X-bit Labs Presents: Power Supply Unit Testing Methodology In-Depth

This article provides a detailed description of our methodology for testing power supplies in three major parts. The first one lists PSU parameters we check out and specifies the test conditions. In the second section you’ll find terms often voiced by PSU manufacturers for marketing purposes and their definitions. The third part will be most interesting for people who’d like to know how our PSU testbed is designed and operates.

by Oleg Artamonov
01/03/2008 | 10:43 AM

This article provides a detailed description of our methodology for testing power supplies. The last article discussing our methodology indepth dates back to January 16, 2005 (X-bit Presents: Power Supply Unit Testing Methodology), and since then additional info on methodology changes has been rather inconveniently scattered among several reviews. So it is high time we posted a new article answering all questions about our testing procedures and techniques.

 

As our testing methods have evolved and improved over time, some of them may not be reflected in our older PSU reviews just because the method was developed after the publication of the review. The list of changes to this article is provided at its end.

There are three main sections in this article. The first one lists PSU parameters we check out and specifies the test conditions. In the second section you’ll find terms often voiced by PSU manufacturers for marketing purposes and their definitions. The third part will be most interesting for people who’d like to know how our PSU testbed is designed and operates.

When developing our methods we based ourselves on the ATX Power Supply Design Guide standard the latest version of which is available at FormFactors.org. By today, it has become part of the more comprehensive document called Power Supply Design Guide for Desktop Platform Form Factors in which PSUs of not only ATX but also other form-factors (CFX, TFX, SFX, etc) are described. Although PSDG is not formally obligatory for PSU manufacturers, we think that if not stated otherwise, a PC power supply (selling in retail and meant for general use rather than for specific PC models of a specific manufacturer) must comply with the PSDG requirements.

You can refer to the appropriate section on the site for the reviews of particular PSU models.

Visual Examination

Of course, each test session begins from the tester’s taking a look at the PSU. Besides an aesthetic delight (or disappointment), such visual scrutiny provides some clues as to the quality of the product.

First, it is the quality of the housing as is indicated by the thickness of the metal, rigidity, special features of the assembly (for example, the PSU case may be made from thin steel but fastened with seven or eight screws instead of four as usual), the quality of paint, etc.

Second, it is the quality of the internal component mounting. Every PSU coming to our labs gets opened up and scrutinized and photographed. We don’t focus on minute details and don’t enumerate all the components we find in the PSU along with their ratings. This would make the review more academic, but wouldn’t make much sense for the end-user. However, if the PSU has a non-standard circuit design overall, we try to describe it in general and explain why the developer may have chosen it. Of course, we also draw your attention to any serious defect in the quality of manufacture such as sloppy soldering.

Third, it is the specified parameters of the PSU. The specs are often indicative of quality when it comes to inexpensive PSUs, for example when the total output power marked on the label proves to be much higher than the sum of products of the currents and voltages marked on the same label.

We also provide a list of cables and connectors the PSU offers indicating their length. The latter is written as a sum of numbers, the first of which is the distance from the PSU to the first connector, the second number is the distance between the first and second connectors, etc. The cable pictured above would be described as follows: a detachable cable with three SATA power connectors, 60+15+15cm.

Operation under Full Load

The full output power of a PSU is the most intuitively comprehensible and, consequently, the most popular parameter among end-users. The PSU label shows the so-called continuous output power. Sometimes, the peak output power is also indicated – the PSU can yield it for no more than one minute. Some unconscientious manufacturers specify either peak or continuous output power at room temperature only. When such a PSU is installed into the real PC, in which the air temperature is certainly higher than the room temperature, the permissible max output power lowers. According to ATX 12V Power Supply Design Guide, the fundamental document on all PSU-related issues, the PSU must operate at the output power indicated on its label up to an air temperature of 50°C. Some manufacturers even mention this temperature explicitly to avoid ambiguities.

