Isikhulu se-NVIDIA uJensen Huang wethule amakhadi evidiyo esizukulwane esilandelayo alindelwe ngabomvu avela ekhishini lakhe. Njengoba kulindelekile, izixazululo ezindala zimenyezelwe namuhla: I-GeForce RTX 3090, i-GeForce RTX 3080 ne-GeForce RTX 3070. Amakhadi evidiyo akhiwe kuma-GPU esizukulwane se-Ampere enziwe kusetshenziswa ubuchwepheshe benqubo ye-Samsung's 8nm, kuyilapho abanduleli bawo be-Turing bekhiqizwa kusetshenziswa ubuchwepheshe be-12nm i-TSMC.
I-GeForce RTX 3090
Ukushintshela kunqubo entsha, emincane yezobuchwepheshe kwenze kwaba nokwenzeka ukubeka ama-transistors amaningi ku-chip, ngenxa yokuthi inani lamabhulokhi asebenzayo landa kakhulu. Ngeshwa, u-Jensen Huang akazange acacise ukulungiselelwa kwe-GPU, futhi waqamba kuphela amapharamitha wesistimu engaphansi yememori phakathi kwezici. I-GeForce RTX 3090 ifakwe inkumbulo engu-24 GB ye-GDDR6X enejubane elisebenzayo lewashi elingu-19,5 GHz. Kanye nebhasi elibanzi elingu-384-bit, lokhu kunikeza umkhawulokudonsa wesistimu engaphansi yememori engu-936 GB/s. I-GeForce RTX 2080 Ti, khumbula, yayinememori engu-11 GB kuphela ye-GDDR6 ene-bandwidth engaphansi kwezikhathi eziphindwe nesigamu.
Ukusebenza okuphezulu kwe-GPU okucatshangelwayo kwekhadi elisha levidiyo ngama-36 TFLOPS “ngokwakhiwa kwezithombe zendabuko kusetshenziswa ukufiphaza (imithunzi)” (Shader-TFLOPS). Ukusebenza lapho usebenza ngokulandela umsebe kushiwo ku-69 TFLOPS (RT-TFLOPS), futhi ukusebenza uma usebenzisa ama-tensor cores kufinyelela ku-285 TFLOPS (Tensor-TFLOPS).
Ngokusho kwe-NVIDIA, leli khadi levidiyo lenzelwe ukudlala ngokulungiswa kwe-8K, lapho likwazi ukuhlinzeka nge-FPS engu-60 ezinzile. Kuphinde kwaphawulwa ukuthi lesi “silo” sicishe sibe ngama-50% ngokushesha ekulungisweni kwe-4K kune-Titan RTX.
I-GeForce RTX 3080
Inhloko ye-NVIDIA ibize ikhadi levidiyo le-GeForce RTX 3080 ngokuthi ifulegi elisha, idlula umbuzo wokuthi iyini i-GeForce RTX 3090 kuleli cala. Nokho, i-GeForce RTX 3080 ibukeka “njengefulegi,” kokubili ngokubukeka nokusebenza.
Phakathi kwezici, ukuba khona kwe-10 GB yenkumbulo ye-GDDR6X enemvamisa esebenzayo ye-19 GHz kwaphawulwa. Inkumbulo ixhunywe ngebhasi elingu-320-bit, eligcina lihlinzeka ngokuphuma kwe-760 GB/s. Masikhumbule ukuthi i-GeForce RTX 2080 yangaphambilini yayinememori engu-8 GB kuphela ye-GDDR6 enomkhawulokudonsa ongu-448 GB/s.
Ukusebenza kwe-GeForce RTX 3080 ngokunikezwa kwendabuko kusetshenziswa ama-shader yi-30 TFLOPS (Shader-TFLOPS). Lapho kucutshungulwa ukulandelwa kwe-ray, ikhadi levidiyo lihlinzeka ngama-58 TFLOPS (RT-TFLOPS), futhi ekusebenzeni kuma-tensor cores linikeza kufika ku-238 TFLOPS (Tensor-TFLOPS). Ngokusho kwenhloko ye-NVIDIA, ikhadi levidiyo le-GeForce RTX 3080 likhiqiza ngokuphindwe kabili kunamanduleli walo, i-GeForce RTX 2080.
I-GeForce RTX 3070
Ikhadi elilula le-trio ethuliwe, i-GeForce RTX 3070, njengeyandulela i-GeForce RTX 2070, inememori engu-8 GB ye-GDDR6.
Izinga lokusebenza le-GeForce RTX 3070 accelerator laliyi-20 TFLOPS ngokunikezwa kwendabuko ngama-shader (Shader-TFLOPS). Ukusebenza lapho usebenza nge-ray tracing kushiwo ku-40 TFLOPS (RT-TFLOPS), futhi ukusebenza ekusebenzeni ngama-tensor cores kufinyelela ku-163 TFLOPS (Tensor-TFLOPS). I-NVIDIA ithi i-GeForce RTX 3070 entsha ikhiqiza kakhulu kune-flagship yangaphambilini - i-GeForce RTX 2080 Ti.
Owokuqala wemikhiqizo emisha kuzoba yi-GeForce RTX 3080. Lokhu kuzokwenzeka kule nyanga - Septhemba 17. Izindleko zizohlala ezingeni labanduleli bayo - $ 699 (eRussia - ruble 63). I-GeForce RTX 500 ethengeka kakhulu izovela emashalofini ngo-Okthoba, futhi intengo yayo izoba ngu-$3070 (ama-ruble angu-499) - futhi, ezingeni lomanduleli wayo. Ekugcineni, i-GeForce RTX 45 izoqala ukudayiswa ngoSepthemba 500, etholakala ngokushesha kubo bobabili i-NVIDIA nabalingani bayo, ngentengo... $3090 (24 ruble). Akukho ukukhuphuka kwentengo lapha.
Source:
Source: 3dnews.ru