CPU, GPU

Varying the CPU

To start with, here is the consumption of the configuration when varying the processor model or in overclocking.


As you can see, its influence is significant. The difference between the highest and lowest figures in standby in Windows desktop is 64 watts and 131 watts in load with two to four sessions of Prime95. You may have noticed that with the X6800, it’s more the rise in voltage in overclocking more than the increase in frequency that adds to power use.

Of course, in addition to the processor, Prime95 also affects these figures on the memory and chipset level. If we want to isolate the processor, we will need the ampere meter, which allows us to measure energy use of the CPU block on the mother card, whose supply varies between 80 and 90%. Here, we had a figure of 45.3 watts for the E4300 in load, versus 158.4 watts for the overclocked QX6700.

Varying the GPU

The other significant parameter in computer energy use is the graphic card. We took measurements in the Pixel Shader test in 3DMark 2006 with Prime95 in order to have a constant CPU use independent of the execution speed of 3DMark06. This choice wasn’t arbitrary, because based upon our experience, this is what offers the highest most stable use.


Even with just Windows desktop, a 8800 GTX consumes 50 watts more than a X1950 Pro. This gap increases to 92 watts in 3D. As much as this last figure is comprehensible, in 2D there is some effort to be made by manufacturers to limit a consumption that really isn’t justified. The transition to SLI noticeably increases the load and allows the deduction of the consumption of one 8800 GTX : 82 watts in 2D, 172 watts in 3D.

Memory, HDD, CD, Sound, Ventilation

Varying memory

We now come to memory. From the start the computer was equipped with 2x1 GB RAM. We opted to increase the quantity by adding chips as well as increasing consumption to 2.25V instead of 1.8V.


Given the fact that it really isn’t possible to load the memory independently of the processor, the best option was to see the effect of going from 2 to 4 GB. At 1.8V, this varied between 9.6W in stand by and 10.5W in load, versus 10.8 and 12.8W in activity at 2.25V. Nothing too significant here.

Varying HDD & CD

For the hard drive, we added a Raptor 150 GB as well as a Samsung 250 GB and measured consumption in rotation and also in access via IOmeter.


Adding a Raptor 150 GB results in a surplus of 10.5W in standby when measuring from the plug and 14.5W in load. The Samsung 250 GB, functioning at 7200 Rpm, is logically lower and gave us 9.3 and 12.3W.

With the help of the ampere meter, we looked at how the +5V and +12V were distributed and see that the Raptor uses 4.1 and 7.2W, respectively, on its lines, while for the Samsung these figures are 3.2 and 7.2W.

What is the consumption of the CD-RW Lite-On while reading a CD ? Without a CD inserted, it’s at 2 watts (on the 5V), a CD in rotation at 7.4W (4.9 on the 12V), and in reading at 14W (10.9 on the 12V.

Varying sound

As you can see, this has little impact. In stand by, the sound card adds a use of 3.8W (measured at the power source) and an additional 3W when reading an AAC 128 kbps in Itunes. A test in DirectSound 3D didn’t show a significant increase given these low numbers.

Varying ventilation

Given the low consumption of fans, we measured this parameter solely with the ampemeter. Two models were tested, an Enermax 120mm with its dial set at 0.30A in 12V, and a Scythe 120mm, which was at 0.18A. In practice, the Enermax varied between 0.19 and 0.29A (or 2.28 and 3.48W) versus 0.12A for the Scythe (or 1.44W). As you can see the impact is negligible.

The configuration

The configurations

To go further in our tests, we then looked at the precise consumption of 3 configurations. To do this, we took measurements in stand by and load of Prime95+3DMark06 at the power source with our wattmeter. We then found the different amperage directly at the power output. In fact, when a power supply is specified at 400 watts, this isn’t the maximum power that it consumes, but rather the maximum that it can supply. With an efficiency of 80%, for 400 watts it takes 500 watts for its maximum supply.

Our first « mid range » configuration for practical reasons had the same card and main drive, which are more high end:

- ASUSTeK P5N32-E SLI Deluxe (nForce 680i SLI)
- Intel Core 2 Duo E4300
- Ventirad Intel box
- Sapphire Radeon X1950 Pro 256 MB
- 2x1 GB DDR2 800
- Western Digital Raptor 74 GB
- Lite-On CD-RW 48x
- Enermax 120mm at its minimum
- Creative Sound Blaster Audigy

At rest in Windows desktop, consumption reached 130.1 Watts and went as high as 218.1 watts in load. This is what we got when looking at the different lines:


In the end, the power source actually supplied 183.5 watts in load with 72% on the 12V. The power efficiency was therefore at a very good 84.1%. Just to be clear, a 250 watts supply, (so that we have a little room) is largely enough to power this configuration, under the condition of course that power supply specifications are respected.