Our conditions are relaxed, though. We test the PSU at full load under typical room temperature, i.e. 22-25°C. The PSU works at full load for half an hour at least. If nothing bad happens to it during that time, we consider it as having passed the test.

Our current testbed can fully load PSUs with a specified output power up to 1350W.

Cross-Load Characteristics

Although the PC power supply acts as the source of several voltages, the main of which are +12V, +5V, and +3.3V, there is a common regulator for the former two voltages in many PSU models. This regulator is oriented at the arithmetic mean between the two controlled voltages, this design being known as joint voltage regulation.

The pros and cons of it are obvious. On one hand, the cost of the PSU is reduced. On the other hand, the voltages depend on each other. For example, if the load on the +12V rail is increased, the appropriate voltage goes down and the regulator tries to pull it up again to the previous level. But controlling both voltages simultaneously, the regulator increases the +5V voltage as well. The regulator considers the situation normal when the average deflection of both voltages from the nominal value is zero, but it means that the +12V voltage is slightly below the nominal, and the +5V voltage is slightly above the nominal value. If the former voltage is increased higher, the latter will increase as well. If the second voltage is reduced, the first one will lower, too.

PSU developers are trying to find ways to solve the problem. Their attempts can be evaluated by means of cross-load diagrams.


Example of a cross-load diagram

The X-axis of the diagram shows the load on the PSU’s +12V rail (the combined load on all of its +12V lines if there are several of them in the given PSU). The Y-axis shows the combined load on the +5V and +3.3V rails. Thus, each point of the diagram corresponds to a certain load distribution among these power rails. For better readability we paint the diagram different colors denoting the deflection of the voltages from the nominal values, from green (a deflection below 1%) to red (a deflection of 4-5%). A deflection of higher than 5% is considered as unacceptable.

For example, the cross-load diagram above shows that the tested PSU keeps the +12V voltage rather stable, most of the diagram being green. It is only when the load distribution is misbalanced towards the +5V and +3.3V rails that the +12V voltage becomes red.

Moreover, the diagram is limited from the left, bottom and right with the minimum and maximum allowable load on the PSU, but the uneven top border is due to the voltages exceeding the 5% deflection. According to the industry standard, the PSU should not be used at such loads.


Typical workload area on the cross-load diagram

It is also important in what exactly area the voltages deflect the most. The hatched area in the diagram above denotes the power consumption typical of modern PCs: today, all high-power components (graphics cards, CPUs) are fed from the +12V line, which can be under a very high load. The +5V and +3.3V rails, on the contrary, are only responsible for hard disks and mainboard components now. Their load cannot be higher than a few dozen watts even in a top-end PC system.

Comparing these two diagrams built for two PSUs, you can see that the first PSU goes red in the area that is unimportant for modern PCs whereas the red zone of the second PSU is located differently. So, even though the two PSUs have similar results considering the whole range of loads, the first one is going to be preferable for practical applications.

We are tracking the three main power rails of the PSU during this test: +12V, +5V and +3.3V. The cross-load characteristics are presented in our articles as an animated three-frame image, each frame corresponding to one of the mentioned rails.

There have recently been more and more PSUs with dedicated regulation of the output voltages in which the classic design is complemented with additional saturated-core regulators. Such PSUs show much weaker interdependence between the output voltages. Their cross-load diagrams are mostly green.

Fan Speed and Temperature Growth

The efficiency of the cooling system of a PSU can be characterized by two parameters: noise and temperature growth. Obviously, it is difficult for a PSU to be good from both aspects. You can achieve a small temperature growth by installing a faster fan, but you lose in terms of noisiness then, and vice versa.

To evaluate the efficiency of the cooling system of the tested PSU, we are changing its load from 50W to maximum, giving the PSU 20-30 minutes to warm up at each step – the temperature stabilizes during that time period. After that we use an optical tachometer Velleman DTO2234 to measure the speed of the PSU fan and a dual-channel digital thermometer Fluke 54 II to measure the difference of air temperatures at the PSU’s input and output. Ideally, both numbers should be small. If the temperature and the fan speed are both high, the cooling system design is poor.

