3DMark 11 is a synthetic benchmark designed to test a GPU's performance at various aspects of DirectX 11 such as tessellation and DirectCompute. The diminutive card was just slightly behind our downclocked Palit card that mimics stock GeForce GTX 670 performance. In fact, the margins that separate them on both Performance and Extreme settings were just about 1%. Our reference NVIDIA GTX 680 card naturally pulled ahead of the GTX 670 ones by a comfortable margin of 8%, showing that the GTX 670 is a fairly powerful card and within striking distance of what the GTX 680 can achieve.
Crysis 3 is currently one of our toughest gaming benchmark due to its extreme amounts of tessellation, per-pixel per-object motion blur, Bokeh Depth of Field, displacement mapping, particle and volumetric lighting and fog shadows. However, due to some technical glitch, we couldn't get the GeForce GTX 670 cards to complete the test scenario when we set the display resolution to a maximum of 2560 x 1600 pixels resolution. However, they were able to run the test scenario smoothly when we disabled our frame rate recording software, Fraps.
We can see that at the maximum resolution with anti-aliasing turned off, both the cards were churning out average frame rates that are below 30 frames per second mark; hence, we can safely conclude that with anti-aliasing enable at 8x, the game play will be choppy. The GeForce GTX 680 lagged in its scores due to its outdated drivers as its results were derived from our new test rig prior to the availability of the current driver release, NVIDIA ForceWare 314.22.
At the end of our overclocking exercise, we managed to boost the ASUS GTX 670 DirectCU Mini OC to 1098MHz, with its memory overclocked to 6308MHz GDDR5; we were actually able to push these levels higher but we experience performance degradation due to GPU throttling. By overclocking, we managed to garner performance boosts that range from 10.5- to about 12% gains. This is in line with what we had expected.