SmokSmog
Member
Not the same, different architecture, additional FP32 on Ampere/Ada are actually usable.Lol Nvidia did the same double tflop thing with Ampere
RTX 2000 SM = 64 Int32 + 64 FP32
Ampere SM = 64 Int32/FP32 + 64 FP32
It can switch to Int or FP if needed.
It was something around 30% boost to performance per SM if I'm correct.
Games are 25-35% Int32 and rest FP32 operations if I'm correct again.