We attempted to check the overclocking capacities of all the boards, trying to obtain the highest possible clock for different voltages.
The notion of voltage varies a great deal from one motherboard to another, with each manufacturer manipulating the effective voltage (that measured with the sensor) according to the voltage requested (the VID sent to the processor). In its VRM 12/12.5 spec, in load Intel provides for a lowering of the voltage supplied as demand increases (the Vdroop concept). The effective voltage in load is then lower.
For some time manufacturers have been able to manipulate Vdroop in the BIOS using something known as 'LoadLine Calibration’. This option, which can sometimes be regulated, allows you to mitigate natural ‘loss’ of voltage in load, which can give the impression that the overclocking for a given VID is easier. In practice however this actually only consists of a manipulation of the final voltage, but the algorithm used by each manufacturer varies significantly and this can result in marked differences. Unless otherwise stipulated, we set the option at 50% on all the motherboards. We also activated Internal PLL Overvolting on all the boards. In practice the option didn’t allow us to stabilise further increases in clock.
In spite of the differences in the internal management of voltages for each board, we tried to look at the attainable clocks (we overclocked our 2700K using the multiplier, with a ratio of 47 below consisting of 4.7 GHz effective) by increasing voltage in 0.05V steps. Each mobo must be considered independently and can’t be compared lane by lane. In each case we give:
- The voltage requested in the BIOS
- The voltage read in load on the sensor
- The VID reading in load
- Maximum ratio reached (multiply by 100 to obtain the clock, 47 = 4.7 GHz)
- The energy consumption at the platform socket
We checked clock stability of each time in Prime95. We managed higher clocks in Windows that weren’t completely stable. Before starting, we wish to thank Martin Malik (author of the excellent hwinfo software) for his help on the subject of voltages.
No point pretending otherwise, the ASRock motherboard was the most complicated to overclock!
In spite of level 3 of LLC, the difference between the voltage requested and that obtained is high. In spite of everything, it was impossible to get up to 5 GHz at a requested voltage of 1.5V (which corresponds to more than 1.4 in load). By increasing voltage compensation (LLC2) we stabilised the processor at 5 GHz, but at the price of a very high real voltage and ridiculous power consumption. As with the missing automatic overclocking, ASRock doesn’t seem to have finished its BIOS with respect to overclocking.
The Asus mobo did better than the ASRock board, though it was slightly recalcitrant at the end.
It was hard to stabilise it at 5 GHz in spite of the fact that the voltage taken by the sensor was higher than that requested in the BIOS (though lower than the VID), with energy consumption remaining very respectable for 5 GHz however.
In spite of the scattered menus, it was simple enough to overclock the Gigabyte motherboard.
Note that we weren’t able to take a voltage reading on the sensor, which stayed blocked at 1.056. This voltage is reported using an additional chip that wasn’t yet correctly supported by the standard tools. Strangely we weren’t able to get a stable 4.9 GHz at 1.4V and had to go up to 1.425 to attain this clock and 5 GHz.
Along with the Gigabyte board, the MSI mobo was one of the simplest to overclock.
The increases in voltage are very gradual and energy consumption remains at reasonable levels.
Unsurprisingly, having 8, 10 or 12 phases on a board doesn’t make much of a difference. We managed to stabilise the processor at 5 GHz on all the motherboards tested.
Only the ASRock motherboard stands out for its less than optimum voltage, as we can see on the graph above.