You are here

Power Supply Fundamentals

by Mike Chin

  • April 4, 2007 - Added a new high power system and an A64X2 system to section on Real System Power Requirements (page 4).
  • Dec 1, 2006 - The Lists of Recommended PSUs have been separated from this article. This article was just getting too long and cumbersome. Separating them means it's a bit easier to update them more frequently.
  • Sept 2, 2006 - Added section on how to identify the PSU maker and a counterpoint to correct PSU sizing.
  • Aug 31, 2006 - Updated much of the text on Intel PSU guidelines based on latest Power Supply Design Guide for Desktop Platform Form Factors (Revision 1.0 - June 2006). Also expanded discussion on multiple 12V lines, efficiency, power factor correction, etc.
  • Aug 24, 2006 - Added discussion about efficient PSUs that don't start with some motherboards.
  • Aug 14, 2006 - Seasonic S12-330 review added. PicoPSU and Silentmaxx Fanless 400W MX460-PFL01 added. Antec NeoHE430 reinstated. Nexus 4090 & Zalman 400B retired.
  • Jan 3, 2006 - Zalman ZM460-APS, Fortron Green 400W and Seasonic SS300-SFD 80 Plus added. Enermax Noisetaker 325 & 475 retired; Antec NeoHE430 retired until proven reliable.
  • Oct 28, 2005 - Antec NeoHE430 added, adjustments in recommended list comments made to reflect new efficiency findings reported in Corrected Efficiency Results from PSU Test Rig V.3.
  • Major Update Sept 8, 2005 - Revisions in selected portions of the text, update version info on PSU design guides, added info on 12V reliance, dual 12V lines, more system power examples, added new models and retired some old ones.
  • April 12, 2005 - Numeric ranking system of PSUs revised.
  • Major Update April 5, 2005 - Mass retirement of many older PSUs. Added Seasonic S12, Enermax Noisetaker 701, Nexus 4090. Major revisions of the entire article.
  • Update February 4, 2005 - Added detailed information about real desktrop system power requirements. Minor adjustments in comments to PSU tables; removed two PSUs no longer available.
  • Update October 17, 2004 - Added three fanless PSUs: Antec Phantom, Silverstone ST30NF and CoolMax Taurus CF-300.
  • Update June 5, 2004 - Raised Seasonic Super series ratings after Rev.03 reviews, added other Enermax Noisetaker and Seasonic Super series models.
  • Update May 31, 2004 - Added Zalman ZM400B, links to useful sticky posts in the PSU forum.
  • Update April 4, 2004 - Added Enermax NoiseTaker 475. RSG Electronics PSUs removed: Not enough validation from firsthand or trusted sources about these models.
  • Update March 10, 2004 - Added section on Power Factor Correction.
  • Updated Feb 1, 2004 - Major changes to text content, inlcuding information about BTX Form Factor, and PSU output power ratings. Added Nexus NX3500. Seasonic Silencer rev. 02 inconsistency noted.
  • Updated Dec 8, 2003 - Seasonic Tornado downgraded for inconsistency; Fortron 350 Aurora added.
  • Updated Sept 2, 2003 - Added new Seasonic "Super" models, refined comments.
  • Updated July 10, 2003 - Quoted section "5.7 Acoustics" from ATX12V 1.3 guide.
  • Updated June 26, 2003 - proSilence PCS-350 fanless PSU added.
  • Updated June 19, 2003 - Reorganized ranking tables into 4 categories, refined comments, reshuffled rankings based on latest info and results.
  • Updated June 17, 2003 - Added Nexus NX-4000, info on new Intel PSU standards and microATX system design guide.
  • Updated May 11, 2003 - Added SilenX 14dB 400.
  • Updated April 7, 2003 - Added Verax 300.
  • Updated Dec 10, 2002 - Minor adjustments in table comments, note to ATX spec.
  • Updated Dec 8, 2002 - Antec TrueControl 550 added.
  • Updated November 29, 2002 - Nexus NX-3000 PSU added.
  • Updated November 15, 2002 - Enermax and SH models removed; quieter alternatives now plentiful.

* * *

Often one of the biggest noise makers in a PC, the Power Supply Unit delivers regulated DC voltages to various components. Computer PSUs are switching mode types, which provide relatively high efficiency at low cost. They utilize forced air cooling, usually an 80mm fan, and sometimes incorporate a second fan. The fan is the primary source of noise in a PSU. Coils in a PSU can buzz and hum, especially when pushed under high loads, but usually fan noise masks coil noise. Typically, the fan is rated for higher than the maximum airflow needed to keep the PSU cool. It's cheap engineering insurance for the manufacturer and also fulfills the case cooling role of the PSU fan.

In the ATX case specification, cool air is drawn into a typical case from vents in the front panel. The incoming air helps cool components as it moves through the case, becoming warm in the process. It is evacuated through the PSU and out the rear by the PSU fan. So the loud, fast fans do help to keep a case cooler. Manually varying a high airflow PSU fan can cause CPU temperature to be affected as much as 5-6 degrees C.

The ATX specification was created some years ago at a time when desktop CPUs generated no more than ~30W. Now, they are up over 130W. The airflow arrangement of exhausting hot case air out through the PSU no longer makes as much sense as it did in the past. The PSU has to handle both its own self-generated heat, which is naturally higher than before, as well as the heat generated by the other components.

Fast-spinning fans make a lot of noise, especially when confined in a small space with nearby airflow obstructions. It is not unusual for the noise of a PSU to be 12-15 dB higher than the rated noise of its fan in free air. The noise is further exacerbated by the way a PSU is mounted in a typical tower case: the typical 4~5 pound weight "hangs" off the top of the back panel on 4 screws. In this mounting configuration, excitation of case panel resonances by direct transfer of PSU fan vibrations is almost unavoidable. The end result is more noise, especially as a droning type of hum in the lower frequencies.

The quietest PSUs on our list (on page six if you want to jump straight there) feature either no fan at all or a fan that spins at low speed under most conditions. Keep in mind that components will tend to run a bit hotter than usual as a result of reduced airflow. This can be a concern if the normal ambient room temperature is high or if very hot components are used. The best fan-cooled models have low normal fan speed, and allow the fan to ramp up to full speed only when really necessary.