We now move on to our high end where a X6800 replaces the E4300, we add an HDD Samsung 250 GB, Enermax ventilation is at a max (given that its efficiency equals two silent fans), and the X1950 Pro is left aside for an Albatron GeForce 8800 GTX.

- ASUSTeK P5N32-E SLI Deluxe (nForce 680i SLI)
- Intel Core 2 Duo X6800
- Ventirad Intel box
- Albatron GeForce 8800 GTX
- 2 x 1 GB DDR2 800
- Western Digital Raptor 74 GB
- Samsung 250 GB
- Lite-On CD-RW 48x
- Enermax 120mm to its max
- Creative Sound Blaster Audigy

Measurements gave us 191.1 and 347.7 watts at rest and in load, respectively. The power supply was divided between the following:


Compared to our previous example, we notice a decrease in the 12V, which comes from the ATX plug. This is related to the fact that the 8800 draws less current than the X1950 Pro directly from the PCI Express port, which is compensated by PCI-E plugs. In the end, 285.8 watts is supplied, or an efficiency of 82.3%, 80% of which is on the +12V. We noted that with the transition of 8800 GTX in SLI with this configuration, the power source should supply 416 watts in load. Out of SLI, 350W was sufficient here.

Now, onto the ultra high end, where we have a quad core QX6700, which in addition is overclocked to 3.2 GHz in 1.5V (actually the CPU supports this frequency starting at 1.4V, but here this involved increasing power consumption to a max). A second 8800 GTX was added as well as another Raptor 74 GB and Maxtor 250 GB drive. We also moved up to 4 GB of DDR2-800, powered in 2.25V, and CPU cooling was provided by a Scythe Infinity.

- ASUSTeK P5N32-E SLI Deluxe (nForce 680i SLI)
- Intel Core 2 Duo QX6700 @ 3.2 GHz @ 1.5V
- Ventirad Scythe Infinity
- 2 x Albatron GeForce 8800 GTX
- 4 x 1 GB DDR2 800 @ 2.25V
- 2 x Western Digital Raptor 74 GB
- Samsung 250 GB + Maxtor 250 GB
- Lite-On CD-RW 48x
- Enermax 120mm at max
- Creative Sound Blaster Audigy

Consumption doubled compared to the previous configuration, because we are at 384.7 watts in stand by and 687.6 watts in load. The first surprising point was that in stand by this system consumed more than the previous one did in load. Of course, this is the ultra high end, but this proves that manufacturers might want take a look at the situation for these components under these conditions.


The power source actually supplies 567.6 watts, of which 85.3% are on the +12V, efficiency being 82.6%. So this configuration justifies the purchase of a 600-650 watt supply, and even one a little higher if you like to have a little room. The load on the +12V is relatively significant, and so it’s important to have a source capable of supplying enough amperage on this line. We noted that changing the CPU from 1.5 to 1.4V, which is more reasonable, allows us to gain a little more in consumption and the +12V goes from 40.3A to 37.9A.

Conclusion

Conclusion

As we can see in this article, the arrival of power supplies devoted to PC desktops and servers capable of 1200, 1000 or even 800 watts have little justification. In the worst case scenario, we couldn’t go beyond 687.6 watts of consumption, and in the end the actual power supplied was only 567.6 watts. This was in spite of using an overclocked and ultra high end configuration. Of course, certain extremely specialized systems in all areas, with water cooling and overclocked graphic cards modified in voltage could go further and justify a 700 or even 750 watt source.

On more normal configurations, tests showed that a power source of 250 watts was sufficient in the mid range, and 350 watts would be enough for the a high end configuration.


This takes us to the quality of the power supply. On the one hand, we can’t forget that +12V is predominant, as it represents 72, 80.4 and 85.3% of the power supplied in our test. It’s important that there is a proper division between lines and a 350 Watt supply that can only provide 200 Watts on the +12V would not be a good choice. Also, we need to think about not choosing a system that provides the bare minimum where tension is less stable and ventilation is noisy.

The difference in price between 400 and 350 watts isn’t that great (if the latter is available), and it would be a shame to deny ourselves the extra fifty watts. Only having an SLI based configuration really justifies the need for 500 watts, however, this is also the minimum power, at which manufacturers offer modular models, another possibly desirable characteristic. Anything higher than this would only be useful for ultra high end overclocked computers, as we saw above.

We finish this article with a final regret. It concerns consumption in stand by for desktops. If the figures are comprehensible in load, it seems that there could be a lot of room for improvement without reaching sleep mode levels. Knowing that a good part of even high end PC use isn’t necessarily too heavy in terms of load, lowering consumption seems to be a priority, from an economic as well as environmental point of view.