Our Intel Core i7-920 Overclocking Guide claimed that systems on Intel Core i7 processors are fairly easy to overclock. I seem to have a different opinion on this matter. First of all, I have to say that a completely new approach to overclocking and the use of a totally new platform did make things a little complicated in the beginning. Moreover, we usually do not know the potential of only one system components out of three (CPU, memory, mainboard). We either investigate the overclocking potential of a new processor on a well familiar mainboard, or try to find out how well a mainboard can overclock our familiar CPU. This time I had to solve an equation in all unknowns by trial and error. I am not very well familiar with the overclocking potential of the new Intel Core i7-920 as well as with the overclocking abilities of both new mainboards.
Even though we have used OCZ memory modules for testing mainboards with DDR3 SDRAM support for a while now, we had to check out their overclocking potential all over again. We know what they are capable of at 1.95V voltage. However, it is not recommended to push the voltage over 1.65V during Core i7 CPUs overclocking and I haven’t yet tested our memory in this mode. Besides, it turned out that the third OCZ DDR3-1800 Platinum Series memory module that I took for the triple-channel configuration was slightly different from the pair I already had. Its SDP reported higher frequencies and more aggressive timings, which you could see in the BIOS screenshots above, because the settings for each memory channel are adjusted independently. The mainboard set 7-7-7-20-1T timings for the first two models and 6-6-6-16-1T timings for the third one.
Let me briefly remind you how we overclock Core i7 processors. We increase their so-called “base frequency” of 133MHz. All the other frequencies increase with it, because they are calculated by applying a corresponding multiplier to the base frequency.
The processor core frequency is calculated using base frequency and clock multiplier, which is in our case 133x20=2.66GHz. The frequency of UnCore unit that includes L3 cache memory, voltage regulators, QPI and memory busses equals 213MHz (133x16). The frequency of the QPI bus between the CPU and the chipset North Bridge equals 4.8GHz (133x36). Default memory frequency is set at 1066MHz (133x8).
You should lower the corresponding multiplier in order to prevent any excessive increase in one of the above listed frequencies. However, you can in fact lower only the memory and UnCore frequency, because QPI bus already uses minimal multiplier possible. So, we will have to increase the voltage to ensure that the system remains stable at higher frequencies. Gigabyte did a great job preparing a manual on voltages and their role in overall system stability (the illustration below has been taken from this manual):
At first, this is exactly what I did: lowered memory and UnCore frequencies and increased the voltages. However, then it turned out that I didn’t have to lower UnCore frequency, because a lot of processor units will work at a lower frequency and it will affect the overall performance quite noticeably. The system, however, tolerates UnCore frequency increase during overclocking just fine. Then it turned out that I didn’t have to increase the voltages, either. I didn’t have to increase any of the voltages that could theoretically improve the mainboard stability, namely QPI, VTT, PLL or chipset NB voltage. Although all frequencies got higher during overclocking, the system worked perfectly fine with the default voltage settings. At least, it was the case up until the base frequency hit 200MHz. However, I have to point out two things about the voltages. The first one is evident: you do have to raise the processor core voltage during significant overclocking. The second one deals with the memory.
Many manufacturers have already announced memory modules designed specifically for work in LGA 1366 platforms in nominal and overclocked modes. Namely, they can work at high frequencies and almost nominal voltage of 1.5V for DDR3. Our modules are not like that. They require a serious voltage increase during overclocking, up to 1.95V, which is unacceptable for Core i7 CPUs. Therefore, we increased the memory voltage to the maximum possible value of 1.64V. But even in this case we had to use the lowest x6 memory frequency multiplier. As a result, the memory frequency was around 1100MHz during our overclocking experiments. It is very low for DDR3 in LGA775 systems, but turned out quite ok for LGA1366. The memory controller is integrated into the CPU, so you have to b careful when you increase the memory voltage. However, it means that the memory subsystem latencies in LGA1366 platforms are considerably lower than in LGA775 platforms. Besides, the wider 192bit memory bus has its positive input, too. Since we couldn’t increase the frequency, we managed to lower the timings to 6-6-6-18-1T and reach acceptable performance level.