There are many different motherboard/case form factors, such as Mini-ATX, LTX, Flex-ATX, AT, Mini-ITX, etc. There are also many proprietary cases that don't conform to any general form factor. ATX power supplies can often be used, but some cases require different PSU form factors, such as STX for Flex-ATX cases. In time, we will expand our list to include different form factor PSUs.

UPDATES to INTEL's PSU DESIGN GUIDES

Intel's power department has been very busy in the last few years as rapid deloyment of higher power and different form factor devices keep adding to the PC mix. They have made frequent updates and additions their design guidelines for PSUs in the ATX12V, SFX12V, TFX12V, LFX12V and CFX12V PSU Design Guides on their site Desktop Form Factors. These guides are not standards that must be adhered to by regulation, but specifications that almost every PSU maker in the industry follows very closely in order to ensure compatibility between their products and new motherboards, graphics cards and other peripherals.

In summary:

  • NEW! Power Supply Design Guide for Desktop Platform Form Factors (Revision 1.0 - June 2006) Consolidated guide for CFX12V, LFX12V, ATX12V, SFX12V, and TFX12V PSU form factors
  • ATX12V (Revision 2.2 - March 2005) is for standard tower and desktop form factor systems.
  • SFX12V (Revision 3.1 - March 2005) is primarily intended for use with small form factor microATX and FlexATX systems.
  • TFX12V (Revision 2.01 - June 2004) - Thin Form Factor - for SFF system designs (9-15 liters in total system volume).
  • LFX12V(Revision 1.0 - April 2004) - Lowprofile Form Factor - for Balanced Technology Extended (BTX) form factor systems with ultra small form factor system designs (6-9 liters in total system volume).
  • CFX12V (Revision 1.2 - May 2004) - Compact Form Factor - for BTX system designs with ultra small form factor system designs (10-15 liters in total system volume).

Common themes in most of the PSU Guides are:

  • Increased +12 VDC output capability
  • Dual 12V rails for PSUs that have >20A current capability on the 12V line
  • 2x12-pin main power connector for additional 75W PCI Express requirements (Most v2.2 PSUs are coming with 20/24 pin ATX connectors for backward compatibility)
  • Minimum efficiency for typical and light load.
  • Higher recommended efficiency targets for typical and light load.
  • Details on an S-ATA power connectors
  • Definitions of the new 24-pin ATX connector
  • Acoustic guidance to support low noise systems.

The Recommended Acoustics has been revised since ATX 12V version 2.0. Section 5.1, page 33 of Power Supply Design Guide for Desktop Platform Form Factors, v.1.0 states:

5.7. Acoustics - RECOMMENDED

It is recommended that the power supply be designed with an appropriate fan, internal impedance, and fan speed control circuitry capable of meeting the acoustic targets listed in Table23.

The power supply assembly shall not produce and prominent discrete tone determined according to ISO 7779, Annex D.

Sound power determination is to be performed at 43 C, at 50% of the maximum rated load, at sea level. This test point is chosen to represent the environment seen inside a typical system at the idle acoustic test condition, with the 43 C being derived from the standard ambient assumption of 23°C, with 20°C added for the temperature rise within the system (what is typically seen by the inlet fan). The declared sound power shall be measured according to ISO 7779 and reported according to ISO 9296.

Table 23. Recommended Power Supply Acoustic Targets




Idle



Typical - 50% load



Maximum



Minimum


3.5 BA

4.0 BA

5.0 BA


Target


3.0 BA

3.8 BA

4.5 BA

The acoustic targets recommended here are not particularly low (quiet). As far as I know, no power supply maker is testing or reporting acoustics in the way recommended above.

DESKTOP PSU DESIGN GUIDE: Jan / June 2006

The latest PSU Guide consolidated all the desktop form factors into one: Power Supply Design Guide for Desktop Platform Form Factors revision 0.5 was released in January 2006. It was updated as Revision 1.0 in June 2006. This guide combines design guidelines for the ATX12V, CFX12V, LFX12V, TFX12V and SFX12V power supply form factors into one comprehensive power supply design guide.

The highlights of revision 0.5 were:

  • Combined CFX12V, LFX12V, ATX12V, SFX12V, and TFX12V content into one desktop power supply design guide
  • CFX12V content derived from revision 1.2

    - Updated 12V1 current for 300 W configuration

    - Updated efficiency loading for 300 W configuration
  • LFX12V content derived from revision 1.1
  • ATX12V content derived from revision 2.2
  • SFX12V content derived from revision 3.1
  • TFX12V content derived from revision 2.1

    - Updated 12V1 current for 300 W configuration

    - Updated efficiency loading for 300 W configuration
  • Updated Capacitive Load section to use standard capacitor values
  • Updated 5 VSB efficiency recommendations for Digital Office platforms
  • Removed power-down warning from power supply timing diagram
  • Marked sections with labels to indicate REQUIRED, RECOMMENDED, or OPTIONAL items

Changes in in revision 1.0:

  • Added 12V2 Current for Processor Configurations table
  • Added revision numbers to form factor specific chapters
  • Changed Input Line Current Harmonic Content to OPTIONAL to better reflect geographical requirements

THE PUSH TO HIGHER EFFICIENCY

Those who followed the evolution of the ATX12V Power Supply Design Guide (now v2.2) authored by Intel's power division may recall that it was the release of v1.3 in April 2003 that added efficiency guidelines at typical (50%) and light (20%) loads, and saw the minimum efficiency increased from 68% to 70%. Details for Energy Star and standby efficiency were also added in that version.

Version 2.0 of the ATX12V guide released February 2004 underwent a few more significant changes. Aside from those related to higher power requirements for more power hungry systems, section 3.2.5. for Efficiency listed not only the required minimum efficiency, but for the first time, the recommended minimum efficiency. Those numbers are substantially higher. The v2.01 spec of June 2004 remained unchanged from v2.0, but the current v2.2 spec of March 2005 saw substantial increases in both Required and Recommended efficiency.


