And now let’s run the same exact tests in overclocked mode. The differences in system settings are recorded in the following table:
All mainboards overclocked our test processor to 4.2 GHz, but the gaming Gigabyte G1.Sniper board could accomplish it only at a lower base clock frequency and therefore its memory also worked at a lower speed. As we have expected, overclocking parameters on Gigabyte GA-X58A-OC and Intel DX58SO2 were identical, sow e expected them to perform at about the same speed with Gigabyte G1.Sniper falling behind. However, in reality things turned out totally different.
The gaming Gigabyte G1.Sniper is indeed always behind overclocker Gigabyte GA-X58A-OC, which is quite logical judging by the overclocking results. However, despite our high expectations, Intel DX58SO2 was unable not only to maintain the same performance level as the overclocker mainboard, but it simply turned out the slowest board of the three, even though the overclocking settings of the gaming board are way worse. All this seems very strange at first glance. However, the explanation is simple enough. The AIDA64 screenshots that we used as illustration of our overclocking achievements do not show one very important parameter – the frequency of the North Bridge built into the processor also referred to as Uncore, which contains memory controller and L3 cache memory. All frequencies on an LGA1366 platform are derived from the base clock by applying different multipliers to it. The Uncore multiplier should be at least twice as high as the memory clock multiplier and by default I equals 16x. It means that with the nominal base clock of 133 MHz, the frequency of this unit is 2133 MHz. When we overclocked our system to 200 MHz base clock, overclocker Gigabyte GA-X58A-CO board increased this multiplier to 19x thus raising the Uncore frequency to 3800 MHz.
The gaming Gigabyte G1.Sniper board turned out even more aggressive than the overclocker one and raised the Uncore multiplier to 20x.
Intel DX58SO2 worked in the same conditions as Gigabyte GA-X58A-OC. It lowered the clock frequency multiplier to 12x thus reaching the lowest Uncore frequency and therefore finishing last.
Actually the fact that we overlooked such an important parameter as Uncore frequency and let the mainboard set it on their own is a serious omission on our part. All boards set their Uncore frequency by themselves and although all of them allow users to manually configure multipliers, we could have easily corrected the Uncore values. However, the obtained results are extremely valuable, because they once again showed us that leaving some important parameters in Auto mode is not just unreasonable, but sometimes even harmful.