係窮呀哈利
40系開始同Quadro 個cuda同tensor core/transformer core比例唔同咗,PCIE bandwidth ,vram同float point 容量大提升做market segmenting
30系卡仲係同當季運算卡分别無咁大所以仲係抵玩,新架構就運算卡真係100%主打AI市場,同樣電力唔使推埋RTX core效能會更高
GeForce 保留tensorcore更大程度係留俾DLSS用,tf pytorch by default 係行tensor core唔係CUDA core,直接有hardware acc