AXT12V v2.01: "Minimum PSU Efficiency Vs Load"
Loading

Full

Typical

Light
Required Min

70%

70%

60%
Recommended Min

75%

80%

68%


AXT12V v2.2: "Minimum PSU Efficiency Vs Load"
Loading

Full

Typical

Light
Required Min

70%

72%

65%
Recommended Min

77%

80%

75%

In the current PS Design Guide for Desktop Platform Factors v1.0, efficiency guidelines have been raised even higher, and power factor correction has finally been introduced. The Optional Minimum Efficiency numbers come straight from the 80 Plus program guidelines. (More on PFC and 80 Plus below)


"Efficiency Vs Load"
for all Desktop PS types

in PS Design Guide for Desktop Platform Factors v1.0

Loading

Full

Typical

Light

PFC
Required Min

70%

72%

65%

-
Recommended Min

74%

77%

72%

-
Optional Min

80%

80%

80%

0.9

What is efficiency in a power supply? It is defined as the power loss in AC-to-DC conversion, expressed as a percentage of total AC input power. For example, a power supply that requires 100W AC input to produce 70W DC output has an efficiency of 70%. In this example, 30W is lost as heat within the PSU. A power supply that requires 100W AC input to produce 80W DC output has an efficiency of 80%. 20W is lost as heat. The 10W difference in these examples seems trivial. However, at higher output levels, differences in efficiency becomes quite significant in terms of how much energy is lost as heat, as the table below shows. The ideal efficiency is 100%, where AC input and DC output are the same, and there is no loss to heat in the power supply.


THE EFFECTS OF EFFICIENCY

Efficiency


Power Lost as Heat


@100W
output


@200W

output


@400W
output



85%


18W


35W


70W



80%


25W


50W


100W



75%


33W


67W


133W



70%


43W


86W


174W



65%


54W

108W

216W

The above table show clearly how much more heat is generated at higher power levels with even just a five percentage point drop in efficiency. When you consider the 10 percentage point differences, you can see nearly a doubling of power loss or heat generation. Cooling a power supply quietly becomes progressively easier as the power supply efficiency is improved.

Part of the push for higher efficiency comes from sheer necessaity. Intel (and AMD) systems are drawing so much more power now than they were even a few scant years ago that the 65% typical efficiency is no longer viable from a thermal management point of view. When the total AC power consumed was 100W, the heat generated by the PSU at 65% efficiency was just 35W. With some powerful systems drawing over 400W DC at high loads, the heat from a 65% efficient PSU would amount to >200W. This is a lot of extra heat heat to eliminate from computers. The generally shrinking size and tight internal spacing of PC cases makes cooling even tougher. An 80% efficient PSU in the same >400W DC system would generate less than one third of the heat, 100W. With system integrators everywhere concerned about the cost and complexity of adequate CPU and component cooling, reduction in heat had to start in the obvious places.

There is a certain confluence of factors leading to similar efficiency targets for PC power supplies. There's the aformentioned need to curb heat, being recognized and addressed by Intel's power team. Then there are increasing concerns about energy efficiency in light of world ecology and the challenges of electricity delivery. Here are several relevent orginizations and programs that are all pushing the efficiency envelope:

The Energy Star program of the US Environmental Protection Agency is proposing a new tough low target for AC consumption of computers in idle (50~60W for desktops) effective 2007. We first covered this story in March 2005, A New Energy Star... in 2007, and included an update on page 3 of the article, The State of the Industry, March 2006: Through Silent Eyes.

The 80 Plus Program run by Ecos Consulting is certifying high efficiency power supplies for eligibility in a national buy-down rebate program that returns $5 and $10 to system integrators who incorporate these PSUs in complete systems.

The related web site Efficient Power Supplies aims to "initiate a global dialogue about energy efficient power supplies." Run jointly by EPRI Solutions and Ecos Consulting, the site is an influential central clearing house for all things related to efficient power supplies, and even sponsors international competitions for high efficient PSU designs.

There's no question that higher efficiency is the trend in computer power supplies. In the past 24 months, the number of PSU samples that have been tested by SPCR to have > 80% efficiency has risen dramatically. As one commentator in the SPCR forums noted recently, the last three decades were a complete waste regarding PSU efficiency; there was almost no improvement. Then, in the last two years, it jumped from an average of 80% for higher quality PSUs. Looked at in reverse — from the poiint of how much electricity is wasted as heat — going from >30% to

The big payoff for those who loathe PC noise is that a PSU that generates less heat needs less forced airflow to stay cool. This means lower fan noise. It also means less heat leaking out from the PSU into the rest of the system, especially to the already hot CPU, which is usually located right next to the PSU, within a couple inches. Lowering the heat radiating from the PSU can help to lower the airflow requirement for cooling the CPU as well, leading again, to less fan noise.

HIGH 12V LINE RELIANCE

The high reliance of current systems on the 12V is dramatic compared to even just a couple of years ago, and the evolution of the ATX12V spec reflects this change. Almost any system assembled from current components will draw the vast majority of current from 12V, in some cases, as much as 90% at load. This is one of the factors in increasing efficiency. It is generally easier to obtain higher efficiency in the conversion from 120VAC to 12VDC rather than 5VDC or 3.3VDC.

We recently studied the power distribution in half a dozen systems of varied configuration to confirm the high 12V reliance first hand and reported our finding in the article, Power Distribution within Six PCs. In all the systems the current draw on the 5V and 3.3V lines was a maximum of just 5A! Note that in dual 12V line models, 12V2 is supposed to only supply the AUX12V (2x12V) 4-pin plug, which feeds only the CPU. 12V1 is supposed to supply 12V to all the other components that require it. This can be a potential problem in some high end gaming systems; see the sections on Dual 12V Lines on the next page.

BTX FORM FACTOR

BTX prototype, mATX equivalent sizeThe various BTX interface, system design and case specifications and studies released by Intel since September 2003 are major departures from the ATX form factor, but they are only just trickiling into the market now from a few vendors. FormFactors.org describes the BTX as follows:

"The BTX form factor specification gives developers options to balance thermal management, acoustics, system performance, and size in the system form factors and stylish designs that are desired in today's products. The BTX form factor is a clear break from previous ATX form factor layouts and was developed with emerging technologies such as Serial ATA, USB 2.0, and PCI Express*.

