Masu bincike daga Jami'ar Ben-Gurion (Isra'ila), Jami'ar Lille (Faransa) da Jami'ar Adelaide (Ostiraliya) sun kirkiro sabuwar dabara don gano na'urorin masu amfani ta hanyar gano sigogin aiki na GPU a cikin mai binciken gidan yanar gizo. Ana kiran hanyar “Drawn Apart” kuma ta dogara ne akan amfani da WebGL don samun bayanan aikin GPU, wanda zai iya inganta daidaiton hanyoyin bin diddigin abubuwan da ke aiki ba tare da amfani da Kukis ba kuma ba tare da adana mai ganowa akan tsarin mai amfani ba.
Hanyoyin da ke la'akari da fasalulluka na nunawa, GPU, graphics stack da direbobi lokacin da aka gano an yi amfani da su a baya, amma an iyakance su ga ikon raba na'urori kawai a matakin nau'i daban-daban na katunan bidiyo da GPUs, watau. za a iya amfani da shi azaman ƙarin abu kawai don ƙara yuwuwar ganowa. Babban fasalin sabuwar hanyar "Drawn Apart" ita ce, ba ta iyakance kanta ga rarraba nau'ikan GPU daban-daban ba, amma tana ƙoƙarin gano bambance-bambance tsakanin GPUs iri ɗaya na ƙirar iri ɗaya saboda bambancin tsarin kera kwakwalwan kwamfuta da aka tsara don daidaitawa da yawa. kwamfuta. An lura cewa bambance-bambancen da ke tasowa yayin aikin samarwa yana ba da damar ƙirƙirar simintin da ba maimaituwa ba don ƙirar na'ura iri ɗaya.
Ya bayyana cewa ana iya gano waɗannan bambance-bambance ta hanyar kirga adadin raka'o'in kisa da kuma nazarin ayyukansu a cikin GPU. An yi amfani da cak bisa saitin ayyukan trigonometric, ayyuka masu ma'ana da lissafin maki masu iyo a matsayin na farko don gano nau'ikan GPU daban-daban. Don gano bambance-bambance a cikin GPUs iri ɗaya, an ƙididdige adadin aiwatar da zaren a lokaci guda yayin aiwatar da inuwar inuwa. Ana tsammanin cewa tasirin da aka gano yana haifar da bambance-bambance a yanayin zafin jiki da kuma amfani da wutar lantarki na nau'ikan kwakwalwan kwamfuta daban-daban (a baya, an nuna irin wannan tasirin ga CPUs - masu sarrafawa iri ɗaya sun nuna yawan amfani da wutar lantarki yayin aiwatar da lambar guda).
Saboda ana yin ayyuka ta hanyar WebGL ba tare da an daidaita su ba, aikin JavaScript API ɗin.now() ba za a iya amfani da shi kai tsaye don auna lokacin aiwatar da su ba, don haka an gabatar da dabaru uku don auna lokacin:
- kan allo - yana nuna wurin a cikin zanen HTML, yana auna lokacin amsawar aikin sake kira, wanda aka saita ta Window.requestAnimationFrame API kuma ana kira bayan an gama nunawa.
- Offscreen - ta amfani da ma'aikaci da mayar da wurin zuwa wani abu na OffscreenCanvas, yana auna lokacin aiwatar da umarnin convertToBlob.
- GPU - Zana zuwa wani abu na OffscreenCanvas, amma yi amfani da mai ƙidayar lokaci ta WebGL don auna lokacin da ke la'akari da tsawon saƙon umarni a gefen GPU.
Yayin aikin ƙirƙirar ID, ana yin gwaje-gwaje 50 akan kowace na'ura, kowanne yana rufe ma'auni 176 na halaye 16 daban-daban. Gwajin da ya tattara bayanai akan na'urori 2500 tare da 1605 GPUs daban-daban ya nuna haɓakar 67% a cikin ingantaccen hanyoyin gano hanyoyin haɗin gwiwa lokacin ƙara tallafin Drawn Apart. Musamman, hanyar haɗin FP-STALKER ta ba da ganewa a cikin kwanaki 17.5 akan matsakaita, kuma lokacin da aka haɗa tare da Drawn Apart, tsawon lokacin ganowa ya ƙaru zuwa kwanaki 28.
- Daidaiton rabuwar tsarin 10 tare da kwakwalwan kwamfuta na Intel i5-3470 (GEN 3 Ivy Bridge) da Intel HD Graphics 2500 GPU a gwajin kan allo ya kasance 93%, kuma a cikin gwajin kashe allo ya kasance 36.3%.
- Don tsarin 10 Intel i5-10500 (GEN 10 Comet Lake) tare da katin bidiyo na NVIDIA GTX1650, daidaito ya kasance 70% da 95.8%.
- Don tsarin 15 Intel i5-8500 (GEN 8 Coffee Lake) tare da Intel UHD Graphics 630 GPU - 42% da 55%.
- Don tsarin 23 Intel i5-4590 (GEN 4 Haswell) tare da Intel HD Graphics 4600 GPU - 32.7% da 63.7%.
- Don wayoyi shida na Samsung Galaxy S20/S20 Ultra tare da Mali-G77 MP11 GPU, daidaiton tantancewa a cikin gwajin kan allo ya kasance 92.7%, kuma ga Samsung Galaxy S9/S9+ wayowin komai da ruwan da Mali-G72 MP18 ya kasance 54.3%.
An lura cewa yanayin zafi na GPU ya shafi daidaito, kuma ga wasu na'urori, sake kunna tsarin ya haifar da ɓarna na mai ganowa. Lokacin amfani da hanyar tare da sauran hanyoyin tantancewa kai tsaye, ana iya ƙara daidaito sosai. Hakanan suna shirin haɓaka daidaito ta hanyar amfani da inuwa mai ƙididdigewa bayan daidaitawar sabon WebGPU API.
An sanar da Intel, ARM, Google, Khronos, Mozilla da Brave game da matsalar a cikin 2020, amma an bayyana cikakkun bayanai game da hanyar. Masu binciken sun kuma buga misalan aiki da aka rubuta a cikin JavaScript da GLSL waɗanda zasu iya aiki tare da ba tare da nuna bayanai akan allon ba. Hakanan, don tsarin da ya dogara da GPU Intel GEN 3/4/8/10, an buga saitin bayanai don rarraba bayanan da aka fitar a cikin tsarin koyon injin.
source: budenet.ru