Every modern PSU can regulate the speed of its fan but the initial speed (i.e. the fan speed at minimum load, which determines the noisiness of the PSU when the PC is idle and the fans of the graphics card and CPU are working at minimum speeds), and the dependence of speed on load can vary greatly. Particularly, in entry-level PSUs there is often just one thermistor without any additional circuits that regulates the fan speed – and the speed can only be changed by 10-15%, which is almost the same as no regulation at all.

Many PSU manufacturers specify noise in decibel or fan speed in rotations per minute. Both are often accompanied with a marketing trick as these parameters are measured at a temperature of 18°C. The resulting number can look pretty (like a noise level of 16dBA) but carries no practical meaning since the air temperature is going to be 10-15°C higher in a real PC. Another trick we have met with was the specifying of the parameters of only the slower of the two fans for PSUs with a two-fan cooling system.

Output Voltage Ripple

The point of a switching power supply – and all PC power supplies are of the switching variety – is that the step-down power transformer operates at a frequency many times that of the alternating current in the mains. This helps reduce the dimensions of the transformer greatly.

The alternating mains voltage (at a frequency of 50Hz or 60Hz depending on the geographical region) at the PSU input is rectified and smoothed and applied to the switching transistor that converts the direct voltage back into alternating but at a much higher frequency (from 60 to 120kHz depending on the PSU model). This voltage comes to the high-frequency transformer that lowers it to the value we need (12V, 5V, etc) and then gets rectified and smoothed again. Ideally, the PSU’s output voltage should be strictly constant but it is impossible to get it from an alternating high-frequency current. The ATX12V Power Supply Design Guide standard demands that the output voltage ripple at full load be not higher than 50 millivolts for the +5V and +3.3V rails and not higher than 120 millivolts for the +12V rail.

During our tests we record oscillograms of the PSU’s output voltages at full load by means of a dual-channel oscilloscope Velleman PCSU1000 and show them all in a single diagram:

The top, middle and bottom graphs correspond to the +5V, +12V and +3.3V rails, respectively. To the right of the graphs the permissible maximums of the pulsations are given. As you can see, the +12V rail of this PSU meets the requirements easily. The +5V voltage is barely within the norm, and the +3.3V voltage violates the permissible limits. The high spikes of the oscillogram of the latter voltage indicate that the PSU doesn’t cope with filtering highest-frequency noise, which is usually the consequence of using poor electrolytic capacitors whose efficiency degenerates as the operating frequency is increased.

If the output voltage ripple is above the norm, it may affect the stability of operation of your PC system and interfere with audio cards and other such equipment.

Efficiency Factor

We have been discussing the output parameters of the PSU so far but the efficiency counts its input parameters in. It means what percent of the power taken from the mains is converted into the power the PSU yields into the load. The difference is wasted for heating the PSU up.

The current version of the ATX12V 2.2 standard restricts the PSU efficiency from below: minimum of 72% at typical load, 70% at full load and 65% at low load. Besides that, there are optional numbers (an efficiency of 80% at typical load) and the voluntary certification program “80 Plus” which requires that the PSU has an efficiency of 80% and higher at loads from 20% to maximum. The new certification program Energy Star 4.0 has the same requirements as in the 80 Plus standard.

The efficiency of a PSU depends on the input voltage. The higher that voltage is, the better the efficiency. The difference in efficiency between the 110V and 220V power grids is about 2%. Moreover, different samples of the same PSU model may vary in efficiency by 1-2% due to the variations in the parameters of the components employed.

In our tests we are changing the load on the PSU from 50W to the specified maximum in small steps and measure the amount of power the PSU consumes from the mains. The ratio of the output power to the input power is the efficiency of the PSU. This test produces a graph showing the dependence of efficiency on load.

