Conclusion

1rst Scenario: Clocked at the same frequency there is a gap of almost 10 % in performance between the GTX 670 and the GTX 680 in theoretical benchmarks such as 3D Mark 11. While in games this is another thing, we noticed only 3 % difference in this case, nothing to worry about I would say. The reason for thise differences can be found in the specs of the card. The GTX 680 simply has 192 more shaders than the GTX 670.

2nd Scenario: this one shows Almost the same like in the first scenario. You will only notice a big difference in synthetic benchmarks rather than with games where the difference in performance is low. For the record: at this point we have benched the cards like if they were under normal use.

3rd Scenario: We noticed that when you raise the TDP limit of the graphics card it will operate always at the max boost clock. For exemple our sample was always running at 1'241 MHz. No less than 104 Mhz higher than the official boost clock. From ''always at'' 1'241 Mhz and up to 1'241 MHz the performance gap is very low and you almost don't see it in games.

4th and last Scenario: Wow, in this situation where the cards don't have any Turbo there is quite a difference in both synthetic benchmarks and games. In fact it is 8 % and 15 % more performance for the GTX 680 on the GTX 670.

Conclusion: the GTX 670 gives the GTX 680 an hard time. Unless you have a banker neighbor we don't see why one should spend 120 - 150 Euros (est. CHF 144.- up to 180.-) more on a graphics card that performs almost the same in games. If you like to run synthetic benchmarks, like me, then you have a point and a real problem (just like me)!
Meanwhile we prefer to spend 20 - 50 more bucks on a GTX 670 that comes with a custom PCB, a factory overclocking and a better cooling system than nVidia's one. In this case you are just winning on all sides, performance, noise level and design at a lower price and power consumption.