Nagorak
Well-known member
Instead of cramming everything into one GPU, I wish we could use the full die for traditional rendering (no tensor cores) and then an add-in card that is exclusively tensor cores for RTX rendering. Give people the option of RTX, not force the technology down our throats for the minimal games that use it.
Imagine how much faster the 2080TI would be if it was a full die and not cut down for tensor cores. How much faster RTX could render on a die of ONLY tensor cores..
I was hoping AMD would go down this route but it appears they are going to follow suite.
I agree in principle, but I wonder how much faster Turing would actually be? The thing is, these chips are already running at higher power draw than Pascal. If they had even more of the die devoted to rasterization surely that would increase performance, but wouldn't higher power draw also go along with it? I'm not sure there's that much more power envelope available above what we're already drawing.