Let's recap. The GTX 670 2GB is essentially the same 28nm Kepler GK104 GPU featured in the GTX 680 2GB, but with one of its SMCs removed, thus leaving it with a total of 1,344 stream processors and 112 texture units. It also sports seven of Nvidia's Polymorph engines for tesselation duties, and four rasterisers.
At stock frequencies, the card has a 915MHz core clock speed, with a boost speed of 980MHz (although the card will boost past this under heavy load). Gigabyte, however, has applied a modest 7 per cent overclock, taking the core clock speed to 980MHz for a boost clock of 1,058MHz, though the card actually reached 1,162MHz in our testing.
Skyrim tends to favour Nvidia hardware, so it's unsurprising to see GeForce cards do better than their Radeon counterparts. Nonetheless, the Gigabyte GTX 670 2GB clearly outperforms the HD 7970 3GB in every test, and either beats or equals the costlier GHz Edition in each test too, which is an excellent result.
Gigabyte GeForce GTX 670 2GB X Turbo - Overclocking
As Gigabyte's overclock is a fairly modest one, we were hoping to be able to squeeze some extra juice and higher frame rates from its card, especially as it houses an extra power phase. We fired up EVGA's Precision tool for the job, and once we'd settled on the highest stable overclock, repeated the Unigine benchmark and the Battlefield 3 2,560 x 1,600 test to see the impact.
We ramped up the voltage the maximum available, which was 1175mV, and also increased the available power to the maximum of 112 per cent. With these settings, we were able to add an extra 100MHz to the core clock, taking it to 1,080MHz. This represents roughly a 10 per cent overclock, or an 18 per cent one over stock frequencies, which is not too shabby. It's also 40MHz higher than we were able to overclock the stock GTX 670 2GB to. The claimed boost clock at this setting is 1,159MHz, but we saw it boost to 1,254MHz in use.
As for the memory, we were able to clock it up by a whopping 350MHz (1.4GHz effective) before the card begun to become unstable. This gave us a memory frequency of 1.85GHz (7.4GHz effective), a 23 per cent increase, far higher than the 1.7Ghz (6.8GHz effective) we managed with our stock sample. At these frequencies, our power draw test took our peak system power consumption up to 284W.
Sauron_Daz wrote:The best way to teach a lesson: allowing people to make their own mistakes.
This does not apply to politicians!!!
Sauron_Daz wrote:I'd guess you don't have any children in their "teens".
Users browsing this forum: No registered users and 5 guests