Testbed and Methods
We carried out our tests on a testbed that included the following components:
- Gigabyte GA-X58A-UD5 rev. 2.0 mainboard (LGA1366, Intel X58 Express, BIOS version FB);
- Intel Core i7-930 CPU (2.8 GHz, Bloomfield D0);
- 3 x 1024 MB Kingston HyperX DDR3-1866, KHX14900D3T1K3/3GX, (1866 MHz, 9-9-9-27 timings, 1.65 V voltage);
- HIS HD 5850, H585F1GDG graphics card (ATI Radeon HD 5850, Cypress, 40 nm, 725/4000 MHz, 256-bit GDDR5 1024 MB);
- Kingston SSD Now V+ Series (SNVP325-S2, 128 GB);
- DVD±RW Sony NEC Optiarc AD-7173A optical drive;
- Scythe Mugen 2 Revision B (SCMG-2100) CPU cooler;
- Zalman CSL 850 thermal interface;
- CoolerMaster RealPower M850 PSU (RS-850-ESBA);
- Open testbed built using Antec Skeleton system case.
We used Microsoft Windows 7 Ultimate 64 bit (Microsoft Windows, Version 6.1, Build 7600) operating system, Intel Chipset Software Installation Utility version 126.96.36.1995, ATI Catalyst 10.9 graphics card driver.
We want to note that there are changes in our testbed configuration. We replaced our old OCZ GameXStream OCZGXS700 power supply with a higher-wattage and higher-efficiency CoolerMaster RealPower M850. Our quiet, fast and high-capacity (2 terabytes) hard disk drive Seagate Barracuda XT ST32000641AS is now replaced with a 128GB Kingston SSD Now V+ Series SNVP325-S2/128GB solid state drive, which is faster and absolutely silent. We have also switched back from the new but not quite perfect memory kit Kingston KHX12800D3LLK3/6GX to our old and time-tested Kingston HyperX DDR3-1866 KHX14900D3T1K3/3GX.
Operational and Overclocking Specifics
We didn’t have any problems assembling our testbed configuration around our Gigabyte GA-X58A-UD5 (rev. 2.0) mainboard and running it in default mode. However, there are a few peculiarities that must be mentioned as they may have a certain effect on performance and power consumption. You can see one as soon as you launch some diagnostic utility. The mainboard sets its base clock rate higher than usual in default mode.
The difference is small as the base clock rate is set at 135 MHz instead of 133 MHz, but the related frequencies grow up as well. We mean the CPU, memory, QPI and Uncore frequencies. There is nothing wrong with that and we have seen many mainboards from many manufacturers do the same. Such overclocking has no effect on stability. However, if this Gigabyte mainboard turns to be slightly faster than its opponents in default mode, we will know why.
A second thing we must mention is not specific to this mainboard. It is about the Intel Turbo Boost modes as they are implemented in nearly all LGA1336 mainboards: the CPU works not at its default frequency most of the time. The Turbo Boost technology increases its frequency multiplier by x1 even at maximum load.
It doesn’t mean there is a conspiracy among mainboard makers who violate the principles of Intel Turbo Boost and make CPUs work at higher clock rates by using some tricks. The multiplier should be increased by x1 even when all the CPU cores are working if the CPU’s power consumption and heat dissipation do not go beyond certain limits. During our tests we use high-performance CPU coolers and junior CPU models, so we suspect we meet that requirement. Therefore, Intel Turbo Boost indeed has a reason to keep the CPU frequency increased.
Then, there is a BIOS parameter that allows the CPU to enter deeper power-saving modes, but it is usually turned off by default. This option comes under different names: “Intel C-STATE Tech” in ASUS mainboards or “C3/C6/C7 State Support” in Gigabyte ones. The point of the technology is in switching more CPU subunits into power-saving mode when the CPU is idle. These changes can be easily spotted without any tools. For example, we have shown you above a screenshot of the AIDA64 program where the CPU is idle, its multiplier is reduced to x12, and its voltage is 0.928 volts. It’s because we had enabled the C3/C6/C7 State Support before capturing that screenshot. If we had left that option at its default, the voltage would have been 0.944 volts. The difference in voltage seems negligible, yet the resulting difference in power consumption for our CPU is about 6 watts. Saving this much power for months and years will sum up to large numbers eventually, especially as there are millions of computers in the world whose CPUs are idle most of the time.
Moreover, with that power-saving technology turned off by default, Intel’s Turbo Boost cannot perform fully. It’s only when you enable C3/C6/C7 State Support that idle CPU cores may be turned off when only one core is necessary for the current task. The frequency multiplier of the single active core is increased by x2 rather than x1 then, which improves the CPU’s performance in single-threaded applications. So, we can have lower power consumption and higher performance – why aren’t the mainboards set up this way by default? The single explanation we can find is that recovering from the deep-sleep modes takes more time. However, the difference is just a fraction of a second and can hardly affect your everyday computer activities.
In our tests we always allow the CPU to switch into the deep-sleep modes when benchmarking at the default frequencies to make full use of Intel’s power-saving technologies and Turbo Boost. However, the periodic increase of the CPU frequency multiplier by x2 may provoke some instability in an overclocked system, so we turn the Intel C-STATE Tech option (or its equivalent) off and use Turbo Boost only for overclocking.
The Gigabyte GA-X58A-UD5 (rev. 2.0) had no problems overclocking our CPU to 3.9 GHz, just like other mainboards we had tested.
When the CPU is idle, its multiplier and voltage are reduced by means of power-saving technologies.