I kēia manawa, ua hoʻolauna ʻo NVIDIA i hoʻokahi wale nō kaʻina hana kiʻi kiʻi Ampere - ka hae nui GA100, ka mea i hoʻokumu i ke kumu o ka NVIDIA A100 computing accelerator. A i kēia manawa ua hōʻike ke poʻo o OTOY, kahi hui loea i ka hoʻolilo ʻana i ke ao, i nā hopena hōʻike mua o kēia accelerator.
ʻO ka ʻōnaehana kiʻi Ampere GA100 i hoʻohana ʻia i ka NVIDIA A100 me 6912 CUDA cores a me 40 GB o HBM2 RAM. Hana ʻia ka GPU me ka ʻenehana kaʻina hana 7-nm ma nā hale TSMC. Loaʻa ka hoʻokele helu helu i nā mana me nā pilina PCIe 4.0 a me SXM4. I ka wā mua, loaʻa nā accelerators NVIDIA A100 ma ke ʻano he ʻāpana o nā ʻōnaehana computing NVIDIA DGX A100 proprietary, e komo pū ana i ʻewalu GPU.
Ua hoʻāʻo ʻia ka NVIDIA A100 computing accelerator ma ka papa kuhikuhi OctaneBench kaulana ʻole, e hoʻāʻo ana i ka hana GPU i ka wā e hoʻohana ai i ka mīkini kiʻi Octane Render. Ke hilinaʻi nei ia i nā ʻenehana NVIDIA CUDA, ʻo ia ka mea hiki ke hāʻawi wale me ka hoʻohana ʻana i nā NVIDIA GPU. A ʻo ka hui i ʻōlelo ʻia ʻo OTOY e hoʻomohala nei i kēia mīkini.
Ua hōʻike ʻia ua hōʻike ka NVIDIA A100 accelerator i kahi hopena moʻolelo ma OctaneBench, ʻo ia ka 446 mau helu. Ma ka hoʻohālikelike ʻana, loaʻa ka NVIDIA Titan V i hoʻokumu ʻia ma Volta i nā helu 401 (11% haʻahaʻa), ʻoiai ʻo ke kāleka kiʻi wikiwiki Turing-gen, ʻo ka Quadro RTX 8000, loaʻa nā helu 328 wale nō (43% haʻahaʻa).
No laila, ua unuhi maoli ka hana theoretical kiʻekiʻe o ka mea hana Ampere i nā wikiwiki hoʻololi wikiwiki. E hoʻomanaʻo mākou iā ʻoe ʻo ka hana kiʻekiʻe o ka NVIDIA A100 ʻo 19,5 a me 9,7 Tflops ma ka hoʻokahi a me ka pālua. I ka manawa like, hiki i ka Turing hanauna Quadro RTX 8000 i ʻōlelo ʻia ma luna ke hāʻawi wale i nā wikiwiki o 16,0 a me 0,5 Tflops.
Source:
Source: 3dnews.ru