Power Consumption and Performance in Nominal Mode

However, since we came to talk about energy-efficiency, let’s measure the power consumption using Extech Power Analyzer 380803 device. This device is connected before the system PSU, i.e. it measures the power consumption of the entire system without the monitor, including the power losses that occur in the PSU itself. When we took the power readings in idle mode, the system was completely idle: there were even no requests sent to the hard drive at that time. We used LinX program to load the CPU. This time we didn’t just measure the power consumption in idle mode and during maximum CPU utilization. For more illustrative picture we also created a graph showing the power consumption growth depending on the increase in CPU utilization as the number of active computational threads in LinX changed. We are going to compare the performance of DFI LANPARTY DK X58-T3eH6 against that of Gigabyte GA-EX58-Extreme. To ensure that the comparison is fair, we enabled “Set VR Current Limit Max” parameter on DFI LANPARTY DK X58-T3eH6, i.e. even under heavy load both mainboards increased the frequency of our Intel Core i7-920 processor to 2.8 GHz.

The results turned out pretty interesting. In idle mode the power consumption difference between the two boards is not very big and makes only 10 W. However, as the workload increases, DFI LANPARTY DK X58-T3eH6 loses more and more, so that the maximum difference starts to exceed 40 W. It is a pretty serious number already. Of course, there can be multiple explanations to that. For example, the efficiency of boards’ processor voltage regulator circuitries may differ dramatically. However, I believe that the increased Vcore on DFI LANPARTY DK X58-T3eH6 also contributed to this defeat. The higher the CPU utilization, the lower drops the processor core voltage on Gigabyte GA-EX58-Extreme, while on the DFI board it, on the contrary, increases above the nominal leading to increased power consumption.

But how will the two boards compare in terms of performance? During our DFI LANPARTY JR X58-T3H6 tests, this comparison didn’t produce the best results. They could be partially explained by the known issue with Intel Turbo Boost technology implementation, which wouldn’t allow the board to increase the processor clock multiplier under heavy load. Therefore, this time we are not going to deal with this implementation and will go back to multiplier increase to 21 with the help of the above described “Set VR Current Limit Max” parameter. Let’s check out the results now:

Don’t’ be surprised at the significant advantage DFI LANPARTY DK X58-T3eH6 mainboard demonstrated in Everest tests. At first we tried to explain the lower performance of Gigabyte mainboards in these tests by the fact that they do not increase the clock frequency multiplier to 22 often enough. However, looks like this peculiarity is typical of all Gigabyte mainboards on Intel X58 Express chipset, because they always lose to their competitors in these tests. Why do we refer to this problem as “peculiarity” but not “drawback”? Well, the thing is that it only shows in these synthetic memory tests and nowhere else. The lower performance in Everest tests doesn’t prevent Gigabyte solutions from running at least as fast or even faster than their competitors in all other benchmarks based on real applications. Things turned out exactly the same this time, too: DFI LANPARTY DK X58-T3eH6 is not always too far behind Gigabyte GA-EX58-Extreme, but it is a pretty noticeable difference.

As a result, DFI LANPARTY DK X58-T3eH6 performance tests in nominal mode turned out not very pleasing. If we leave all settings at their default values, the board will be relatively economical, but will lose in performance tests because of specific Intel Turbo Boost implementation. If we change the way this technology works, the mainboard power consumption increases significantly, but performance-wise it gets close to the competitor but still cannot really catch up with it. Loet’s see if things get any better during overclocking.