Well, that's a good question and a tough one. Since games have become more memory bandwidth, it's really not about optimizing as much as the old days with doing assembly language functions.
P
psorcerer
would probably disagree with me.
Let's take a look at one thing that everyone wants.. 4k.. true 4k.. well, that's just a switch but it makes the entire pipeline hungry for more bandwidth. The computations increase according to how many pixels you have to render to the game per second. That automatically stresses out a low bandwidth graphics card. Knowing the TFLOP isn't going to help with guesstimating how a game will look because it's not taking into account how the game was programmed or how busy the traffic is going to be going from the bus to the graphics card.. and then on the graphics card there are several other memory banks (L1, L2 cache, shared memory, unified memory) that have their own sizes and speeds. It's a moving target and hard to put a finger on. I wouldn't be shocked if Sony nor MS give out TFLOPS.
In answer to your question, I think the top priority these days is going to be which hardware gives them the best bandwidth. That way they can use higher res textures, they can use more texture filtering samples higher than the PS4's 4x. They can incorporate more intersection tests for ray-tracing, etc.. Right now, the Nvidia boards give them that. We haven't seen the AMD boards yet, but hopefully they will reach parity.
