Power Distribution in Three PCs (2012)

Six years ago we conducted an anlysis of power consumption and distribution in several desktop PCs. Today we repeat the analysis with three modern machines to see how things have changed.

February 26, 2012 by Lawrence Lee

A few years ago we analyzed the power consumption of six desktop PCs to uncover just how much energy they consumed and how the power was distributed within each machine along the power supply’s various rails. These computers were based on Pentium 4, Pentium D, Athlon, and Athlon 64 processors, dinosaurs by today’s standards. The results of our testing gave critical insight into power supply requirements at the time.

Since then energy efficiency has become an important factor in hardware selection. Today, a CPU, GPU, motherboard, hard drive, or complete PC review isn’t complete without power consumption measurements — it has become standard operating procedure for most tech sites. Not only does this help end-users pick out the right PSU, it also allows them to weigh the financial and environmental ramifications. Given this new climate we thought it would be interesting to revisit the subject with a set of modern PCs to see how things have changed.

An Extech True RMS Power Analyzer.

The first piece of equipment we’ll be using for our tests is an Extech True RMS Power Analyzer/Datalogger. This is an expensive tool that displays four readings (watts, power factor or VA, voltage or frequency, and current) simultaneously using True RMS voltage and current measurements of sine, square, triangular and distorted wave forms. It’s overkill as we’ll only be using it to measure the total AC power draw from the wall, but it does deliver better accuracy than simple wall meters like the Seasonic Power Angel and Kill-A-Watt.

This measurement will tell us exactly how much electricity is needed to use the system as well as give us a rough estimate of the output required for a power supply to run it. A higher draw not only means a heftier electric bill but also translates into a larger impact on the environment (from the resources consumed by the utility company to generate the extra energy) and a louder cooling system to deal with the increased waste heat of the various components inside.

A Fluke 36 Clamp Meter.

Another key to our testing is a Fluke 36 Clamp Meter. This device
measures the electromagnetic field around any wire carrying electricity
and translates it into a current readout in Amperes. It’s not a precision lab tool, being sensitive to RF fields, with a margin of error of about 1.9% (far less accurate than that of a common digital multimeter), but it is a safe and easy way to isolate the power being delivered along each rail (+12V, +5V, and +3.3V) to the various components of the system. Clamping the meter around each set of power cables is certainly more convenient than exposing wires to measure the current directly.

Determining the power draw along the three main rails will tell us whether the distribution within modern power supplies is appropriate. When we did this test six years ago, 75~90% of the total power consumption was on the +12V line, but most power supply units at the time had outputs just about even in terms of current between the three main rails (+12V, +5V, +3.3V). Today the current available on the +12V line is frequently twice that of the other two, more so on high capacity units.

TESTING METHODOLOGY

This time around we’ll be testing three desktop PCs of varying power requirements to get a good cross section of data. We have a high power Nehalem based system with a 200W+ HD 5870 graphics card, a medium power Sandy Bridge machine with an HD 6870, and finally a low-end Athlon II PC with an entry level HD 5550 GPU.

Intel LGA1366 Test System:

• Intel Core i7-965 Extreme
  processor. 3.2GHz, 45nm, 130W TDP.
• Gelid Tranquillo CPU cooler.
• Asus
  P6X58D Premium motherboard. ATX, X58 chipset.
• 3GB QiMonda
  memory. 3 x 1GB DDR3-1066 in triple channel.
• HIS HD Radeon 5870 iCooler V Turbo
  graphics card.
• Scythe Musashi GPU cooler.
• ATI Radeon HD 4870
  graphics card(s). (for CrossFireX configuration)
• Kingston
  SSDNow V solid state drive. 30GB, 2.5".
• Seasonic X-650 SS-650KM
  power supply. ATX, 650W.
• Microsoft
  Windows 7 operating system. Ultimate, 64-bit.

Intel LGA1155 Test System:

• Intel Core i5-2500K
  processor. 3.3GHz, 32nm, 95W TDP.
• Thermalright MUX-120 CPU cooler.
• Asus P8P67 motherboard. ATX, P67 chipset.
• Corsair
  XMS3 memory 2x2GB, DDR3-1600 @ 1333MHz, 9-9-9-24.
• AMD Radeon HD 6870 graphics card.
• GELID Icy Vision GPU cooler.
• WD VelociRaptor hard drive. 600GB, 10,000 RPM, 32MB cache.
• Seasonic X-650 SS-650KM
  power supply. ATX, 650W.
• Microsoft
  Windows 7 operating system. Ultimate, 64-bit.

