Tare da sabon processor core ARM ta gabatar da na'urar sarrafa hoto da aka ƙera don tsarin guntu guda ɗaya na wayar hannu mai zuwa. Mali-G77, wanda bai kamata a rikita shi da sabon na'ura mai sarrafa nuni ba , yana nuna canji daga gine-ginen ARM Bifrost zuwa Valhall.
ARM ta ayyana gagarumin haɓakar ayyukan zane-zane na Mali-G77 - da kashi 40% idan aka kwatanta da na yanzu na Mali-G76. An cimma wannan duka ta hanyar tsarin fasaha da haɓakar gine-gine. Mali-G77 na iya samun daga murdiya 7 zuwa 16 (aunawa daga 1 zuwa 32 abu ne mai yiyuwa a nan gaba), kuma kowannen su kusan girman G76 ne. Saboda haka, manyan wayowin komai da ruwan za su kasance suna da adadin adadin abubuwan GPU iri ɗaya.
A cikin wasanni, kuna iya tsammanin haɓaka aiki tsakanin 20 zuwa 40%, ya danganta da nau'in nauyin aikin zane. Yin la'akari da sakamakon sanannen gwajin GFXBench na Manhattan, babban fifikon sabon GPU akan ƙarni na yanzu zai tilasta abokin hamayyarsa Qualcomm ya damu da gagarumin ci gaba a cikin ayyukan zane-zane na Adreno.
A kan kansa, sabon tsarin gine-gine na Mali-G77 yana ba da matsakaicin haɓaka kashi 30 cikin 16 na inganci ko aiki, in ji ARM. Ƙarni na biyu na ARM Valhall scalar architecture yana ba GPU damar aiwatar da umarnin 76 a kowane zagaye a layi daya akan CU, idan aka kwatanta da takwas a cikin Bifrost (Mali-G1.3). Sauran sabbin abubuwa sun haɗa da cikakken tsarin tsarin koyarwa mai ƙarfi da kayan masarufi da sabon saitin koyarwa gaba ɗaya yayin da ake ci gaba da dacewa da baya tare da Bifrost. An kuma ƙara goyan bayan tsarin matsawa na ARM AFBC16 da sauran sabbin abubuwa (FPXNUMX masu sa ido, maƙasudin ma'auni da kayan aikin inuwa na vertex).
Bifrost CU ya ƙunshi injunan zartarwa guda 3, kowannensu ya haɗa da cache na koyarwa, rajista, da sashin kula da Warp. Rarraba a cikin waɗannan injunan guda uku sun ba da izinin aiwatar da umarnin FMA 24 a daidaitaccen ma'aunin iyo 32-bit (FP32). A Valhall, kowane CU yana da injin aiwatarwa guda ɗaya kawai, wanda aka raba tsakanin raka'o'in ƙididdigewa biyu masu iya sarrafa umarnin Warp 16 a kowane agogo, wanda ya haifar da jimillar kayan aiki na 32 FMA FP32 umarnin kowane CU. Godiya ga waɗannan sauye-sauyen gine-gine, Mali-G77 na iya yin ƙarin lissafin lissafin kashi ɗaya bisa uku a cikin lissafta daidai da Mali-G76.
Bugu da kari, kowane ɗayan waɗannan CUs ya ƙunshi sabbin tubalan aikin lissafi guda biyu. Sabuwar injin jujjuyawa (CVT) yana ɗaukar ainihin lamba, ma'ana, reshe, da umarnin juyawa. Sashin Aiki na Musamman (SFU) yana haɓaka haɓaka lamba, rarrabuwa, tushen murabba'i, logarithms, da sauran hadaddun ayyukan lamba.
Daidaitaccen toshewar FMA yana da saitunan da yawa waɗanda ke goyan bayan umarnin 16 FP32 a kowane zagaye, 32 don FP16, ko 64 don Samfurin Dot8 na INT60. Waɗannan haɓakawa na iya samar da haɓaka aiki har zuwa XNUMX% a aikace-aikacen koyon injin.
Wani mahimmin canji a cikin Mali-G77 shine ninka aikin injin rubutu, wanda yanzu ke sarrafa texels bilinear 4 a kowace agogo idan aka kwatanta da biyun baya, 2 trilinear texels a kowace agogo, yana ba da damar FP16 da FP32 mai sauri.
ARM ta yi wasu sauye-sauye da dama, tare da Mali-G77 da Valhall suna yin alƙawarin inganta ingantaccen aiki don ayyukan wasan kwaikwayo da na'ura. Mahimmanci, ana kiyaye amfani da wutar lantarki da yanki na guntu a matakan Bifrost, masu ba da alƙawarin na'urorin hannu tare da mafi girman aiki ba tare da ƙara yawan wutar lantarki ba, zubar da zafi da buƙatun girman.
source: 3dnews.ru