Njengengxenye yengqungquthela ye-GDC 2019, iGoogle ibambe owayo umcimbi, lapho yethule khona insizakalo entsha yokusakaza umdlalo i-Stadia. Sesivele sikhulume ngesevisi ngokwayo, futhi manje sithanda ukukutshela kabanzi mayelana nokuthi uhlelo olusha lwe-Google lusebenza ngani, ngoba lisebenzisa izixazululo ezihlukahlukene ezenzelwe lolu hlelo.
Isici esibalulekile sohlelo lwe-Google, vele, ama-GPU. Izixazululo zangokwezifiso ze-AMD zisetshenziswa lapha, ezisekelwe ekwakhiweni kwezithombe ze-Vega. Kubikwa ukuthi i-GPU ngayinye inama-56 Compute Units (CUs), futhi ifakwe nenkumbulo ye-HBM2.
Umuntu angacabanga ukuthi i-Google isebenzisa amakhadi ehluzo eduze nomthengi i-Radeon RX Vega 56. Nokho, eqinisweni, izixazululo zangokwezifiso ze-AMD zinokuhluka okuningana okubalulekile. Okokuqala, inkumbulo esheshayo ene-bandwidth engu-484 GB / s isetshenziswa lapha. Umthengi u-Radeon RX Vega 64 unenkumbulo efanayo, kuyilapho i-Radeon RX Vega 56 encane isebenzisa inkumbulo esheshayo encane (410 GB / s). Ngokushesha, siqaphela ukuthi inani eliphelele lememori ohlelweni yi-16 GB, ingxenye yayo, ngokusobala, inkumbulo yevidiyo ye-HBM2, kanti enye i-DDR4 RAM.
Kodwa okubaluleke nakakhulu, i-Google ifuna ama-teraflops angu-10,7 okusebenza kwama-GPU ayo, ngokusobala ngezibalo ezinembile (FP32). Umthengi iRadeon RX Vega 56 ikwazi ukuhlinzeka kuphela ngama-teraflops angu-8,3. Kungaba okunengqondo ukucabanga ukuthi izixazululo ze-Google zisebenzisa ama-GPU anemvamisa ephezulu. Futhi lokhu, ngokulandelayo, kuphakamisa ukuthi i-AMD idale iphrosesa yezithombe ze-Stadia ngokusekelwe ekwakhiweni kwe-Vega II ebuyekeziwe, futhi ikhiqizwa kusetshenziswa ubuchwepheshe benqubo ye-7nm.
Ngokuqondene nephrosesa, i-Google ayicacisi ukuthi yisiphi isisombululo somkhiqizi esisisebenzisile ezinhlelweni zesevisi ye-Stadia. Kuthiwa lokhu kuyiphrosesa evumelana ne-x86 ene-frequency engu-2,7 GHz, eno-9,5 MB wenqolobane ezingeni lesibili nelesithathu, kanye ne-multithreading (Hyperthreading) nokusekelwa kwemiyalelo ye-AVX2. Inani lenqolobane negama le-multithreading njengokuthi "HyperThreading" libonisa ukuthi lena yi-Intel chip. Kodwa-ke, ukusekelwa kwe-AVX2 kuphela ngaphandle kwe-AVX512 yesimanjemanje kusibhekisa ku-AMD, okuthi, ngaphezu kwalokho, kwaziwa kangcono ngama-chips ayo angokwezifiso. Kungenzeka kakhulu ukuthi ama-processor amasha e-AMD asuselwa ku-7nm Zen 7 nawo asetshenziswe kanye ne-2nm Vega GPU.
Lawa amasistimu i-Google ezowahlinzeka cishe kubasebenzisi besevisi yayo entsha yokudlala, i-Stadia. Amandla amaningi e-computing, kufanele ngisho, kodwa ayadingeka ukuze kuqinisekiswe ukusebenza okuphezulu emidlalweni. Ngaphezu kwalokho, i-Google ihlela ukuqinisekisa ukwenziwa kwemidlalo ngezinqumo ezifika ku-4K ngemvamisa ye-60 FPS.
Source: 3dnews.ru