Okwamanje, i-NVIDIA yethule iphrosesa yehluzo yesizukulwane esisha se-Ampere - i-flagship GA100, eyakha isisekelo se-NVIDIA A100 computing accelerator. Futhi manje inhloko ye-OTOY, inkampani esebenza ngokuhlinzeka ngamafu, yabelane ngemiphumela yokuqala yokuhlolwa yalesi sisheshisi.
Iphrosesa yezithombe ze-Ampere GA100 esetshenziswe ku-NVIDIA A100 ihlanganisa ama-cores angu-6912 CUDA kanye no-40 GB we-RAM ye-HBM2. I-GPU ngokwayo yenziwa kusetshenziswa ubuchwepheshe benqubo ye-7-nm ezikhungweni ze-TSMC. I-accelerator yekhompyutha iyatholakala ngezinguqulo ezine-PCIe 4.0 kanye ne-SXM4 interface. Ekuqaleni, ama-accelerator e-NVIDIA A100 ayatholakala njengengxenye yezinhlelo zekhompyutha ze-NVIDIA DGX A100, ezihlanganisa ama-GPU afinyelela kwayisishiyagalombili.
Isisheshisi sekhompyutha se-NVIDIA A100 sihlolwe kubhentshimakhi ye-OctaneBench engadumile, ehlola ukusebenza kwe-GPU lapho inikezela kusetshenziswa injini yezithombe ze-Octane Render. Ithembele kubuchwepheshe be-NVIDIA CUDA, okusho ukuthi inganikezela kusetshenziswa ama-NVIDIA GPU. Futhi inkampani eshiwo i-OTOY ithuthukisa le njini.
Kubikwa ukuthi i-accelerator ye-NVIDIA A100 ibonise umphumela werekhodi ku-OctaneBench, efinyelele amaphuzu angu-446. Ngokuqhathanisa, i-Volta-based NVIDIA Titan V ithola amaphuzu angu-401 (11% ephansi), kuyilapho ikhadi lemifanekiso le-Turing-gen elisheshayo, i-Quadro RTX 8000, lithola amaphuzu angu-328 nje (43% ngaphansi).
Ngakho-ke, ukusebenza okuphezulu kwethiyori kwephrosesa ye-Ampere empeleni kuhumushela kuzivinini zokunikezela ezisheshayo. Ake sikukhumbuze ukuthi ukusebenza okuphezulu kwe-NVIDIA A100 kungu-19,5 no-9,7 Tflops ngokunemba okukodwa nokukabili, ngokulandelana. Ngesikhathi esifanayo, isizukulwane se-Turing i-Quadro RTX 8000 okukhulunywe ngayo ngenhla singanikeza kuphela isivinini esingu-16,0 no-0,5 Tflops.
Source:
Source: 3dnews.ru