The efficiency of typical switching PSUs grows up quickly along with the load, reaches the maximum, and then slowly lowers. This non-linearity has one interesting consequence: from the efficiency standpoint, it is better to buy a PSU whose specified output power is adequate to the load. If you take a PSU with a large reserve of power, it won’t be very efficient at low loads (like the 730W PSU whose efficiency is shown above is at a load of 200W).

Power Factor

The AC electric mains can be considered as having two types of power: active and reactive. Reactive power is generated in two cases: when the load current and the mains voltage are out of phase (that is, the load is inductive or capacitive) or when the load is non-linear. The PC power supply is a pronounced example of the second case. It will normally consume the mains current in short high impulses that coincide with the maximums of the mains voltage.

The problem is that while active power is fully transformed into useful work in the PSU (meaning both the power the PSU yields into the load and its own heating up), reactive power is not consumed at all. It is driven back into the mains. It is kind of wandering to and fro between the generator and the load, but it heats up the connecting wires as well as active power does. That’s why reactive power must be got rid of.

The circuit called active PFC is the most efficient way to suppress reactive power. It is in fact an impulse transformer that is designed in such a way that its instantaneous consumed power is directly proportional to the instantaneous voltage in the mains. In other words, it is linear and consumes active power only. The voltage from the output of the active PFC device goes right to the switching transformer of the power supply which used to be a reactive load due to its non-linearity. But now that it receives direct voltage, the non-linearity of the second transformer doesn’t matter anymore because it is detached from the electric mains and cannot affect it.

The power factor is the measure of reactive power. It is the ratio of reactive power to the total of active and reactive power. It is about 0.65 for an ordinary PSU, but PSUs with active PFC have a power factor of 0.97-0.99. So, the active PFC device reduces reactive power almost to zero.

Users and even hardware reviewers sometimes make no difference between the power factor and the efficiency factor. Although both these terms describe the effectiveness of a power supply, it is a gross mistake to confuse them. The power factor describes how efficiently the PSU uses the AC electric mains, i.e. what percent of power the PSU consumes from it is actually put to use. The efficiency factor describes how efficiently this consumed power is transformed into useful work. There is no connection between these two things because, as I said above, reactive power, which determines the value of the power factor, is not transformed in the PSU into anything. You cannot apply the term “conversion efficiency” to it, so it has no effect on the efficiency factor.

Generally speaking, it is the power supply companies rather than the users that profit from active PFC because it reduces the computer’s load on the electric mains by a third or more. This amounts to big numbers today when there is a PC standing on every office desk. From an ordinary user’s point of view, active PFC makes no difference even when it comes to electricity bills. Home electricity supply meters measure only active power as yet. The manufacturers’ claims that active PFC can in any way help your computer are nothing but marketing noise.

A side effect of active PFC is that it can be easily designed to support a full range of input voltages, from 90 to 260V, thus making it a universal PSU that can work in any power grid without a manual selection of the input voltage. Moreover, PSUs with manual switches can only work in two input voltage ranges, 90-130V and 180-260V, and you cannot start them up at an input voltage of 130-180V. A PSU with active PFC covers all those input voltage ranges without any gaps. So, if you have to work in an environment with unstable energy supply, when the AC voltage may often bottom out to below 180V, a PSU with active PFC will allow you to do without a UPS or will make the UPS’ battery life much longer.

Well, the availability of active PFC does not guarantee that the PSU supports the whole range of input voltages. It can be designed to support a range of 180-260V only. This is sometimes implemented in PSUs to be sold in Europe because the use of such narrow-range active PFC helps reduce the manufacturing cost of the PSU somewhat.

Besides active, there are passive PFC devices. Passive PFC is the simplest way to correct the power factor. It is just a large choke connected in series with the power supply. Its inductance is smoothing out the pulsation of the current consumed by the PSU and is thus reducing the level of non-linearity. There is a very small effect from passive PFC – the power factor grows only from 0.65 to 0.7-0.75. But while implementing active PFC requires a deep redesign of the PSU’s high-voltage circuitry, passive PFC can be easily added into any existing power supply.

