| JEMC said: I'm not arguing or questioning your numbers, I'm only saying that they don't tell the whole story or take everything into account because you're looking at it purely from the point of the manufacturing process while ignoring any architectural improvements that Nvidia can/will introduce with Blackwell. What if the increased focus on ray/path tracing leads to bigger SMs? That would affect density. What if the architecture is designed to run at faster clocks? That will impact performance. What if Nvidia doesn't want to go insane with power consumption and Broadwell is more energy efficient? That will also play a part in the end result, and neither of those three architectural decisions have anything to do with the jump to a new node. They will "only" leverage them to bring the end product to a whole new level. |
I mean, it's all related since the determinant of energy efficiency and clock speeds is still the manufacturing process due to feature physical size and signal dissipation. That was the entire reasonale behind Intel's 'ticks' in the past.
I'm not too hung up on IPC here since Nvidia's IPC gains in the recent past have also correlated with higher power consumption (not entirely, but mostly) so that's the lesser factor at play here (probably).
