Power Consumption of Today’s Computers

I assembled two computers for my tests. They differ in their processors and, accordingly, in their mainboards and memory. These are typical configurations of modern top-end gaming systems, i.e. a single top-end graphics card, one gigabyte of memory, and a well-overclocked processor (the last thing may be not very typical, I should confess). The systems lack a DVD-RW drive, but this device doesn’t have a big share in the total power consumption of a computer much (moreover, it is usually used when the CPU and the graphics card are idle; I mean few people play Doom 3 burning DVDs in the background).

So, the Athlon 64 platform consisted of:

• ABIT Fatality AN8 SLI mainboard
• AMD Athlon 64 3800+ CPU (overclocked to 2.8GHz with Vcore increased by 0.1V)
• Zalman CNPS7700Cu cooler
• 2 x 512MB modules of DDR SDRAM from Corsair
• GeForce 6800 GT graphics card
• WD Raptor WD740GD hard disk drive

The Pentium 4 computer was assembled out of:

• ASUS P5WD2 Premium mainboard
• Intel Pentium 4 530 CPU (overclocked to 4.0GHz with default Vcore)
• Zalman CNPS7700Cu cooler
• 2 x 512MB modules of DDR2 SDRAM from Corsair
• GeForce 6800 GT graphics card
• WD Raptor WD740GD hard disk drive

To measure the consumption, I cut four standard 75-millivolt/20-amperes shunt resistors (that is, the voltage drop of a 20 amperes current is 75 millivolts on this shunt) into the power supply’s cables (in +5, +3.3, and in both +12V rails; I will denote the mainboard and hard disk’s +12V line as 12V1 and the processor’s as 12V2). The shunts don’t affect the work of the system as the voltage drop on them is much lower than the allowable fluctuations of the PSU’s voltages. In other words, the computer just doesn’t notice them. I measured voltage on the shunts with a Uni-Trend UT70D multimeter at an 80-millivolt margin. The multimeter’s accuracy is ±(0.05% + 1 position); the declared accuracy of the shunts is 0.5%, but this is quite acceptable for our purpose.

The main danger for the tester is that the multimeter is intended for measuring direct current, while the current consumed by a computer includes periodic fluctuations as well as separate impulses at various frequencies besides the constant constituent. For example, the graphics card changes its power consumption a little depending on the frame rate and movements of the disk drive’s heads generate a group of pulses at frequencies up to several megahertz, and so on. This is why precise measurements of the consumed current can only be done with an oscilloscope with a passband of 10MHz (I don’t invent this number, but take it from the manufacturers’ recommendations). But we are not interested in measuring the consumption with an up to a fraction of a watt precision. This precision makes no sense for us as it doesn’t give us new and useful information, so it is quite possible to use a multimeter instead of an oscilloscope, but with the following reservation: the computer must be tested in a sustained state, i.e. in periods of uniform CPU and graphics card loads, with idle heads of the hard disk drives and so on.

That’s why I won’t test transient processes like when the computer is powered up or the hard drive is spinning up. Each test was running continuously, and in the case of game tests I performed the measurements on a second launch of the game since at the first launch the game usually loads up from the hard drive parts of the game level as it advanced through it.

I installed Windows XP on both computers and this OS served as the first test – power consumption in the idle mode. Then I used two CPU-loading tests (S&M and BurnK7) and two tests that loaded both the CPU and the graphics card (Doom 3 and 3DMark03).