but can it play bf3 ultra with eyefinty with aa
Here's your answer:
HardOCP's Review goes more in-depth than the others, tests out triple-monitor standard configs.
Metro2033 is one game where 7970 shows strongly but even then it's behind by 5-7%.
As it happens, Metro isn't one for a consistent benchmark scenario. Even in normal gameplay the max framerate you'll see often changes through separate playthroughs and often level reloads (play, save, exit, enter and re-load). The way the 4A Games Engine works is very similar to the Crytek one - in order to keep the level and experience fresh, small things are randomly changed like light sources and other environmental effects. Also I guess there's a difference in the motherboards and processors used in the various systems set up in reviews. I notice a few that have higher Metro 2033 scores use SSDs, while ones more erratic like Tom's used a mechanical drive early on. Too many variables, not many answers.
Still an amazing game though, can't wait for the sequel.
But back to the GTX680; it rocks. Technically, and in Direct Compute capabilities Kepler and GCN stand as equals, though its as if AMD and Nvidia had a mind switch. GCN introduces an efficient design and new hardware that sorts through code dependencies and executes those first, while Kepler cuts out the hardware that did exactly that in Fermi and replaced it with something a bit more streamlined and less wasteful on power and heat generation.
That does bring Direct Compute performance down, but in most cases it doesn't matter. In one of the reviews I saw, a GTX680 (alone) beats the combination of an i7 2600k and a GTX580 to re-code a video from HD to SD video suitable for Youtube, even taking out Intel Quicksync in the process. Nvidia has got a winner here, and I know if I had the money, it'd be on buying a new Kepler card for my machine.
Until the GK110 comes out, it's all just talk.
From what we see with the GTX 680, the performance is better than the 7970 (by how much depends on which review sites you visit).
Actually, using GK104 we can easily see what GK110 is going to be like. Take the exact same design of GF110: triple the amount of CUDA cores, double the amount of SMX modules, double the Raster count, lower the TDP by 40watts, double the Load/Storage Units, Interpolation and Special Function SFUs and then add in double texture units and extra CUDA Floating-Point cores. I predict it'll result in a 25% increase over the GTX590.
Although here's where it gets interesting: with the GTX690, we may see that the card is held back by current-gen hardware. An i7 2600k isn't going to be enough for that beast - for that you'll need PCI-E 3.0, oodles of memory bandwidth and an X79 board and a 6-core Intel Ivy Bridge processor. Triple-monitor 1920x1200 displays will play very nicely, I think.
Also, don't forget FXAA! That in itself is a tremendous achievement.
.
.
