Examples of Our Overclocking Experiments
Algorithms, flow-charts, all this sounds very contemporary and very appealing, but sometimes you can’t see the wood for the trees – the picture doesn’t shape up from the succession of seemingly logical actions. Therefore, we decided to take an even closer look at Lynnfield CPU overclocking than we would normally do. Some specific examples could appear more illustrative than theoretical discussions and will depict the overclocking principles more clearly.
So, we have an Intel Core i7-860 processor. Its nominal frequency is 2.8 GHz, which means that it works at the default base frequency of 133 MHz with the multiplier of 21x. In reality, we haven’t seen this CPU ever work at its nominal speed at all. The static Intel Turbo Boost implementation is enabled by default and under any load the CPU clock multiplier increases to 22x, which results into 2.93 GHz frequency. If we enable dynamic implementation, we will see the same exact multiplier when three or all four cores are loaded with work. When only two cores are utilized, the CPU runs at 3.33 GHz with 25x multiplier, and with only one core working the multiplier hits 26x and the CPU works at 3.46 GHz.
We have determined in advance that Asus P7P55D Deluxe mainboard works at 210 MHz base clock frequency, which requires IMC voltage to be increased to 1.2 V. the memory voltage was set to 1.65 V, the frequency remained the same, we used the same 10x the memory frequency multiplier, because the nominal frequency of our Corsair Dominator GT CM3X2G2000C8GT modules was 2000 MHz and in reality this memory can do even better than that. We set the timings at 8-8-8-22-1T. All other voltages were at their default settings and the CPU Vcore was increased by 0.13125 V with enabled “Load-Line Calibration” protection against voltage drop under heavy load. There is a simple explanation of why we resorted to such an “unrounded” setting: the nominal Vcore for our CPU sample is 1.16875 V, which will provide us with a pretty “rounded” 1.3 V final Vcore.
At first let’s find out what our mainboard and processor are capable off with static Turbo Boost, when the CPU clock frequency multiplier increased only to 22x. We started with 175 MHz base clock, which produced 3.85 GHz CPU frequency. It is higher than 3.5-3.7 GHz recommended by our own methodology. However, we mentioned in the beginning of this article that we also tested the new processors in Gigabyte GA-P55-UD3 mainboard, so, we have already been through everything and know very well what our CPU is capable off.
When we launched LinX utility, its eight computational threads increased the CPU core temperature to 90 °C in just three test cycles. It was way too high. We stopped our tests and adjusted the processor core voltage increasing it by only 0.125 V instead of 0.13125 V. After that we launched the test one more time. Once again we hit 90 °C, but this time after ten test runs. However, it is still too high. Now we will only add 0.11875 V to the nominal Vcore setting but at the same time will increase the base frequency to 177 MHz. System passes the test but the temperature again hits 90 °C. We lower the voltage increase to 0.1125 V and this time the test finishes at 87 °C. Not bad, but could we increase the base frequency to 179 MHz at the same Vcore setting? No, we couldn’t, because the utility started to report errors, so we rolled back to 177 MHz base clock. Maybe we would be able to lower the core voltage even more then? Nope, we get errors again. So, this is how we found the maximum possible processor core voltage setting and maximum frequency at this voltage.
Finally, let’s optimize all other system parameters. We increase the memory frequency, launch LinX utility for a final test run and then a one-hour-run of Prime95 in Blend mode. This way, by adding 0.1125 V to the nominal CPU core voltage we overclocked it to 3.9 GHz. The memory also did very well and agreed to work fine at 2124 MHz with 8-8-8-22-1T timings. Asus P7P55D Deluxe mainboard reports slightly higher base clock that is why the processor and memory frequencies are even higher.
I believe it is a very decent result. We gained 1.1 GHz of frequency. At the same time, we retained all power-saving technologies, so the processor clock frequency multiplier and core voltage will lower in idle mode.
Now let’s try to overclock our processor using all advantages of dynamic Turbo Boost. Obviously, we can’t just switch to dynamics: at 177 MHz base clock with 26x multiplier the processor frequency will increase to 4.6 GHz and in our testing conditions the system will never be stable with these settings. So, we lower the base frequency to 161 MHz, but increase the voltage back to 1.3 V by adding 0.13125 V to the nominal. Our tests prove that under maximum load created by LinX utility the CPU core temperature remains within acceptable range. That is why at this point we get to tests with only one-two computational threads, when the processor clock multiplier increases to its maximum value of 26x.
All preliminary tests have been successfully passed, so we increase the base frequency to 165 MHz, but we see errors. We add 0.14 V and then 0.15 V to the CPU Vcore, but the errors are still there that is why we lower the base clock to 163 MHz. Unfortunately, we couldn’t achieve stability at this frequency, too, so we go back to 161 MHz. after a number of tests we find out that we need to increase the processor Vcore by 0.1375 V to ensure stable operation of our processor with 26x multiplier. Another run of LinX under maximum load – the CPU core temperature barely gets past 80 °C, so in this respect the voltage setting will work just fine. Now let’s increase the memory frequency, lower the timings and launch a one-hour run of Prime95 with maximum CPU utilization and use of eight computational threads. Another successful pass with 77 °C maximum core temperature. Now we repeat the test with only one computational thread: no errors and the core temp is only 60 °C.
As a result, the CPU will work with 22x multiplier at 3.55 GHz frequency under maximum load.
When only one core is loaded with work, the processor frequency will increase to maximum 4.2 GHz.
In idle mode both, multiplier and core voltage go down, due to power-saving technologies.