NVIDIA yatulutsa zotsatira zatsopano, zochititsa chidwi kwambiri pantchito yogwiritsa ntchito mitundu yayikulu yazilankhulo (LLM) mu benchmark ya MLPerf Inference 4.0. M'miyezi isanu ndi umodzi yapitayo, zotsatira zapamwamba zomwe zawonetsedwa kale ndi kamangidwe ka Hopper muzochitika zongopeka zasinthidwa pafupifupi katatu. Zotsatira zochititsa chidwi zotere zidatheka chifukwa chakusintha kwa hardware mu H200 accelerators ndi kukhathamiritsa kwa mapulogalamu. Generative AI yaphulika kwambiri makampani: pazaka khumi zapitazi, mphamvu zamakompyuta zomwe zimagwiritsidwa ntchito pophunzitsa ma neural network zakula ndi maulamuliro asanu ndi limodzi, ndipo ma LLM okhala ndi ma trilioni sakhalanso zachilendo. Komabe, kulingalira kwa mitundu yotereyi ndi ntchito yovuta, yomwe NVIDIA imayandikira mozama, pogwiritsa ntchito, m'mawu akeake, "kukhathamiritsa kwamitundumitundu."
Source: 3dnews.ru