AMD AM3 Test System:

• AMD Athlon II X4 630 processor. 2.8GHz, 45nm, 95W TDP.
• Xigmatek HDT-SD964
  CPU cooler.
• Asus M4A785TD-V EVO motherboard. ATX,
  785G chipset.
• 2GB
  Corsair Dominator memory. 2 x 1GB DDR3-1800 in dual channel.
• HIS HD 5550 Silence graphics card.
• Kingston
  SSDNow V solid state drive. 30GB, 2.5".
• Seasonic X-650 SS-650KM
  power supply. ATX, 650W.
• Microsoft
  Windows 7 operating system. Ultimate, 64-bit.

Measurement and Analysis Tools

• Cyberlink
  PowerDVD to play an H.264 video clip.
• HandBrake to re-encode an XVID file into H.264.
• Prime95
  to stress the CPU.
• FurMark
  to stress the GPU.
• Extech 380803 Data Analyzer to measure the total AC power.
• Fluke 36 Clamp Meter to measure DC power along the +12V, +5V, and +3.3V rails.

Testing Procedures

Each system is examined in various load states:

• idle
• playing H.264 video
• encoding video with HandBrake
• CPU load with Prime95
• CPU+GPU load with Prime95 and FurMark simultaneously

System power consumption (AC) is measured with an Extech 38080. Current flowing
through the +12V, +5V, and +3.3V rails is monitored using a Fluke 36. The individual
wires in the various cable sets from the power supply were separated and then
recombined according to the voltage they carried. In this way the total current
from each individual voltage line could be measured separately. The -12V and
+5VSB lines were not measured, as they carry so little current that they are
insignificant (typically well under 5W).

Estimating DC Power

Ambient temperature at the time of testing was 22°C.

LIGHT LOAD POWER TEST RESULTS

Most systems, especially those in the home, sit idle or are put under very little load during the majority of operation, performing mundane activities like checking e-mail, browsing the web, listening to music, and watching movies. For this type of usage pattern, saving 10W when idle is much more beneficial than 20W on load.

Power Distribution: Idle
System	i7-965 XE + HD 5870	i5-2500K + HD 6870	X4 630 + HD 5550	Pent. D 820 + GMA 950	Athlon 64 3500+ + 6800GT
+12V (8-pin CPU)	0.8A	0.7A	0.8A	4.1A	0.6A
+12V (6-pin GPU)	0.4A	0.3A	N/A	N/A	N/A
+12V Total	3.9A	3.2A	1.7A	4.6A	4.6A
+5V	0.9A	1.6A	1.8A	3.6A	3.0A
+3.3V	10.2A	2.5A	2.4A	0.7A	3.9A
Total DC Output	86W	55W	38W	76W	83W
Total AC Input	110W	77W	54W	–	–
Approx. Efficiency	78%	71%	69%

At idle, the Athlon II X4 630 system had the lowest total power draw at 54W AC (38W DC). The AMD CPU was paired with a low-end graphics card, but it should be noted that we previously estimated the idle power consumption of the HD 5550 at 15W, with the HD 6870 and HD 5870 at 18W and 22W respectively so the difference isn’t huge. The i5-2500K rig was also equipped with a VelociRaptor rather than an SSD, accounting for a 4W difference (the SSD used with the other two system uses almost nothing).

The Athlon II X4 630 system had the most even power distribution, as did the
Athlon 64 3500+ machine from six years prior, while the Core i7-965 XE machine
relied heavily on the +3.3V rail. The Pentium D 820 and 3500+ drew a substantial
amount off the +5V rail, but not so for today’s systems. For a dual core CPU,
the D 820’s total power consumption was incredibly poor compared to its modern
day quad core counterparts, especially as the Pentium system lacked discrete
graphics.

The efficiency of our Seasonic X-650 power supply was noticeably poor at lower
levels, but this is true of all power supplies. The efficiency curve of most
PSUs typically peaks at the halfway point of the unit’s maximum rated output.

Power Distribution: H.264 Playback
System	i7-965 XE + HD 5870	i5-2500K + HD 6870	X4 630 + HD 5550
+12V (8-pin CPU)	0.8A	0.7A	1.1A
+12V (6-pin GPU)	0.8A	0.5A	N/A
+12V Total	4.9A	5.3A	2.7A
+5V	1.5A	1.6A	2.1A
+3.3V	10.0A	2.7A	2.6A
Total DC Output	100W	82W	52W
Total AC Input	126W	105W	72W
Approx. Efficiency	80%	78%	72%

Little CPU power was used in the playing of H.264 video thanks to hardware acceleration being present in the Radeon HD 5000/6000 series. The minor increase in power draw over idle was mostly restricted to the +12V rail of all three systems with the GPU doing much of the heavy lifting.