"Thermal improvements come primarily from taking advantage of in-line airflow. The BTX defined in-line airflow layout allows many of the main board components (i.e.: processor, chipset, and graphics controller) to utilize the same primary fan airflow, thereby reducing the need for, and noise from, additional system fans. In some cases this also allows fewer and/or less expensive heat sinks to be used when compared to ATX solutions. The system level acoustics are also improved by the reduced air turbulence within the in-line airflow system. The BTX layout supports better component placement for back panel I/O controllers — important as the signal speed of external devices continues to increase. In addition to smaller than microATX system sizes, BTX was designed to scale up to tower size systems using the same core layout by increasing the number of system slots included."

We will provide analysis of BTX cases, PSUs and systems as they become more widely available.

POWER SHMOWER

or How PSU Power Ratings Mean Almost Nothing

A frustrating fact about PSUs is that there does not appear to be a stringent or regulated standard for reporting, advertising and labeling rated power. This is despite the existence of standards like ATX2.03 or Intel ATX12V.

There are well-established standards for measuring and rating HDD capacity, an engine's horsepower, or the heat generated by a furnace... but not one for how much power a PSU can deliver. There are so many cases of people with "450W" PSUs having power stability issues running a system that can't possibly draw more than 150W. And "300W" units that keep running where the "450W" units are faltering.

It's not just about bad PSUs vs better ones. It's a dumb situation caused by uncontrolled marketing competition. Real regulation would bring PSUs out of snake oil territory and into a more sensible consumer-friendly terrain.

There are many ways PSU makers fudge to make their units seem more powerful.

1) Out and out lying. You add up the power on all the lines in many PSUs and they fall short of the rated power by 10, 20 30W or even more.

There are more sophisticated ways:

2) Limit the AC input voltage to a very narrow tolerance. The best PSUs are able to deliver their rated power given a decent range of AC input power, say 90~130V for a 120V unit. It's much more demanding to produce 300W w/90VAC input than with 120VAC, so what some PSU makers will detail in their tech specs (usually not in their consumer brochures) is to specify 115-120VAC for input power. A PSU specified this way will not deliver full power if the AC voltage sags, if there is a brown-out. Surely it causes instability more often than a PSU rated to deliver full power with 90-130VAC.

3) Specify a low operating temperature for rated output. This is quite common, but again not often seen in consumer brochures, but rather tech spec sheets provided usually only on demand by engineers or corp buyers. A typical PSU operating temp statement is somthing like this:



0°C~25°C for full rating of load, decrease to zero Watts O/P at 70°C

Examine what that says. Full power (let's say 400W) is available when the unit is at 0°C~25°C. Hmmm. Think about this.

Have you ever felt air blown out of a PSU in a PC running absolutely full tilt (which it would have to do to get anywhere near 400W output) that felt cool to the fingers? 25°C airflow would feel exactly that: Cool, given that normal body temperature is 37°C.

So this PSU cannot deliver full rated power when its temperature goes over 25°C. OK, what happens to the max power output capacity above that temp? It decreases gradually so that by the time the PSU temp reaches 70°C, the PSU cannot deliver any power at all. So if you assume that this power drop as temp rises is linear, then max power capacity will drop by ~9W for every degree over 25°C.

Now having examined as many PSUs as I have over the last 2~3 years, I have to say there's not a single PSU in ANY PC I have ever used or examined that would not measure at least 30~35°C almost anywhere inside the PSU under almost any kind of load. And if/when it is pushed, 45°C is nothing at all, especially for or near hot running components like voltage regulators.

So let's say 40?C is a fairly typical temp inside a PSU. This 400W rated unit would actually be able to deliver a max of just 220W at that temp. Hmmm. Interesting, isn't it? At 50°C, the available power would drop to just 130W. No wonder some PSUs have 3 fans each capable of 50 cfm!!

Here's a simple fact: Really high quality PSUs are actually rated for full power output at as high as 50°C. The trick is get a hold of the spec sheets that tell such information so you can compare apples to apples. Or ask.

HOW MUCH POWER IS ENOUGH?

300W models have replaced 230W and 250W models as baseline units since the introduction of the AMD Athlon. They feature a fan (or two) rated for 35~40 cubic feet per minute (CFM) airflow. Presumably, this level of airflow is required for adequate cooling at full power output to pass safety approvals under UL, CSA, CE and other regulations. In early 2005, retail PSU models rated for increasingly higher power, as much >600W, are being introduced by many brands.

Our own experience indicates that despite all the new power hungry components such as >75W video cards and >120W CPUs, it is still rare to find a desktop computer than draws much more than 200W DC under typical demanding applications. Around 300W DC looks to be about the highest power draw from a single CPU full-bore high end system at this time (Feb 2005). Although some headroom is always good to have, there seems little question that consumers are being persuaded to pay for power capacity that is never used. One of the nasty side effects is the fan noise of the high airflow required to keep the PSU adequately cooled when delivering maximum power. High speed fans generally make more noise than slower ones even when they are slowed by undervolting.

Why this state of affairs exists is a matter of marketing and technical obfuscation, probably more by accident than any massive conspiracy. With relatively low current requirements prior to the AMD Athlon processor, the aforementioned 230W and 250W were perfectly adequate for PC systems, even if the power supplies didn't deliver full rated performance. That changed with the Athlon and then the P4. PSU makers were quick to introduce higher rated models said to be required for the new power hungry processors. It was a good marketing opportunity. Rather than "Our 250W PSU is better than theirs," it is easier to sell the message "Our 300W PSU is better than their 250W PSU." Bigger is always better, isn't it? It also allowed higher prices to be charged.

A counterpoint is AMD's system builder's guide, which suggests higher numbers: up to ~180W DC for a typical system and ~250W DC for a high performance system, but these numbers are obtained by adding the maximum power rating for each component, then taking 20% off to account for real-world conditions. It is almost impossible for any application to demand 80% of maximum power draw from each component simultaneously. Intel's PSU recommendations are similar.

Suffice it to say that as manufacturers, both AMD and Intel are looking at worst-case secenarios. As custom builders, enthusiasts and system integrators can make choices based on real needs and applications.

Even so, Is Higher Power Better?