We measure the power factor of a PSU by the same method as the efficiency – increasing the load from 50W to maximum in steps. The results are presented together with the efficiency in a single diagram.

Operation with UPS’s

The above-described active PFC has one drawback. Some of its implementations cannot work normally with uninterruptible power supplies. When the UPS switches to its batteries, such an A-PFC device increases its consumption greatly, making the UPS shut down to prevent overload.

To check out the implementation of active PFC in each particular PSU we connect it to an APC SmartUPS SC 620VA and test them both in two modes: powered from the mains and powered from the batteries. In both cases the load on the PSU is being steadily increased until the UPS reports overload.

If the PSU is compatible with the UPS, the max load on the PSU is typically 340-380W and 320-340W when powered from the mains and batteries, respectively. When the load is higher at the moment of switching to the batteries, the UPS reports overload but doesn’t shut down.

If the PSU has the mentioned problem, the maximum load the UPS can work from the batteries at is far below 300W, and if it is exceeded, the UPS shuts down right after switching to the batteries or five to ten seconds afterwards. You should avoid such PSUs if you are planning to use a UPS.

Fortunately, there are fewer power supplies incompatible with UPS’s these days. For example, this incompatibility had been a problem with FSP Group’s PLN/PFN series but was eliminated in the subsequent GLN/HLN series.

If you have a PSU that is unable to work with UPS’s normally, you have two solutions (besides improving the PSU itself which would call for a good knowledge of electronics): change the PSU or change the UPS. The former solution is usually cheaper since the UPS would have to have a large reserve of wattage or even be the online variety, which is expensive and not really appropriate for home applications.

Marketing Noise

Besides the technical characteristics that can and should be checked out in tests, the manufacturers are prone to cover their products with a veil of pretty words and labels touting all kinds of technologies. The meaning of those technologies is sometimes distorted or trivial. Some of them refer to the specifics of the internal circuit design and have no effect on the “external” parameters of the PSU, others are employed for cost-saving or manufacturability reasons. In other words, such pretty-looking labels are often nothing but marketing noise. It can also be viewed as white noise since it carries no valuable information. There is no sense in testing most of such marketing innovations but we’ll try to enumerate the most frequent ones for you to know what they mean. If you think we miss something, post a comment and we’ll make an addition to this article.

Dual +12V Output Circuits

In good old times PC power supplies used to have one power rail for each of the output voltages (+5V, +12V, +3.3V, and a couple of negative voltages), and the maximum output power on each of the rails was not higher than 150-200W. It’s only in some high-wattage server-oriented power supplies that the load on the +5V rail could be as high as 50A, i.e. 250W. This situation was changing as computers required ever more power and the distribution of power consumption among the different power rails was shifting towards +12V.

The ATX12V 1.3 standard recommends a max current of 18A for the +12V rail and this is where a problem occurred. It was about safety regulations rather than about increasing the current load further. According to the EN-60950 standard, the maximum output power on user-accessible connectors must not exceed 240VA. It is thought that higher output power may with a higher probability lead to various disasters like inflammation in case of a short circuit or hardware failure. Obviously, this output power is achieved on the +12V rail at a current of 20A while the PSU connectors are surely user-accessible.

So, when it became necessary to push the allowable current bar higher on the +12V rail, Intel Corporation, the developer of the ATX12V standard, decided to divide that power rail into multiple ones, with a current of 18A on each, the 2A difference being left as a small reserve. Purely out of safety considerations, there was no other reason for that solution. It means that the power supply does not necessarily have to have more than one +12V power rail. It is only required that an attempt to put a load higher than 18A on any of its 12V connectors would trigger off the overcurrent protection. That’s all. This simplest way to implement this is to install a few shunts into the PSU, each of which is responsible for a group of connectors. If there’s a current of over 18A on a shunt, the protection wakes up. As a result, the output power of none of the 12V connectors can exceed 18A*12V=216VA, but the combined power on the different 12V connectors can be higher than that number.