HEAVY LOAD POWER TEST RESULTS

The amount of power consumed on heavy load is a determining factor in how much capacity is required for a power supply and how much cooling is required to keep things running smoothly. Like all electrical components, CPUs, GPUs, and motherboard VRMs aren’t 100% efficient, resulting in waste heat during operation. The more energy the system draws, the more heat needs to be dissipated, so bigger heatsinks and faster, louder fans might be required.

Power Distribution: HandBrake Video Encoding
System	i7-965 XE + HD 5870	i5-2500K + HD 6870	X4 630 + HD 5550
+12V (8-pin CPU)	7.5A	4.2A	5.2A
+12V (6-pin GPU)	0.4A	0.3A	N/A
+12V Total	11.2A	6.5A	6.2A
+5V	1.5A	1.7A	2.9A
+3.3V	9.7A	2.4A	2.4A
Total DC Output	175W	96W	97W
Total AC Input	204W	123W	119W
Approx. Efficiency	86%	78%	82%

Our first heavy load test is encoding a video using HandBrake, a tool commonly used to re-encode video formats so that they’re compatible with various devices. It’s purely a software encoder, using CPU power to do all the work. As a result, almost all the extra load came on the +12V lines with some marginal increases on the +5V rail for the i7-965 and X4 630.

For many users, this may be the most stressful thing performed on their system and yet the total system power consumption was quite low, about 120W AC for both the Sandy Bridge and Athlon II based machines. The i7-965 is a bit antiquated and inefficient and even it used only ~200W.

Power Distribution: Prime95
System	i7-965 XE + HD 5870	i5-2500K + HD 6870	X4 630 + HD 5550	Pent. D 820 + GMA950	Athlon 64 3500+ + 6800GT
+12V (8-pin CPU)	8.9A	4.8A	6.2A	11.0A	4.0A
+12V (6-pin GPU)	0.4A	0.3A	N/A	N/A	N/A
+12V Total	12.7A	7.3A	7.2A	11.4A	7.8A
+5V	2.1A	2.6A	3.5A	3.5A	3.8A
+3.3V	10.2A	2.2A	2.6A	0.6A	3.8A
Total DC Output	198W	109W	113W	155W	125W
Total AC Input	229W	134W	134W		–
Approx. Efficiency	86%	81%	85%
Note: The older Pentium D and Athlon 64 systems were tested with CPUBurn rather than Prime95.

Using Prime95, a synthetic CPU stress tool, again most of the extra power is derived from the +12V rail with some moderate increases on the +5V rail. The total power consumption was not that far off from our video encoding test, about 10~25W more.

Power Distribution: Prime95 + FurMark
System	i7-965 XE + HD 4870 CrossFireX	i7-965 XE + HD 5870	i5-2500K + HD 6870	X4 630 + HD 5550
+12V (8-pin CPU)	9.0A	9.2A	4.9A	6.0A
+12V (6-pin GPU)	15.2A	12.1A	7.4A	N/A
+12V Total	36.0A	28.3A	17.8A	9.3A
+5V	2.3A	2.3A	2.5A	3.5A
+3.3V	12.2A	10.6A	2.6A	2.7A
Total DC Output	484W	389W	237W	139W
Total AC Input	547W	438W	271W	162W
Approx. Efficiency	88%	89%	87%	86%

The real party didn’t begin until we threw FurMark into the mix, a tool used to stress graphics cards more than actual gaming. Once again, the +12V rail ruled supreme, but this time the 6-pin PCI-E power cables provided much of the extra load. The HD 5870 and 6870 are fairly power hungry, doubling the total power consumption of their respective systems, while the HD 5550 didn’t make much of a dent.

Our video card selection is somewhat limited (the HD 5870 is our highest draw card, using about 220W on load) so to show you what you can expect with a more high-end configuration we added results with a pair of HD 4870s in CrossFireX on the i7-965 machine. The older 4870s are notoriously inefficient, taking the AC input of the Nehalem PC up to 547W. Even combining components with intentionally power sucking hardware, our 650W power supply was only pushed to about 75% capacity.

This is the type of activity that allows power supplies to reach peak efficiency
levels, with our test unit exceeding 85%. It should be noted that in our X-650
review, efficiency reached up to 92%. The discrephancy is likely the result
of a combination of three factors: the difference in load distribution in our
power supply testing, the different type of load (resistive) we use to stress
power supplies, and to a lesser degree, the inaccuracy of our clamp meter. As
a result the real world efficiency was up to 10% lower in some cases.