Without getting into technical details, the nature of a switching power supply is that it delivers as much power as is demanded by the components. This means that when installed in a PC whose components require 200W, a 400W PSU and a 250W PSU will each deliver 200W. Does this mean the 400W is coasting while the 250W is struggling? Not if they are both rated honestly and if they have the same efficiency. If one has lower efficiency than the other, then it will consume more AC to deliver the same power to the components, and in the process, generate more heat within itself. As long as there is adequate power, higher efficiency is the key to cooler, quieter PSU operation.

The main benefit of higher power PSUs is when the airflow in the PSU is deliberately set very low in order to minimize noise. This usually means the PSU components will run hotter. If all other things are equal, a higher rated PSU may be a better choice in such an application because its parts are generally rated for higher current and heat than a lower rated model.

What are the Key Aspects to Good PSU Performance?

There is a great deal of fuzzy and unclear thinking about what constitutes a good power supply. The obsfucation caused by competitive marketing is certainly one cause of this confusion. Another is the proliferation of computer hardware web sites that publish "reviews" of PSUs without much notion of what should be examined or how or why.

These parameters are the keys to good PSU performance:

  1. Stable power delivery under load
  2. High efficiency
  3. Good cooling
  4. Low noise operation
  5. Long term reliability

The truth is that a computer power supply is a complex electronic device with a complex role that is little appreciated by most hardware reviewers. Most system integrators don't really appreciate it either, either. This is due partly to the assemble-and-sell nature of the PC industry, where manufacturers build components in accordance to an accepted standard specification for "universal" compatibility with other components. Such piecemeal component manufacturing does not nurture or reward system thinking, which has been much more the norm for Apple.

DUAL 12V LINES: SPECS

Version 2.0 of Intel's ATX12V Power Supply Design Guide began recommending dual 12V lines for PSUs that can deliver more than 18A at 12V. In the latest guide, this recommendation has been reworded in section 3.5.7:

The 12 V rail on the 2x2 power connector should be a separate current limited output to meet the requirements of UL and EN 60950.

The requirements of UL and EM 60950 are related to safety. It stipulates that not more than 240VA is carried on any wires or exposed traces.

What is the safety reason for the 240VA maximum? It's the maximum recommended for an electronic device that a consumer will have reasonable likelihood of access. In plain terms, it might be to keep people from zapping themselves inside a PC, or more likely, accidentally creating a fire risk. This safety "rule" does not apply to any electronic or electrical devices where the chance of consumer exposure is low, such as a TV or CRT monitor, for example.

In PSUs that conform strictly to ATX12V v2.xx, it's important to know that even though there are two "independent" 12V lines, they still draw from the same main source. It's highly unlikely that there are two separate 120VAC:12VDC power conversion devices in a PSU; this would be too costly and inefficient. There is only one 12VDC source, and each of the two lines draw from the same 12VDC source, but through its own "controlled gateway".

PSU makers' specs are misleading in that they rate the current capacity of each 12V rail independently. What really matters is the total 12V current: Generally, up to 20A is available on any one 12V line assuming the total 12V current capacity of the PSU is not exceed.

What the above means is that you don't need to worry about imbalances in power draw on the 12V lines — as long as no single line is asked to deliver more than 20A. PSU makers seem to mark each line for max current on a purely arbitrary basis, probably more for marketing reasons than any other. A PSU rated for 32A max on the 12V lines can be labelled many different ways:

  • 12V1: 18A, 12V2: 14A
  • 12V1: 17A, 12V2: 15A
  • 12V1: 16A, 12V2: 16A
  • 12V1: 15A, 12V2: 17A
  • 12V1: 14A, 12V2: 18A

It could be marked 20A + 12A, but being a cautious bunch, the engineers will probably not specify more than 18A on any one line. This gives 2A headroom to allow some room for error for the current limiting circuit.

DUAL 12V LINES: REALITIES

Note that 12V2 is supposed to supply only the AUX12V (2x12V) 4-pin plug, which feeds only the CPU. With PSUs that adhere strictly to the ATX 12V v2.xx Guide, 12V1 then must supply 12V to all the other components that require it. This might lead to a problem with very high power gaming systems that utilize two high power video cards in SLI or Crossfire mode. Current high end VGA cards by themselves can draw >90VA each. Much of this comes from the 12V line via the 6-pin PCIe connector for the VGA card. If you add several hard drives and optical drives, the 240VA limit may be too low.

The current ATX12V v2.2 spec was created before dual VGA card gaming configurations for Intel boards were announced. SLI, being an nVidia feature on nForce 4 chips for AMD CPU motherboards that came many months earlier, may have been ignored by Intel's PSU design guide team.

Not all PSUs with 6-pin PCIe connectors follow ATX12V v2.xx to the letter. In fact, they can't, as the guide does not cover the 6-pin 12V PCIe outputs. This connector and its current delivery capacity was specified by nVidia, the originator of the SLI concept. nVidia maintains a list of power supplies that they have certified as being suitable for SLI systems. The question is, Where should this 12V come from? More to the point, which line DOES it come from?

I interviewed a number of engineers from several power supply manufacturers to pose this very question. The answers were surprising. All of the engineers I spoke with wished to remain anonymous. This is a summary of what they told me:

  • Some PSU makers are using 12V2 to supply more than just the 2x12V or 4x12V connectors. It is often used to power the 6-pin 12V PCIe outputs as well.
  • Many PSUs marked as having dual (or more) 12V lines actually have only a single 12V line — they do not feature two 240VA current limiters specified by ATX12V v2.xx; they have only one Over Current Protection (OCP - current limiter) for the single 12V line.
  • The 240VA current limit is considered a high cost, useless annoyance by most PSU makers. If multiple 12V lines are used, because the vast majority of components now use mostly 12V, the 18~20A limit for any line means that the precise power distribution to the various 12V output connectors can become critically important in some cases.
  • The engineers point to the many high power pre-V2.xx ATX12V PSUs that had as much as 30A on a single 12V line. As a product class, those have not proven to be any more dangerous in any way than other ATX12V PSUs. Even if exceeding 240VA in a single wire run was dangerous, this is extremely unlikely to occur in a PC because 12V is distributed to many different components on many different wire runs.

