Waqtigan xaadirka ah, NVIDIA waxay soo bandhigtay hal jiil cusub oo naqshadeeyayaasha Ampere - calanka GA100, kaas oo aasaas u ah xawaaraha kombuyuutarada ee NVIDIA A100. Oo hadda madaxa OTOY, oo ah shirkad ku takhasustay samaynta daruuraha, ayaa la wadaagay natiijada tijaabada ugu horreysa ee dardargelintan.
Ampere GA100 garaafyada loo isticmaalo NVIDIA A100 waxaa ku jira 6912 CUDA cores iyo 40 GB ee HBM2 RAM. GPU laftiisa waxaa lagu sameeyay iyadoo la adeegsanayo tignoolajiyada habka 7-nm ee xarumaha TSMC. Xawaaraha xisaabinta waxaa lagu heli karaa noocyo leh PCIe 4.0 iyo SXM4 interfaces. Marka hore, xawaaraha NVIDIA A100 ayaa la heli karaa iyada oo qayb ka ah nidaamyada xisaabinta ee NVIDIA DGX A100, oo ay ku jiraan ilaa siddeed GPUs.
Xawaaraha kombuyuutarada ee NVIDIA A100 waxaa lagu tijaabiyay bartilmaameedka OctaneBench ee aan caanka ahayn, kaas oo tijaabiya waxqabadka GPU marka la isticmaalayo mashiinka sawirada Octane Render. Waxay ku tiirsan tahay tignoolajiyada NVIDIA CUDA, taasoo la micno ah inay bixin karto oo keliya iyadoo la adeegsanayo NVIDIA GPUs. Oo shirkadda la sheegay OTOY ayaa horumarinaysa mishiinkan.
Waxaa la sheegay in xawaaraha NVIDIA A100 uu muujiyay natiijada rikoodhka ee OctaneBench, taas oo u dhiganta 446 dhibcood. Marka la barbardhigo, Volta-based NVIDIA Titan V buundooyinka 401 (11% hoose), halka kaarka sawirada Turing-gen ee ugu dhaqsaha badan, Quadro RTX 8000, uu dhaliyay kaliya 328 dhibcood (43% hoose).
Sidaa darteed, waxqabadka sare ee aragtida processor-ka Ampere wuxuu dhab ahaantii u tarjumayaa xawli-bixin degdeg ah. Aan ku xasuusino in waxqabadka ugu sarreeya ee NVIDIA A100 uu yahay 19,5 iyo 9,7 Tflops hal iyo labanlaab sax ah, siday u kala horreeyaan. Isla mar ahaantaana, jiilka Turing Quadro RTX 8000 ee kor ku xusan waxay bixin karaan oo keliya xawaaraha 16,0 iyo 0,5 Tflops.
Source:
Source: 3dnews.ru