Xitaa inta lagu guda jiro ku dhawaaqida kaararka fiidiyoowga taxanaha ah ee GeForce RTX 20, qaar badan ayaa rumaysnaa in Turing GPUs aysan ku lahayn dhammaan cabbirrada yaryar joogitaanka unugyo dheeri ah: Koofiyadaha RT iyo kombuyuutarrada. Hadda, mid ka mid ah isticmaalayaasha Reddit ayaa falanqeeyay sawirada infrared ee Turing TU106 iyo TU116 GPUs oo ay ku soo gabagabeeyeen in unugyada cusub ee xisaabinta aysan qaadanin boos badan sidii markii hore loo maleeyay.
Si aan u bilowno, aan dib u xasuusanno in Turing TU106 GPU uu yahay kan ugu da'da yar uguna is haysta NVIDIA chips oo leh koojyo gaar ah oo RT ah oo loogu talagalay raadinta rayiga iyo koodhka tensor si loo dardargeliyo hawlaha sirdoonka macmal. Dhanka kale, Turing TU116 processor-ka garaafyada, ee la xidhiidha, ayaa laga reebay cutubyada xisaabinta ee gaarka ah, waana sababta loo go'aansaday in la barbardhigo.
NVIDIA Turing GPU-yada waxay u qaybsan yihiin unugyo TPC, oo ay ku jiraan labo-processors multiprocessors oo qulqulaya (Streaming Multiprocessors), kuwaas oo ay ku jiraan dhammaan kombuyuutarrada. Oo sida ay soo baxday, Turing TU106 GPU waxay leedahay kaliya 1,95 mm² aagga TPC ka badan Turing TU116, ama 22%. Aaggan, 1,25 mm² waxaa loogu talagalay koojiyeyaasha tensor, iyo 0,7 mm² kaliya ayaa loogu talagalay koojiyeyaasha RT.
Waxay soo baxday in la'aanteed tensor-ka cusub iyo RT cores, calanka Turing TU102 sawir-qaade, kaas oo hoosta ka xariiqaya GeForce RTX 2080 Ti, uusan qabsan doonin 754 mm², laakiin 684 mm² (36 TPC). Taa baddalkeeda, Turing TU104, oo ah saldhigga GeForce RTX 2080, wuxuu qabsan karaa 498 mm² halkii uu ka ahaan lahaa 545 mm² (24 TPC). Sida aad arki karto, xitaa la'aanteed tensor iyo RT cores, Turing GPU-yada da'da weyn waxay noqon lahaayeen chips aad u weyn. Si aad ah u badan Pascal GPUs.
Haddaba waa maxay sababta tiradaas badan? Bilawga, Turing GPUs waxay lahaayeen cabbirro kayd oo waaweyn. Baaxadda hadhka ayaa sidoo kale la kordhiyey, iyo Chips Turing waxay leeyihiin habab waxbarasho oo waaweyn iyo diiwaanno waaweyn. Waxaas oo dhami waxay suurtogal ka dhigeen in si weyn loo kordhiyo ma aha aagga oo keliya, laakiin sidoo kale waxqabadka Turing GPUs. Tusaale ahaan, isla GeForce RTX 2060 ee ku saleysan TU106 waxay bixisaa ku dhawaad heer la mid ah waxqabadka GeForce GTX 1080 oo ku saleysan GP104. Midda dambe, habka, waxay leedahay tiro badan oo ah 25% oo ah koofiyadaha CUDA, in kasta oo ay ku jirto aag 314 mm2 ah oo ka soo horjeeda 410 mm2 ee TU106 cusub.
Source: 3dnews.ru