Without Frame Generation, the gen-on-gen increase in performance is relatively modest (by historical standards for GPUs). Just 20% improvement between a 4070 and 5070.

It was rather obvious Nvidia was fudging performance results via framegen. Now let’s see those numbers verified independently and we’ll land in a typical generation to generation improvement probably. I wouldn’t be surprised if neural shaders demo was done by an actual demoscene coder because it looked like something someone could do in 64k demo competition and not like something that a developer could do at scale without a horde of artists.

I hope that things like input lag, frame persistence and frame time consistency become metrics we use to judge GPUs as soon as possible because current ones are no longer representative of hardware performance.