POWER DISTRIBUTION

Total power is one factor to consider when choosing a power supply, but equally important is the distribution along the three main rails, +12V, +5V, and +3.3V. In years past the available power was divided almost evenly (in terms of current) between the rails, though most components (e.g. CPUs, GPUs, hard drives) now draw the majority of their power from the +12V line. If half of a power supply’s rated output is on rails that are mainly unused, its effective capacity can decrease significantly.

Power Distribution by Percentage: CPU Load
System	i7-965 XE + HD 5870	i5-2500K + HD 6870	X4 630 + HD 5550	Pent. D 820 + GMA 950	Athlon 64 3500+ + 6800GT
+12V (8-pin CPU)	54%	53%	66%	85%	38%
+12V (6-pin GPU)	2%	3%	N/A	N/A	N/A
+12V Total	77%	81%	77%	88%	75%
+5V	5%	12%	15%	11%	15%
+3.3V	17%	7%	8%	1%	10%
Based on power (watts) drawn from each line.

Despite the passage of time, the power drawn from each line is more or less
similar to the high-end systems from six years past. The +5V and +3.3V lines
play only small supporting roles when it comes to power distribution. The +12V
rail pulls the bulk of the power, though the proportion coming through the 4-pin/8-pin
AUX12V/EPS12V cable has changed. Intel CPUs now pull a greater percentage of
power from the main 24-pin ATX connector, while AMD CPUs seem to be doing the
opposite.

Power Distribution by Percentage: CPU Load + GPU Load
System	i7-965 XE + 2 x HD 4870	i7-965 XE + HD 5870	i5-2500K + HD 6870	X4 630 + HD 5550
+12V (8-pin CPU)	22%	29%	25%	52%
+12V (6-pin GPU)	38%	38%	38%	N/A
+12V Total	89%	88%	91%	81%
+5V	2%	3%	5%	13%
+3.3V	8%	9%	4%	6%
Based on power (watts) drawn from each line.

GPU power consumption is another shift from the time of our previous tests with modern graphics cards often packing more transistors than CPUs. On full CPU + GPU load, 38% of the total power output was delivered through the 6-pin PCI Express cables feeding the graphics cards paired with our Nehalem and Sandy Bridge configurations.

Note that these aren’t particularly high-end graphics cards either, as the HD 6870’s current street price is only ~US$160 and is considered a modest GPU in gaming circles. Using a low-end video card or integrated graphics allows one to cut their power requirements by more than half.

CONCLUSIONS

It is important to keep in mind that the measurements presented here are
continuous loads. Our test equipment does not have the resolution
to measure peaks, which may last for 10 ms or less and may be much higher than
the continuous load. Most power supplies are rated for a continuous
load with allowances for higher peaks, but the internal protection circuits
may still be tripped by an exceptionally high peak. It is wise to leave perhaps 30% headroom
for peaks when sizing a power supply.

With these caveats, some broad, predictable conclusions can be drawn:

1) The +5V and +3.3V lines remain relatively unimportant for desktop PCs.
In our systems the +5V line never drew more than 4A under any circumstances.
The +3.3V line was underutilized as well, drawing less than 3A except in the
case of the Core i7-965 which belongs to the all but obsolete LGA1366 platform.
Even if LGA1366 was still mainstream, the peak draw was only 12A; Most power
supplies, even low wattage models, offer 20A or more.

2) The +12V line is more important than ever thanks to the growing power
requirements of graphics cards. On CPU load alone, our systems were drawing
about 80% of their power from the +12V rail, more or less the same as the
high-end equivalents from six years ago. On combined CPU and GPU load, the
portion was close to 90% and total power consumption effectively doubled.
Even a sub-US$200 card can raise the power requirements of a system
dramatically.

3) Excluding high-end graphics cards and/or multiple GPUs, it appears a 400W power supply is more than sufficient for most systems. Unless you’re choosing components like the Core i7-965 intentionally for inefficiency, even a fairly powerful gaming system will not require a 500W+ power supply. Our Core i5-2500K and HD 6870 combination at full load used less than 240W DC while the +12V rail topped out at just 18A.

NOTE: None of the above conclusions are meant to suggest that power
delivery alone are the only criterion by which a PSU should be chosen. We
are only considering adequate power delivery. We have not touched on noise,
efficiency, cooling, voltage regulation — in short, all of the other
relevant criteria we examine in our PSU reviews.

* * *

Discuss
this article in the SPCR Forums.