That’s why there are virtually no power supplies existing with two, three or four +12V power rails. Why should the engineer pack additional components into the already overcrowded PSU case when he can do with just a couple of shunts and a simple chip that will be controlling the voltage in them (the resistance of a shunt being a known value, the current passing through the shunt can be known if you know the voltage).

But the marketing folk just couldn’t pass by such an opportunity and now you can read on any PSU box that dual +12V output circuits help increase power and stability, the more so if there are not two but three such lines!

You think they stopped at that? Not at all. The latest trend is power supplies that have and don’t have the splitting of the +12V rail at the same time. How? It’s simple. If the current on any of the +12V output lines exceeds the 18A threshold, the overcurrent protection becomes disabled. As a result, they can still embellish the box with the magical text, “Triple 12V Rails for Unprecedented Power and Stability”, but can also add there some nonsense that the three rails are united into one when necessary. I call this nonsense because, as I have written above, there have never been separate +12V power rails. It’s impossible to comprehend the depth of that “new technology” from a technical standpoint. In fact, they try to present the lack of one technology as another technology.

As far as we know, the “self-disabling protection” is currently being promoted by Topower and Seasonic and, accordingly, by the companies that are selling such PSUs under their own brands.

Short Circuit Protection (SCP)

Short circuit protection is obligatory according to the ATX12V Power Supply Design Guide. This means that it is implemented in all power supplies, even those that don’t explicitly mention such protection, that claim to comply with that standard.

Overpower (Overload) Protection (OPP)

This protects the power supply from overload on all of its outputs combined. This protection is obligatory.

Overcurrent Protection (OCP)

This protects the separate PSU outputs from overload (but not yet from short circuit). It is available on many, but not all, PSUs, and not for all of the outputs. This protection is not obligatory.

Overtemperature Protection (OTP)

This protects the PSU from overheat. It is not required and is not implemented often.

Overvoltage Protection (OVP)

This protection is obligatory, but is only meant for critical failures. It works only when some output voltage shoots 20-25% above the nominal value. In other words, if your power supply yields 13V instead of 12V, you must replace it as soon as possible, but its protection is not required to react yet because it is designed for even more critical situations.

Undervoltage Protection (UVP)

As opposed to too-high voltage, too-low voltage cannot do much harm to your computer, but may cause failures in operation of the hard drive, for example. This protection works when a voltage bottoms out by 20-25%.

Nylon Sleeve

Soft braided nylon tubes on the PSU’s output cables help lay them out neatly inside the system case.

Unfortunately, many manufacturers have switched from the undoubtedly good idea of using nylon sleeves to the use of thick plastic tubes, often screened and covered with a paint that shines in ultraviolet. The shining paint is a matter of personal taste, of course, but the screening does not do anything good to the PSU cables. The thick tubes make the cables stiff and unwilling to bend, which makes it hard to lay them out in the system case properly and is even dangerous for the power connectors that have to bear the pressure of the cables that resist the bending.

This is often advertised as a means to improve the ventilation of the system case, but I can assure you that the tubes on the power cables have but a very small effect on the airflows inside your computer.

Dual Core CPU Support

This is nothing but a pretty-looking label. Dual-core processors do not require any special support from the power supply.

SLI and CrossFire Support

Yet another pretty-looking label that means two power connectors for graphics cards and an ability to yield as much power as is considered sufficient for a SLI graphics subsystem. Nothing else stands behind that label.

The PSU manufacturer can get some certificate from a graphics card maker but such a certificate means nothing but the mentioned connectors and high wattage. The latter is often much higher than a typical SLI or CrossFire system may need. After all, the manufacturer must explain the necessity of buying PSUs of indecently high wattage to the customer, so why not do that by simply putting a SLI Certified sticker on the product?

Industrial Class Components

One more pretty-looking sticker! Industrial class components are components that can work in a very wide range of temperatures. But what’s the purpose of installing a chip capable of working under -45°C into a PSU if this PSU will never be used in such cold weather?

