Performance during Overclocking
I believe that you understand well by now why we picked MSI Eclipse Plus mainboard for our performance comparison against Intel DX58SO: they have a lot in common, for example, overclocking results obtained at 185 MHz base frequency using dynamic Intel Turbo Boost implementation. Both mainboards have Intel processor power-saving technologies up and running even though the voltage is increased, so not only the multiplier, but also the CPU core voltage gets lower in idle mode.
During work in heavy-duty multi-threaded applications the processor clock multiplier is at its nominal value of 20x.
Under moderate load Intel Turbo Boost technology increases the multiplier to 21x.
This is pretty much all the boards have in common, because during CPU overclocking on MSI Eclipse Plus mainboard we can no longer control “C State” parameter, because it becomes unavailable to us, while Intel DX58SO mainboard has no restrictions like that. As a result, under low workload when only one processor core was active, its multiplier increased to 22x.
The next graph shows the dynamics of the Intel Core i7-920 processor multiplier change during the tests performed with SuperPI utility. Before the tests started the CPU was idle and its multiplier was lowered to the minimal value of 12x. Right in the beginning of the test the multiplier increased to 21 and got as high as 22 from time to time. Upon completion of the test, the multiplier went back to its minimal value. Therefore, the record-breaking results obtained during the calculation of 8 million PI digits on Intel DX58SO mainboard are not surprising at all, because the CPU frequency sometimes exceeded 4 GHz!
We see a completely different picture when the test is performed on a system built around MSI Eclipse Plus mainboard. Disabled “C State” parameter wouldn’t let us even try increasing the multiplier to 22: it is always at 21x under heavy load.
The differences between mainboards are even greater in multi-threaded applications creating much heavier load. The graph below shows the changes in the clock frequency multiplier of Intel Core i7-920 processor during a one-hour stability test in Prime95. You can clearly see how uneven the load created by this test program is. At first it is relatively low, which allows Intel DX58SO mainboard to increase the multiplier to 21x pretty often. Then we see a long interval of higher load when the CPU is almost always working with its nominal multiplier of 20x. After that the load gets variable again and the multiplier is switching between 20 and 21 all the time.
Now it is clear why you most often see errors only at least 15 minutes into the Prime95 stability test, if there were none during the very first cycles. This is when the load increases and the overclocked system is no longer able to handle it.
And now let’s see how the same exact test went during CPU overclocking on MSI Eclipse Plus mainboard. Back then, when we tested MSI Eclipse Plus mainboard, we mentioned that the only case when the clock frequency multiplier of the Intel Core i7-920 processor would increase to 21x guaranteed, is under relatively low working load created by approximately two threads of Prime95 utility. Unlike Intel DX58SO, MSI Eclipse Plus mainboard is simply unable to increase the processor multiplier when all eight threads of Prime95 are working: it stays at 20x throughout the entire test.
Therefore, it is not surprising that Intel DX58SO mainboard is often ahead of its competitor. Although both mainboards employ dynamic Intel Turbo Boost technology, the specific implementation is different in each particular case. Intel mainboard allows the CPU to work with an increased clock multiplier much more often.
However, I have to make one very important comment about MSI Eclipse Plus as well as Intel DX58SO mainboard. We value our reputation of a credible information source that is why we never report any unconfirmed facts. That is why if we claim that the CPU was stable during overclocking, this is the way it is. It isn’t hard to check out the systems stability with static Intel Turbo Boost implementation. In this case, the multiplier for the Core i7-920 processor increases to 21 under any load, even the heaviest one. Obviously, if the CPU copes with it fine, then stability won’t be an issue under lower working loads.
Things look completely different when dynamic Intel Turbo Boost implementation is employed. The system decided on its own what multiplier to use and changes the processor core voltage accordingly. In fact, the stability tests we use in our suit perfectly for that. LinX utility performs calculations in stages; the load created by Prime95 also changes dynamically that is why the CPU switches the operational modes as well as voltage and multiplier combinations during the same test cycle. Even SuperPI tests checking the ability of our overclocked processor to work with 22x multiplier have been all completed successfully. However, I can’t exclude the probability of finding a certain voltage+ multiplier combination that may cause an error. During dynamic Intel Turbo Boost implementation common tests may be not enough. You may need to work in different applications for quite some time before it would be fair to claim that the overclocked system is in fact stable.