What's really interesting is that Intel has tacitly waived the 240VA limit requirement in its PSU validation program for the better part of a year. Intel maintains a web page listing all the ATX12V they have tested that "meet MINIMUM electrical, mechanical fit and functional compatibility" with Intel desktop boards and processors. For the 32 ATX12V v2.2 PSUs tested in 2005 that are on this list, 17 models are identified as having at least one output line that exceeds 240VA. And yet, these 17 models are on Intel's approved list.

According to the engineers I spoke with, the majority of these 17 models have just one 12V line. They also point out that there are another 20 or so ATX12V v2.0 PSU models on the Intel list, and none of them were tested for the 240VA current limit conformance. My sources say that if these models had been tested, more than half would not conform to the 240VA current limit because they have only one 12V line.

In the last couple of months, my PS engineering sources report, Intel has verbally informed them that the 240VA limit has been removed. A single 12V line is now "officially" approved, never mind what ATX12V v2.2 specifies.

What does all this mean? The safety benefit of dual 12V lines is questioned
by the engineers I spoke with. There are many downsides to multiple 12V lines,
including higher cost and the extra headache of ensuring adequate 12V current
for all the components in complex, high power systems. For the consumer who
is trying to make a choice among the myriad of PSUs available on the retail
market today, the most practical approach regarding dual 12V lines and power
capacity is to consider only the combined 12V current capacity.

This is not to say that there are no advantages of multiple, independent 12V
lines. The fact is that the 12V line is where most of the power is delivered
in a modern PC. In a high power system where many components are pulling on
the 12V line simultaneously (for example, a high power dual video card gaming
rig), independent 12V lines could help improve stability under certain conditions.
The current limits on each 12V line then become important to consider when wiring
up the system. If no specific guidlines are given by the manufacturer regarding
which components should be connected to which connectors and/or cables (especially
with detachable output cable models), then it is probably safe to assume that
the PSU does not have really independent, separate 12V lines.

REAL SYSTEM POWER REQUIREMENTS

While SPCR's reviews test the PSU to full output (even >600W!) in order to verify the manufacturer's claims, real desktop PCs simply do not require anywhere near this level of power. The most pertinent range of DC output power level is between about 50W and 250W, because it is the power range where most desktop systems will be working. To illustrate this point, consider the power consumption of the systems described below. They were tested with the same equipment used to test power supplies.


REAL SYSTEM POWER REQUIREMENTS


System

Total Power Draw


State



AC
Input



DC
Output

A: Low P4

Intel Pentium 4-2.8C

AOpen MX4SGI-4DL2 motherboard

2 x 512 mb OCZ PC3700 DDRAM

Seagate 7200.7 120G HDD

Seagate Barracuda IV 40G HDD

Matrox P650 VGA (dual head mode)

Seasonic Super Tornado 350W PSU

Asus QuieTrack CDRW

6-in-1 card reader / floppy drive

3 low speed fans


idle

71W


54W



Folding @ Home

115W

92W


PCMark04

126W

102W

B: High A64

Athlon A64-3800+ (130nm core)

Soltek SL-K8TPro-939 motherboard

4 x 512 mb OCZ PC4000 DDRAM

ATI 9800-256 Pro VGA

Hitachi 7K400 HDD (400gb)

Samsung P160 HDD

Silverstone ST30NF PSU (fanless)

M-Audio Firewire 410 external firewire-driven sound card

low speed 80mm fan


idle

99W


76W



Folding @ Home

126W

102W


PCMark04

184W

147W

C: Mid P4

Intel Pentium 4-3.2 (Northwood)

Intel D875PBZLK motherboard

2 x 256MB HyperX DDR400 PC3200

ATI Radeon 9800XT 256MB DDR

16x Sony DVD-RW

Zalman 400W PSU

Samsung HDDs

Creative SB Audigy-2 ZS Platinum

2 x 120mm fans and 1 80mm fan


idle

127W


94W



Folding @ Home

194W

146W


PCMark04

236W

180W

NEW! E: A64 Dual Core

AMD A64-4800+ X2 EE (2.4 GHz, 65W TDP) socket 939

DFI RS482 Infinity motherboard

2 x 1024MB Corsair DDR2-6400 RAM

ATI Radeon X1800GTO PCIe graphics

Seagate 7200.7 80GB SATA hard drive

Maxtor Diamondmax 10
300GB SATA HDD

Seasonic SS-400HT ATX PSU

Zalman 9500 HSF
w/ Nexus 92 fan

1 x 120mm fan


idle

88W


69W



Folding @ Home

130W

104W


PCMark05

175W

140W

D: High P4

Intel Pentium 670 (Prescott, 3.8GHz)

Intel D915PBL motherboard

2 x 512MB Corsair DDR2 RAM

AOpen Aeolus 6800GT PCIe VGA

2 x 250 GB Western Digital Caviar SE HDD

Seasonic S12-430W PSU

Creative SB Audigy-2 ZS Platinum

3 x 120mm fans


idle

141W


109W



Folding @ Home

214W

168W


PCMark04

264W

214W

NEW! F: ATI X1950XTX
/ Pentium D950

Intel Pentium D950, overclocked / overvolted 10% (Presler, 3.74GHz)

Asus P5LD2-VM motherboard

4 x 1024MB Corsair DDR2-6400 RAM

ATI Radeon X1950XTX-512 PCIe graphics

Hitachi Deskstar 7K80 80GB hard drive

WD Raptor WD1500ADFD 150GB
10krpm HDD

Seasonic SS-350ET ATX PSU

1 x 120mm fan


idle

119W


95W



Folding @ Home

151W

125W


PCMark05

298W

256W

NOTES:
  • All the systems have two hard drives.
  • The AC input power was measured directly; it is what the PSU draws from the AC line. The DC output power was calculated based on the efficiency of the PSU used in each system. It is the DC power delivered to the components in the system by the PSU. We have measured the efficiency of all the actual PSUs used in the above systems. The results have been posted in previous PSU reviews. The exception is system D, where a clamp meter was used to measure the DC power delivered.
  • It IS possible to obtain about 10% higher peak power draw with the same systems by engaging all the drives in the system to write and read to each other simultaneously. The actual work a PC can do under such multitasking conditions is minimal, and this kind of usage can be considered a gross abuse of the system. However, ensuring some overload margin is not unwise.
  • Folding @ Home is a pretty good realistic maximum power draw number for a system used with many software applications. The power draw during folding also closely approximates turn-on maximum power draw in these systems.
  • The highest power consumption was achieved while running PCMark, a system benchmark which brings the VGA card into play. The recorded wattage is the highest peak seen during this benchmark. Sustained maximum was about 5~10% lower.
  • These numbers are accurate and based on repeated empirical testing, but for argument's sake, you could say they're as much as 10% too low. The max power draw of any system we've discussed here would still be less than 300W DC.

