There's a difference between the maximum speed that a component can achieve and its optimal speed; we were able to push the CPU very hard, but stability in the benchmarks was non-existent. Tuning the system can take a while as trial and error shows what the best results are - the elegant dance between bus speeds, CPU multipliers, voltages and ratios can take some time. It is not uncommon to have the system "get angry" at you and not boot. When this happens, a quick hard power off and clearing of the CMOS gets you back in business.

One issue you can run into is having the system get greedy for higher voltages after pushing it a notch higher. Electrical hysteresis and other effects of physics can cause one setting to not work a second time after a CMOS clear and reboot. More than once we allowed the system to sit for a period of time to allow all the components to assume their natural states. This had the secondary effect of allowing the coolant to lose some energy as well, for more effective cooling when the system was restarted.

In Quad SLI, the highest we could run the graphics card without visual corruption was 610 MHz. This is 110 MHz or 22% above the stock clock frequency. The memory was able to sustain a frequency of 785 MHz or 30.8% above the stock figure. The processor was able to sustain high clock rates up to 3.73 GHz, but was not stable in conjunction with the graphics subsystem. The fastest we could run all of the tests with the cooling configuration we had was 3.66 GHz. While this was the fastest CPU speed, it did not mean that it was the fastest for all of the applications, though.