Ua hoʻopuka ʻo NVIDIA i nā hopena hou, ʻoi aku ka maikaʻi o ka hana ʻana me nā hiʻohiʻona ʻōlelo nui (LLM) ma ka MLPerf Inference 4.0 benchmark. I loko o nā mahina ʻeono i hala iho nei, ua hoʻomaikaʻi ʻia nā hopena kiʻekiʻe i hōʻike ʻia e ka hoʻolālā ʻo Hopper i nā hiʻohiʻona inference a kokoke ʻekolu. Ua hoʻokō ʻia kēlā hopena maikaʻi loa i ka hoʻomaikaʻi ʻana i nā ʻenehana ʻelua i ka H200 accelerators a me ka loiloi polokalamu. Ua pahū maoli ʻo Generative AI i ka ʻoihana: i nā makahiki he ʻumi i hala iho nei, ua hoʻonui ʻia ka mana computing i hoʻohana ʻia no ka hoʻomaʻamaʻa ʻana i nā ʻupena neural e ʻeono mau kauoha o ka nui, a ʻo nā LLM me nā ʻāpana trillion ʻaʻohe mea maʻamau. Eia nō naʻe, he hana paʻakikī hoʻi ka hoʻohālikelike ʻana o ia mau hiʻohiʻona, kahi a NVIDIA e hoʻokokoke ākea ai, me ka hoʻohana ʻana, ma kāna mau ʻōlelo ponoʻī, "multidimensional optimization."
Source: 3dnews.ru