POSTSCRIPT — April 4, 2007

  • The ATI X1950XTX
    / overclocked Pentium D950 system information was added to round out the data. There were snickers that the nVidia 6800GT graphics card of system D was completely outmoded. The 1950XTX still remains about the most power hungry graphics card on the market today. The Pentium D950 actually has a higher TDP (130W) than most current CPUs; we made it run hotter by overclocking and overvolting 10%. This system has a higher DC power demand than just about any single vidcard, single CPU system you can build today. Yet, the total DC power demand is just 256W. (Note: It runs perfectly stable with a new very high efficiency 350W power supply from Seasonic.)
  • A high performance A64X2-4800+ dual-core system with a mid-range graphics card and dual drives was also added. Again, the power demand is very modest, falling well under 200W at the AC outlet.
  • In the context of SPCR, dual-graphic card SLI / Crossfire systems are excessive. For the SPCR audience, the heat, power consumption and noise issues around dual graphics card systems makes them difficult to embrace.

CORRECT PSU SIZING FOR BEST ENERGY EFFICIENCY

This was a topic of discussion in one of the sessions at the Spring 2005 IDF in San Francisco: the concept of choosing a PSU whose efficiency curve is well-matched to the system power load. Such matching can yield incremental improvements in average power consumption and ensure minimum waste heat generation. Correct PSU sizing is very carefully practiced by tier one computer makers concerned with maximum cost effectiveness.

SPCR's own PSU testing has shown that power efficiency in PSUs varies with load, and the load at which the best efficiency is reached varies from model to model. In a given line of PSU models, the maximum and average efficiency tends to be very similar; where the peak occurs depends on power rating. Most PSUs reach peak efficiency between 50~75% loading, tail off a bit at maximum power and drop at least 10% at minimum load.



PSU efficiency data from SPCR review database.

The vertical scale has been truncated for clarity; please see text below for full discussion.

A system that draws ~250W maximum and idles at 150W and peaks at >300W. It would be a substantially worse match for the system of the previous example, as the PSU would be operating at a mediocre

From a PSU heat waste point of view, the differences are significant:

At 200W load,

  • the 300W model would generate 44W of heat (18% of 244W AC input);
  • the 600W model would generate 50W of heat (20% of 250W AC input).

At 90W load,

  • the 300W model would generate 23W of heat (20.5% of 113W AC input).
  • the 600W model would generate 32W of heat (26% of 122W AC input).

Using the 600W PSU with this system is an example of incorrect, costly PSU sizing. It is practised most frequently by gaming enthsiasts who are encouraged to believe that greater power capacity is always better. Whether 480W, 550W or >600W PSUs are suitable for system that cannot possibly draw even 250W is a type of question asked almost daily in the SPCR Forums.

The counterpoint to "correct PSU sizing" comes from very high
efficiency power supplies with very flat power efficiency curves.
The
models in the Seasonic S12 series, for example, differ so little in efficiency
at the same 65~300W power levels that there's no real power consumption cost
in choosing a high power model even if it is going to be used at lower power
levels. With a system that only needs a maximum of 200W, the power consumption
using a S12-500 or a S12-330 is essentially the same. Aside from the initial
cost difference, there's no operational cost due to lower efficiency at lower
power levels. With such power supplies, it makes sense to buy higher power capacity
than currently needed in anticipation of future component upgrades that will
demand more power.

EFFICIENT POWER SUPPLIES THAT DON'T START

Higher efficiency PSUs generally tend to need higher minimum power on the 12V line in order to simply run. Typically, we're talking about 1A or greater. Older, less efficient PSUs have much lower minumum current needs, under 0.5A and often ZERO.

In some recent motherboards, there are various time delays implemented in order to ensure that the PSU (and motherboard) is not subject to a huge current surge when everything turns on all at once. Many Asus boards have been identified as doing this, al though you won't get Asus to talk about it — I tried — they will say it's proprietary information they don't want to share with competitors. They are not the only board makers doing this.

The practice began during the peak of the Prescott era when startup surge became quite serious, but before the 80% efficient power supplies became common. Board makers extended the practice to AMD boards as well.

So this means, for example, that there could be anywhere between tens and hundreds of millseconds between different portions of the board and components being powered up. Just how much delay there is and how much power the CPU/VGA draw affects whether one of these high efficiency PSUs will actually start. Sometimes, adding HDDs will help, sometimes not — they may not pull current soon enough after the power button is pressed to change the current demand the PSU "senses". If the current sensor detects too low a load, the power supply usually does not start.

I don't have concrete information about the time delays involved. However, the Asus boards that would not start with some high efficiency Seasonic PSUs also would not start with some high efficiency PSUs from Antec, Fortron-Source, and Enhance.

A sure-fire way to tell whether too-low 12V start current is the problem is to hook up a known working, older, generic 300W PSU to the afflicted system. If max power was the problem, it would have a hard time starting, or not start at all. But invariably, with these too-low 12V start current situations, such PSUs (even several years old ones that long precede 24-pin ATX outputs, etc.) will start the system fine.

The reality is that most of the better brands like Seasonic and the others mentioned above are going for high efficiency because it is one of the big differentiators between PSUs today, and also very high power output. There are few PSUs that put less than 80% of the total power rating available on the 12V rail. For a 400W PSU, this typically means 320W is available on the 12V lines. You simply don't get a surge that big at startup with most computers, enthusiast or not, so even older PSUs should start fine on most systems

The upside of all this is that most PSU makers are aware of the issues here, and they are implementing solutions. The simplest one is to add just enough internal resistance on the 12V rail to ensure that there is enough current draw to start the PSU even with no 12V draw from the outside. This naturally drops the hard-earned efficiency down a notch, but it is in fact, what some PSU makers have done.

