFirefoxyouneedtochangetheruntimeoption.Type"about:config"intotheaddressbar.InrecentversionsofFirefox,find"webgl.disable-angle"variable.Bydefault,itissetto"false":nativeOpenGLisnotused.Setitto"true"tostartusingOpenGL(browserrestartisnotneeded). Thetextwasupdatedsuccessfully,buttheseerrorswereencountered: 👍 1 hatunaareactedwiththumbsupemoji 🎉 1 julianxhokaxhiureactedwithhoorayemoji hb3p8 pinnedthisissue Jun14,2019 Copylink stefkeB commented Jun17,2019 IsthissupposedtoalsoworkinSafarionmacOS?IcanactivateWebGL2asanexperimentalfeatureandconfirmthisfromtheWebGLwebsite,buttonoavail.Onlyablackscreen. Sorry,somethingwentwrong. Copylink Author denisbogol commented Jun17,2019 Indeed,itissupposedtoworkwithenabledWebGL2.0inSafari,buttothemomentwewerenotabletoidentify&fixtheactualproblem.Asatemporarysolution,youmayuseChromeonMacOS. Sorry,somethingwentwrong. hb3p8 added the hints Usefulinformationfromauthors label Jun19,2019 Copylink ArixCuirax commented Jun27,2019 • edited HelloDenisandDanila,0.9.3doesn'tseemtoworkinChromeforme.It'sshowingablackscreen.ItdoesseemtoworkinFirefox.RunningaVega64onWindows1064bit1903.AllmypreviousrenderswereinChromeon0.9.2. ItseemstoworkfineinFirefoxthough(neveruseditbeforeIstartedgettingtheblackscreeninChrome).ThereisaslightchancethismaybeduetomycustomisedGPUdriversettingsand/orapplicationprofileforChromewhichIjustsetupinRadeonSettings.I'llupdateifImanagetorulethatoutasacause. BTW,herearemydriversettings: Ihavetemporarilyde-activatedtheChromeapplicationprofileandIwillalsobeupdatingmydrivershortlyandrebootingaswell.ThenI'lltryrevertingChromesettingstodefaulttoseeifthathelps. Also,here'swhatI'mgettinginChromeafterupdatingto0.9.3: Sorry,somethingwentwrong. Copylink ArixCuirax commented Jun27,2019 Verypeculiar.SeemsIjustneededtoCtrl+F5.Updatingthevideodriverto19.6.3didn'tfixtheissue.NeitherdiddisablingthecustomprofileI'dmadeforChromeinRadeonsettings.LTalsoworkedfineinOpera.IthentriedCtrl+F5inChromeandnowon0.9.3andLTdisplaysagain. Sorry,somethingwentwrong. Copylink Author denisbogol commented Jun28,2019 • edited Hi@ArixCuirax, IfIunderstoodcorrectly,v0.9.3worksnormallyafterCtrl+F5(clearingthecachedversionofthepage)? Thankyouforreportingthisissue.Beforenextrelease,wewilltrytofindsomewaytonotifythebrowserthatcachedversionisoutdatedandcannotbere-used. Pleaseletusknowifyounoticeanyotherissues. Sorry,somethingwentwrong. Copylink ArixCuirax commented Jul3,2019 Meagain,Ihaveafewquestions. WhatdoestheRNcoherencysettingdo?Settingitto0seemedtoproducebetterrendersquicker. WhatdoestheUnitoflengthsettingdo?Doesitplayaroleinscalingthemodelasfarasthelighttravelpathisconcerned? HowdoesLightTracerbuildtherenderedimage?Doesitaveragetheresultsofallsamplesforagivenscreenpixelovertime?Ifso,doesthismeanthattherenderimprovementisasymptotic-itimprovesquicklyatfirstandmoreandmoreslowlyovertime(becauseeveryadditionalsamplehasaneverdecreasingcontributiontotheaveragevaluewhichhasalreadybeenachievedforthatpixel)? Ifso,couldanadditionalsettingbeaddedwhichcontrolshowmuchweightagivennewsample'scontributiontotheexistingcolouraverageofapixel'scolouris?Andifwewantshorterrendertimeswecanmakethecontributionofeachnewsampletotheexistingpixelcolouraveragelargerandifwewantmoreaccurateendresultswecansetthecontributionofeachnewsampletothepixel'sexistingaveragedcolourtobesmaller? Sorry,somethingwentwrong. Copylink Author denisbogol commented Jul3,2019 