A halin yanzu, NVDIA ya gabatar da sabon ƙarni na Ampere graphics processor - flagship GA100, wanda ya kafa tushen NVIDIA A100 mai sarrafa kwamfuta. Kuma a yanzu shugaban OTOY, wani kamfani da ya ƙware a cikin samar da girgije, ya raba sakamakon gwajin farko na wannan mai haɓakawa.
The Ampere GA100 graphics processor amfani a cikin NVIDIA A100 ya ƙunshi 6912 CUDA cores da 40 GB na HBM2 RAM. GPU kanta ana yin ta ta amfani da fasahar tsari na 7-nm a wuraren TSMC. Ana samun injin sarrafa kwamfuta a cikin nau'ikan tare da mu'amalar PCIe 4.0 da SXM4. Da farko, NVIDIA A100 accelerators suna samuwa a matsayin ɓangare na tsarin sarrafa kwamfuta na NVIDIA DGX A100, wanda ya haɗa har zuwa GPUs takwas.
An gwada na'urar sarrafa kwamfuta ta NVIDIA A100 a cikin ma'aunin OctaneBench wanda bai shahara ba, wanda ke gwada aikin GPU yayin yin amfani da injin zane na Octane Render. Ya dogara da fasahar NVIDIA CUDA, ma'ana yana iya yin amfani da NVIDIA GPUs kawai. Kuma kamfanin da aka ambata OTOY yana haɓaka wannan injin.
An ba da rahoton cewa mai haɓakawa na NVIDIA A100 ya nuna sakamakon rikodin a OctaneBench, wanda ya kai maki 446. Idan aka kwatanta, NVIDIA Titan V na tushen Volta ya sami maki 401 (ƙananan 11%), yayin da mafi sauri katin zane na Turing-gen, Quadro RTX 8000, ya sami maki 328 kawai (43% ƙasa).
Don haka, babban aikin ƙa'idar aikin Ampere a zahiri yana fassara zuwa saurin ma'ana da sauri. Bari mu tunatar da ku cewa mafi girman aikin NVIDIA A100 shine 19,5 da 9,7 Tflops a daidai guda da ninki biyu, bi da bi. A lokaci guda, ƙarni na Turing Quadro RTX 8000 da aka ambata a sama na iya ba da saurin 16,0 da 0,5 Tflops kawai.
source:
source: 3dnews.ru