I know that Seasonic has quietly implemented an active circuit that automatically inserts enough of a load so that the 12V line always sees the minimum load, at least, but then this extra resistance is removed when the load gets higher, so that turn-on is never a problem, and high efficiency is maintained at normal and high power operation.

Which Seasonic models? I believe all the current sleeved output cable S12s, the S12-80+ models and the soon to come M12s.

POWER FACTOR CORRECTION

Increasingly, switched mode power supplies (SMPS) are designed with an active power factor correction (PFC) input stage. This is mainly to meet new regulations aimed at restricting the harmonic content of the load current drawn from power lines. Both users and power companies benefit from PFC, as does the environment.

Power Factor Correction (PFC) can be defined as the reduction of the harmonic content, and/or the aligning of the phase angle of incoming current so that it is in phase with the line voltage. Mathematically, Power Factor (PF) is equal to Real Power (Watts) divided by Apparent Power (Volt*Ampere). The basic concept is to make the input look like a pure resistor. Resistors have a power factor of 1 (unity). This allows the power distribution system to operate at maximum efficiency, which reduces energy consumption.

Non-PFC power supplies use a capacitive filter at the AC input. This results in rectification of the AC line, causes high peak currents at the crests of the AC voltage. These peak currents lead to excessive voltage drops in the wiring and imbalance problems in the three-phase power delivery system. The full energy potential of the AC line is not utilized. Nonlinear peak currents also distort output voltage and create harmonics. There is an international standard for controlling harmonics (IEC100-3-2) and PFC is mandatory for home appliances consuming 70W or more power in EU nations as of January, 2001.

PFC circuits are classified into two types: active and passive.

Passive PFC uses passive elements such as a ferrite core inductor on the input source to create a countering reactance. While easily applied to the existing power circuitry without much modification, the power factor is low (60 - 80%), the AC input must be chosen (115VAC / 230VAC), and the harmonics produced from the difference between the capacitance and the inductance are hard to control. Significant electromagnetic noise can result.

Active PFC uses switching regulator technology with active elements such as IC, FET and diodes, to create a PFC circuit This circuit has a theoretical power factor of over 95%, reduces total harmonics noticeably, and automatically adjusts for AC input voltage. However, it requires a complex EMI filter and an input source circuit, and is more costly to build.

The benefits of high PF for the user comes from the reduced AC current drawn by high PF PSUs, not in any savings from electricity bills, except in the case of commercial utility users who do pay for V(oltage) x A(mperes). There are two broad consequences:

Less stress on the AC electrical wiring: The lower current drawn by a high PF power supply means that there is less stress on the electrical wiring of the building. This can be a big plus in the case of older building with lower capacity AC wiring. It is certainly easy to see the benefits in a enterprise setting where dozens or hundreds of PCs are drawing power. If the total current load from the IT department could be reduced by 30% or more, this would be very signficant in direct electricity savings, reduced airconditioning cost, and possible avoidance of building AC re-wiring.

Lower UPS costs: Lower current draw also means that smaller capacity Uninterruptible Power Supply (UPS) units can be used. As UPS units are priced in direct proportion to their current capacity (VA), a PF of 0.98 versus one of 0.6 can traslate into a 40% reduction in purchase cost. Again, in an enterprise setting with hundreds or thousands of PCs, the savings can be very significant.

PFC Myths

There are myths about power factor correction that continue to be propagated by well-meaning people. Let's tackle the two most common ones:

Does higher PF reduce my electricity bill? No, if you are a home user. If you are an enterprise running hundreds of PCs and pay not only for power but also VA, then yes. For more details, see PFC discussion above.

Does PFC make a power supply more efficient? Not in the normal way that power supply efficiency is defined, which is the power loss (to heat) as a percentage of total AC input in AC-to-DC conversion. However, in the sense that Apparent AC power (VA) is lowered, PFC does reduce energy consumption.

Power factor correction is applied by an input circuit which uses a small amount of input power. With two PSUs that are identical, equipping one with PFC will cause a typical efficiency drop of 2~4%. Many PSUs that have Active PFC also have high efficiency, as APFC is usually found on higher quality PSUs, but the two are not intrinsically related.

WHO REALLY MAKES THAT POWER SUPPLY?

We've often noted in the past couple of years that the retail power supply scene seems to be attracting new brands constantly. We've also noted that most of these new brands are selling products designed and built by other companies, Original Equipment Manufacturers (OEM) and Original Design Manufacturers (ODM) who are happy to customize an existing model for a large volume customer. Some SPCR readers have expressed interest in finding out who is making what. We've known for a long time how to find out, but never thought it important enough to pass this information. While knowing who manufactured a product can be helpful in assessing quality, we feel our reviews provide plenty of information to make that assessment.

Recently, thanks to Hardware Secrets, another PC hardware website, the details of how to identify a PSU's manufacturer have been spelled out for all to see. Here's a quick summary:

Underwriters Laboratories maintains a certifications directory that is accessible from their website, http://database.ul.com/cgi-bin/XYV/template/LISEXT/1FRAME/index.htm. Almost every computer power supply is certified for safety by UL, and the label on most certified power supplies is marked with the UL logo that contains the UL File Number for the manufacturer.





The UL file number on this SilverStone power supply is E166947. It's obviously made by Enhance Electronics... but most labels will not identify the manufacturer so plainly.

Simply enter the number in the "UL File Number" data entry box on the UL Certifications Directory web page linked above, then hit the search button on that page. The results are not infallible, however. Sometimes, the brand owner will have its own UL File Number, in which case, its name will come up in the search. Often, the UL File Number will be that of the manufacturer, in which case, that manufacturer's name will come up in the search.

DISCUSSIONS ABOUT PSUs

There are many informative discussions in the SPCR PSU forum. Several of these threads have been turned into stickies that always appear at the top of the forum so they can be easily found. They are linked directly here:

* * *

Articles of Related Interest

Recommended PSUs

Power Supply Fundamentals


Power Distribution within Six PCs

Desktop CPU Power Survey,
April 2006

*

Discuss this article in the forums.

Google

www SPCR