Hi@ArixCuirax, Infact,0coherencyleadstomoresmoothimageswithmore"natural"noise.But,renderingperformance(i.e.FPS)willbelowerwithsuchaconfig(nottoomuch,maybe~30%).Ifcoherencyishigh(closeto1),raypathshavebecomemorecoherent(similar)whichisgoodforGPUandleadtohigherperformance.But,inbothcases,thefinalimage(whenthereisnovisiblenoise)isthesame.So,pleaseadjustitaccordingtoyourpreferences. UnitoflengthisneededtotreatparametersofCameraLens(e.g.focallength,whichisdefinedinmm).So,theunitoflengthspecifiesunitsusedbyscene(yourmodel),andthusitdefineshowweshouldre-computefocallengthtosceneunits.Changingthissettingdoesnotleadtoanyrecomputationsofgeometry,acceleratingstructures,etc.Itonlyaffectscameratreatment. Youareabsolutelyright,itusesMonte-Carlointegrationtoestimatethecolorofeachpixel(itisaquitecommonapproachforphoto-realisticrenders).Andtheerrorintheimage(noise)decreasesproportionallyto...sqrt("numberofsamples").So,todecreasenoisein2xyouneed4xmoresamples.Inordertominimizecomputationtime,wetrytousesomesmartimportancesamplingtechniqueswheneverpossible. Infact,yourproposalregardingcontrolofminimalcontributionisverynaturalandalreadyplannedforthenextversion. Sorry,somethingwentwrong. Copylink ArixCuirax commented Jul3,2019 Thankyouforyourcomprehensiveanddetailedexplanation.Iappreciateyoutakingthetimetoprovidethisinformation.It'sfascinatingandgivesinsightintohowyourpathtracerworks. IpreferleavingtheRNGcoherencysetto0becausetheimageseemstobebecomenicerquickereventhoughreadingyourexplanationInowunderstandthisisactuallyslowingrendertimes.Thenoisejustseemstobemoredifficulttospot.Iwillbeusingitfromnowon. Also,verynicetohearthatcontrolofminimumcontributionisalreadyaplannedfeatureofthenextversion.Ireallyappreciateyourexcellentwork,thankyou! Sorry,somethingwentwrong. Copylink ArixCuirax commented Jul3,2019 CanwepleasealsohaveRendersettingssavedtoacookieinafutureversionaswell?:) Sorry,somethingwentwrong. Copylink ArixCuirax commented Jul5,2019 Actually,thatmightnotbeagoodideaafterall,atleastforme,becauseIalwaysturnoffthetilebasedrenderingandgenerallyuse32bouncesandeverythinggraphicsrelatedbecomesquitechuggyandskippy. That'sokwhileperformingtheactualrenderbutnotsomethingI'dliketohavestoredinacookieandappliedautomaticallywhenIfirstopentherendererandhaven'tyetimportedmymodel,createdafloororassignedmaterials. Unlesstherewasalsoapausefunctionorbuttonwhichdefaultedtoonwhenyoufirstlaunchtherenderersoastoallowtheusertosetupthesceneinrasterizationmodeonlyrenderingbeforethegpustartsbeingputtoworkactuallypathtracing. Sorry,somethingwentwrong. Copylink necros2k7 commented Sep3,2021 WhyOpenGLisfasterinyouropinion?Isn`tDirectXnativetoWindowssoshouldbefaster?HowtobenchmarkAPIperformanceinbrowser? Sorry,somethingwentwrong. Copylink glenn-slayden commented Jan18,2022 ping@denisbogol...? ...Weneed"ChooseANGLEgraphicsbackend"settingthatshouldbesettoOpenGL(insteadofANGLE)... @necros2k7wrote: WhyOpenGLisfasterinyouropinion?Isn`tDirectXnativetoWindowssoshouldbefaster? Sorry,somethingwentwrong. Signupforfree tojointhisconversationonGitHub. Alreadyhaveanaccount? Signintocomment Assignees Nooneassigned Labels hints Usefulinformationfromauthors Projects Noneyet Milestone Nomilestone Development Nobranchesorpullrequests 6participants Youcan’tperformthatactionatthistime. Yousignedinwithanothertaborwindow.Reloadtorefreshyoursession. Yousignedoutinanothertaborwindow.Reloadtorefreshyoursession.