Overclocking the GTX 690
Overclocking the new Kepler GPUs is a bit different than what many of you are used to. I won’t go over all of the details this time around, but you should check out this page of our original GTX 680 review for more details on how and why overclocking works the way it does on the new architecture. Essentially though, using a combination of increasing the power target and applying a clock speed offset, we can overclock the frequency of the GPU while still utilizing the power characteristics of the core.
For our testing we are using the latest version (3.0.2) of EVGA’s PrecisionX software along with a new, GTX 690-specific skin. You’ll want to pay attention to the sliders labeled "Power Target" and "GPU Clock Offset" and how the corresponding monitor window reflects changes in the GPU clocks and GPU power percentages.
One interesting aspect of overclocking with the GTX 690 (and GTX 680) is that even ONLY adjusting the power target can result in some improvements in performance. Essentially you are telling the GPU that it can override its default TDP of 300 watts in an attempt hit higher clock speed with GPU Boost. Just pushing this up to the maximum of 135% we saw an increase in power consumption of 65 watts – without ever applying a clock speed offset.
With a 100 MHz offset, you can see in the above image that we were hitting clock speeds as high as 1150 MHz on both GPUs, in this case running Batman: Arkham City. There is all kinds of interesting information to be gleaned from these numbers – which we will dive more into later.
A couple of quick tests with our 100 MHz offset (the highest we were able to keep stable at the majority of the time) showed some modest increases in performance. 3DMark11 results increased about 11% while Metro 2033 average frame rates only increased by about 5%.
It would seem that with the advent of GPU Boost, overclocking your GPU might yield less of a benefit.