NVIDIA jumped to the top of our TPR graphs when we posted our GTX 580 in SLI article. It wasn't a surprise, as in single form the GTX 580 1536MB is faster than the HD 6970 2GB. Scaling for NVIDIA has always been strong and while I think AMD have pretty much caught up with both companies offering nearly 100% gains in some areas, NVIDIA still has the raw power of the GTX 580 on their side.
We came to two conclusions really when we tested the GTX 580 in SLI. It's DAMN FAST! - And it's going to play pretty much anything you throw at it, at any resolution, at any level of detail.
So you ask; why bother overclocking a setup so fast? Well, for starters, because we can. Mainly, though, for the simple reason we love power. Overclocking gives us more of that and we want to find out exactly how much more. For so many people a single GTX 580, GTX 570, or HD 6900 series is probably more than enough performance at stock. We ultimately overclock for the same reason people buy two, three or even four of these high end models. We love performance!
When we overclocked the MSI GTX 580 by itself we ended up with a core clock of 900MHz and memory clock of 4200MHz QDR. Now, we know that when we add a second card into the mix we tend not to be able to get the same kind of overclock due to the added heat from running two cards together.
Just for the fun of it, though, we made that our starting point and thought we'd work backwards.
When we set that and ran a full loop of Vantage without an issue, we were pleasantly surprised. Just to make sure everything was ok, I fired up Heaven. This is a great benchmark for testing overclocks; the large pressure on the tessellation engine and just the general intensity of the benchmark means if it can pass this, it'll pass everything.
1680 x 1050 and 1920 x 1200 both completed without an issue. Moving higher on the clocks, we ran into some problems, but at the same clocks we got for the single card, the setup ran perfectly.
What we ended up with were the same clocks which are 900 MHz on the core, 1800 MHz on the Shader and 4200 MHz QDR on the 1536MB of GDDR5 memory; a nice increase from the reference 772 MHz / 1544 MHz / 4008 MHz QDR setup.
Of course, we like to double check that SLI is enabled and you can see in the above GPU-Z screenshot it is. Hopefully the overclock will help yield us a nice performance increase, so let's get into our test system to find out.