GeForce GTX 670 Versus GTX 680 And Radeon HD 7970
Given a GPU with roughly 88% of its CUDA cores enabled and slightly lower clock rates, we’d expect the GeForce GTX 670 to give up at least 12% of its performance compared to GeForce GTX 680. But then you start factoring in the complexities of GPU Boost and add on the fact that frame rates often don’t scale based on shader resources—particularly when an application isn’t bound by shader throughput.
Knowing all of that, it is still surprising to see our GeForce GTX 670 sample perform just 4.5% slower than our reference GeForce GTX 680—certainly less of a gap then we were told to expect. Knowing that overclocking on air is often enough to overcome a sub-5% spread, this almost negates any reason you might have had for buying a GeForce GTX 680, particularly if you were looking to play games at 1920x1080 with the eye candy cranked up.
We figured that upping the comparison to 2560x1600 would increase the gap between GeForce GTX 670 and 680. But the average is only 1% higher. The 670 still looks mighty juicy.
How does it compare to the Radeon HD 7970?
Based on the complete benchmark analysis, we know that Nvidia’s design does best at 1680x1050, while AMD competes most aggressively at 2560x1600. Using 1920x1080 again gives us a good gauge of playable performance at the native resolution of many popular panels.
Nvidia’s architecture establishes a distinct lead in Crysis 2, DiRT 3, and WoW. Its advantage is less pronounced in Battlefield 3 and Skyrim. AMD jumps ahead in Metro 2033.
Of course, you could add games to this list all day long, and the balance would undoubtedly shift. In our suite, though, the GTX 670 averages more than 13% faster. If you leave all of the test results alone and shift that World of Warcraft score to 2560x1600, favoring the Radeon a little more, the average drops to 11.4% faster—still a notable advantage for a card that is expected to sell for $80 less.