In this article we are going to analyse two things: first the clock per clock performance between the GTX 670 and GTX 680 both clocked at 1 GHz. To do so we have bypassed the GPU Boost so the graphics cards have a minimum and maximum clock frequency of 1 GHz. Second, we have raised the TDP of our GTX 670 to see how the GPU Boost will react and what it's impact on performance actually is. So let's find out what's really happening behind the scene, let there be light!

Introduction

Some of you may have noticed that from one website to another the performance of the GTX 670/680 isn't the same. It's mainly due to the GPU Boost. Wazzat? The GPU Boost has been first introduced by nVidia for the GTX 6xx series and acts like the well know Turbo clock on Intel and AMD processors. For example: the graphics processing unit of the graphics card see it's clock raising for a while until it reaches the TDP limit. However while Intel and AMD announce the Max. Turbo frequency nVidia is telling the lowest Boost frequency. This means that all nVidia cards from this series have a Max boost different but the same ''guaranteed'' Boost clock which is the official one you find in the specs and the retail box. The max boost clock maybe different from one card to another because of the manufacturing quality of the GPU, the components on the PCB and the PCB itself. For a - not that good - graphics card the TDP will be reached easily while with a very good card the TDP will be far away allowing the boost clock to go higher. In practice there is a difference in performance from one graphics card to another even if they are/look the same. We noticed that our GTX 680 sample was clocking itself up to 1097 Mhz which is 36 MHz more than the official boost frequency. While our GTX 670 Direct CU II Top was able to raise its boost clock by no less than 104 MHz from the guaranteed one: from 1'137 MHz to 1'241. Quite a boost hu? And you notice it in the performance.

So, how high does your card clock? Well it's simple, download the latest version of GPU-Z (0.6.2). Then go to the monitoring tab (sensors) and clic the GPU Clock box till it shows ''Max''. After that be sure that GPU-Z is set to refresh even in the background and launch any 3D app that is able to stress your GPU so the Boost turn on. the freqeuncy you should see in the gpu clock box match the following: ''Guaranteed Boost Clock'' + k x 13 MHz. With the Guaranteed boost clock equal to 1'059 MHz for the stock GTX 680 for exemple and k a positive integer value between 0 and 20 (let's be mad). Then you may ask why 13 MHz ? Well it seems that the boost clock is working in 13 Mhz increments that we are going to call ''steps''.

It is now a fact, that not all test you can read in the Internet were performed under standard conditions. Therefore there sometimes is quite a big measurement error. Following we wanted to measure the performance impact by making several testing scenario.

First of all a comparison clock per clock GTX 670 against her big sister the GTX 680. The the only difference between the two card will be only the amount of shaders available in the core as all the other specs are exactly the same. For the record the GTX 680 has 1'536 shaders while the GTX 670 gets 1'344.

In the second scenario we are going to measure the performance between a reference GTX 680 and a reference GTX 670. Assuming that the average GTX 680 GPU have a 2 steps greater Boost, ie an increase of 26 MHz over the guaranteed frequency boost we made sure that our card match this during our test. For the GTX 670 is a bit different. Using the GTX 670 DirectCU II we're going to simulate a reference GTX 670. Usually this card increases clocks by 78 MHz.

Last but not least a fight between our GTX 670 Direct CU II Top against herself. With the only difference is that we raised the TDP in one case. Thanks to the option (Power Target) that is available in the app GPU Tweak from Asus (Also available in the softwares from other brands such as MSI AfterBurner and EVGA Precision X) that allows the graphics card to clock the max turbo higher... or not! We will discuss this point in the conclusion.

Huho finally it's not finished. We decided at the last minute to add one more scenario. What if you are as lucky as Donald is? The GTX 670 you just bought doesn't go further than the official boost frequency. In this case you will see how such a bad GTX 670 will perform against a GTX 680. Both cards have been clocked so they don't clock themselves higher than their official boost frequency.

Following you will find a summary regarding the clock frequencies we tested. The test setup and some GPU-Z screenshots can be found on Page 2. While on page 3 & 4 you will find the synthetic benchmarks results. The ingame performances can be found on page 5 to 8. En finally on page 9 we made a summary (average performance) of all results and page 10 there is a little conclusion. Enjoy!