Abacwaningi abavela e-Ben-Gurion University (Israel), iNyuvesi yaseLille (eFrance) kanye neNyuvesi yase-Adelaide (Australia) bathuthukise indlela entsha yokuhlonza amadivaysi abasebenzisi ngokuthola imingcele yokusebenza ye-GPU kusiphequluli sewebhu. Indlela ibizwa ngokuthi "I-Drawn Apart" futhi isekelwe ekusetshenzisweni kwe-WebGL ukuze kutholwe iphrofayela yokusebenza ye-GPU, engathuthukisa ngokuphawulekayo ukunemba kwezindlela zokulandela umkhondo ezisebenza ngaphandle kokusebenzisa Amakhukhi futhi ngaphandle kokugcina isihlonzi ohlelweni lomsebenzisi.
Izindlela ezicabangela izici zokunikezwayo, i-GPU, isitaki sezithombe nezishayeli lapho kuhlonzwa zazisetshenziswa ngaphambilini, kodwa zazilinganiselwe ekhonweni lokuhlukanisa amadivayisi kuphela ezingeni lamamodeli ahlukene wamakhadi wevidiyo nama-GPU, i.e. ingasetshenziswa kuphela njengesici esengeziwe sokwandisa amathuba okuhlonza. Isici esibalulekile sendlela entsha ye-"Drawn Apart" ukuthi ayigcini ngokuhlukanisa amamodeli e-GPU ahlukene, kodwa izama ukukhomba umehluko phakathi kwama-GPU afanayo emodeli efanayo ngenxa yokuhlukahluka kwenqubo yokukhiqiza ama-chips adizayinelwe ukufana okukhulu. ikhompuyutha. Kuyaphawulwa ukuthi ukuhlukahluka okuvela ngesikhathi senqubo yokukhiqiza kwenza kube nokwenzeka ukwenza abalingisi abangaphindi bamamodeli edivayisi afanayo.
Kuvele ukuthi lo mehluko ungabonakala ngokubala inani lamayunithi okubulala nokuhlaziya ukusebenza kwawo ku-GPU. Ukuhlola okusekelwe kusethi yemisebenzi ye-trigonometric, ukusebenza okunengqondo nokubalwa kwamaphoyinti antantayo kwasetshenziswa njengezinto zokuqala ukuze kuhlonzwe amamodeli e-GPU ahlukene. Ukuze uhlonze umehluko kuma-GPU afanayo, inani lezintambo ezisetshenziswa ngesikhathi esisodwa lapho kusetshenziswa ama-vertex shader liye lahlolwa. Kucatshangwa ukuthi umphumela otholiwe ubangelwa umehluko ezimweni zokushisa kanye nokusetshenziswa kwamandla kwezimo ezihlukene zama-chips (ngaphambilini, umphumela ofanayo waboniswa kuma-CPU - amaphrosesa afanayo abonise ukusetshenziswa kwamandla okuhlukile lapho esebenzisa ikhodi efanayo).
Ngenxa yokuthi ukusebenza nge-WebGL kwenziwa ngokuhambisanayo, i-JavaScript API performance.now() ayikwazi ukusetshenziswa ngokuqondile ukukala isikhathi sayo sokwenza, ngakho-ke amaqhinga amathathu ahlongozwayo ukuze kukalwe isikhathi:
- esikrinini β kunikeza isigameko ngekhanvasi ye-HTML, kukala isikhathi sokuphendula somsebenzi wokuphinda ushaye, kusethwe nge-Window.requestAnimationFrame API futhi kubizwe ngemva kokuqedwa kokukhishwa.
- ngaphandle kwesikrini - usebenzisa isisebenzi nokwenza isigameko sibe into ye-OffscreenCanvas, ukulinganisa isikhathi sokwenziwa komyalo we-converterToBlob.
- I-GPU - Dweba entweni ye-OffscreenCanvas, kodwa sebenzisa isibali-sikhathi esinikezwe i-WebGL ukuze ulinganise isikhathi esicabangela ubude besikhathi sesethi yemiyalo ohlangothini lwe-GPU.
Ngesikhathi senqubo yokudala i-ID, kwenziwa izivivinyo ezingama-50 kudivayisi ngayinye, ngakunye kuhlanganisa izilinganiso eziyi-176 zezici eziyi-16 ezihlukene. Ukuhlolwa okuqoqe ulwazi kumadivayisi angu-2500 anama-GPU angu-1605 ahlukene kubonise ukukhuphuka okungu-67% ekusebenzeni kahle kwezindlela zokuhlonza ezihlanganisiwe lapho kwengezwa usekelo lwe-Drawn Apart. Ikakhulukazi, indlela ehlanganisiwe ye-FP-STALKER inikeze ukuhlonza phakathi kwezinsuku eziyi-17.5 ngokwesilinganiso, futhi uma ihlanganiswa ne-Drawn Apart, ubude besikhathi bokuhlonza bukhuphuke baba yizinsuku ezingama-28.
- Ukunemba kokuhlukaniswa kwamasistimu angu-10 anama-Intel i5-3470 chips (GEN 3 Ivy Bridge) ne-Intel HD Graphics 2500 GPU ekuhlolweni kwesikrini kube ngu-93%, futhi ekuhlolweni okungaphandle kwesikrini kube ngu-36.3%.
- Ezinhlelweni eziyi-10 ze-Intel i5-10500 (GEN 10 Comet Lake) ezinekhadi levidiyo le-NVIDIA GTX1650, ukunemba kwakungu-70% no-95.8%.
- Kuzinhlelo ezingu-15 ze-Intel i5-8500 (GEN 8 Coffee Lake) ezine-Intel UHD Graphics 630 GPU - 42% kanye no-55%.
- Kuzinhlelo ezingu-23 ze-Intel i5-4590 (GEN 4 Haswell) ezine-Intel HD Graphics 4600 GPU - 32.7% kanye no-63.7%.
- Kuma-smartphones ayisithupha e-Samsung Galaxy S20/S20 Ultra ane-Mali-G77 MP11 GPU, ukunemba kokuhlonza ekuhlolweni okusesikrinini kwakungu-92.7%, kanti kuma-Samsung Galaxy S9/S9+ ama-smartphones ane-Mali-G72 MP18 bekungu-54.3%.
Kuyaphawulwa ukuthi ukunemba kuthintwa izinga lokushisa le-GPU, futhi kwamanye amadivayisi, ukuqalisa kabusha isistimu kuholele ekuhlanekeni kwesihlonzi. Uma usebenzisa indlela ngokuhambisana nezinye izindlela zokuhlonza ezingaqondile, ukunemba kungakhuliswa kakhulu. Baphinde bahlele ukukhulisa ukunemba ngokusebenzisa ama-shader wokubala ngemuva kokuqiniswa kwe-WebGPU API entsha.
I-Intel, i-ARM, i-Google, i-Khronos, i-Mozilla ne-Brave yaziswa ngenkinga emuva ngo-2020, kodwa imininingwane yendlela iyembulwa kuphela manje. Abacwaningi baphinde bashicilela izibonelo zokusebenza ezibhalwe ku-JavaScript ne-GLSL ezingasebenza futhi ngaphandle kokubonisa ulwazi esikrinini. Futhi, kumasistimu asekelwe ku-GPU Intel GEN 3/4/8/10, amasethi edatha ashicilelwe ukuze ahlukanise ulwazi olukhishiwe kumasistimu okufunda emishini.
Source: opennet.ru