We performed our power consumption measurements in nominal and overclocked modes using Extech Power Analyzer 380803. This device is connected before the PSU and measures the power draw of the entire system (without the monitor), including the power loss that occurs in the PSU itself. In our latest reviews of LGA 1155 mainboards we started using regular applications to increase utilization, in order to emulate different stress levels during every-day work. However, in order to load Haswell processor, we had to turn back to LinX utility, which uses AVX instructions. This utility does load the CPU heavier than normal, but we do not add a directed flow of hot air or open flames to it. If there is a program that can heat up the processor more than anything else, then the chances are that there is another program like that somewhere out there. This is exactly why we test the stability of our overclocked system and then additionally increase the CPU utilization with LinX utility during power consumption tests.
The numbers are quite high. The Haswell needs more power than its predecessors at high loads, both at default settings and when overclocked, even though its overclocking potential is lower. We made our CPU stable at a clock rate of 4.4 GHz by fixing the voltage at 1.180 volts. We couldn’t achieve better results by changing the voltage in any other way.
We want to remind you that the power consumption test is carried out at maximum CPU load. If we take ordinary applications and their CPU load, the power draw is going to be 110 to 115 watts rather than 150 watts as shown in the diagram. On the other hand, we only load the CPU in this test, so a graphics-heavy application may need more power by using the GPU. For example, the total power consumption of our testbed with an AMD Radeon HD 7970 graphics card was much higher than 200 watts when running 3D games.
The computer doesn’t do anything in the idle mode. We wait for it to cease any boot-related activities and stop accessing the disk. Besides measuring the power draw of the system with default and overclocked settings, we also measured it in the eco mode. The latter means the same default settings of the mainboard except that we manually changed the values of power-saving options from Auto to Enabled. This small change produced an impressive outcome.
The Haswell is amazingly economical in the idle mode. While consuming more power than its predecessors at high loads, it need much less than them when idle. Another thing we can note about the results is that there is no difference between the overclocked and the eco mode, although the former has a rather high and constant CPU voltage. The CPU voltage is usually dropped to 0.7 volts in idle mode and remains at 1.180 volts at overclocking, but there’s no difference in power consumption. Why?
The fact is that the CPU doesn’t really use the supplied power when it is idle. A higher voltage means a higher power draw at high loads – the relation is not even linear but quadratic – but when it comes to low loads, the voltage doesn’t matter since no power is consumed and dissipated. It is far more important to have functional power-saving technologies which disable unused CPU subunits. This factor affects the CPU’s power draw in idle mode, which is why we see such a large difference from the default mode where some of those technologies are not enabled. It turns out that the computer may consume considerably less power with enabled power-saving technologies and increased voltage than with reduced voltage and nonfunctioning power-saving technologies.
So we have no reason to worry that we fixed the CPU voltage at a certain level for overclocking instead of increasing it in the offset or adaptive way. It can save us some power and reduce temperature at high loads because the voltage is not raised by the unbalanced CPU-integrated regulator. And at low loads, when the level of voltage doesn’t really affect power consumption, we can achieve maximum savings by simply turning on each and every power-saving technology. And we never forget to do that.