Sometimes the term industrial class components refers to capacitors meant for operation under a temperature up to 105°C, but that’s all clear here, too. The capacitors in the PSU’s output circuits heat up by themselves and also located very close to the hot chokes are always rated for a temperature of 105°C max or their service life would be too short. Of course, there is a much lower temperature inside the PSU, but the problem is that the service life of a capacitor depends on the ambient temperature. Capacitors rated for higher max temperatures are going to last longer under the same thermal conditions.

The input high-voltage capacitors work almost at the temperature of ambient air, so the use of somewhat cheaper 85°C capacitors there doesn’t affect the PSU’s service life much.

Advanced Double Forward Switching Design

Alluring the potential customer with mysterious terms is a favorite trick of the marketing department.

Here, the term means the topology of the PSU, i.e. the general concept of its circuit design. There are quite a number of different topologies. Besides the double forward converter, PC power supplies may use a forward converter or a half-bridge converter. These terms are only interesting for a specialist and don’t mean much for an ordinary user.

The choice of the particular PSU topology is determined by a number of reasons like the availability and price of transistors with required characteristics (they differ greatly depending on the topology), transformers, controller chips, etc. For example, the single-transistor forward converter is simple and cheap but requires a high-voltage transistor and high-voltage diodes on the PSU output, so it is only used in inexpensive low-wattage models (high-voltage diodes and transistors of high power are too expensive). The half-bridge converter is somewhat more complex, but has a two times lower voltage on the transistors. So, this is generally a matter of availability and cost of the necessary components. We can predict, for example, that synchronous rectifiers will be sooner or later used in the secondary circuits of PC power supplies. There’s nothing new in that technology, but it is too expensive as yet and its advantages don’t cover its cost.

Dual-Transformer Design

The use of two power transformers – usually in high-wattage (1000W and higher) power supplies – is a purely engineering solution, too. It doesn’t affect the PSU’s characteristics but it may be just handier to distribute the huge wattage of modern PSUs between two transformers, for example when a full-wattage transformer just wouldn’t fit into the PSU housing. However, some manufacturers present the dual-transformer topology as a means to achieve more stability, reliability, etc, which is not exactly true.

RoHS (Reduction of Hazardous Substances)

This is a new European Union directive that limits the use of certain substances in electronic equipment since July 1, 2006. It restricts the use of lead, mercury, cadmium, hexavalent chromium, and two bromides. For power supplies this mainly means a transition to non-lead solders. Yes, we are all for ecology and against heavy metals, but a too hasty transition to new materials may have unpleasant consequences. You may have heard the story about Fujitsu’s MPG hard drives which would die due to a failure of Cirrus Logic controllers that had a packaging made of some new environment-friendly compound from Sumitomo Bakelite. The elements of the compound facilitated the migration of copper and silver that formed bridges between interconnects inside the chip case. As a result, the chip would fail almost certainly after 1 or 2 years of operation. The compound was abandoned eventually, and the involved companies exchanged lawsuits, but nothing could restore the data that were lost with the hard drives.

Our Test Equipment

In our PSU tests our main task is to check it out at different loads, up to the maximum one. Many reviewers used to employ an ordinary PC for that purpose, installing the tested PSU into it. This method had two drawbacks: you had no control over the amount of power consumed by the PC, and it was hard to load a really high-wattage PSU. The second problem is especially crucial today when the PSU manufacturers have started a race for reaching as high a wattage rating as possible, and the capabilities of their products have by far exceeded the demands of a typical PC. Of course, one may argue that there’s no point in testing PSUs at higher loads if no real PC needs more than 500W, but it would be odd not to check out a high-wattage PSU through its entire load range if we begin to test it at all.

We use an adjustable load with programmable control to test PSUs in our labs. The testbed is based on the well-known feature of metal-oxide-semiconductor field-effect transistors (MOSFETs): a MOSFET limits the current flowing in the drain-source circuit depending on the voltage in the gate.

Above you can see a simple schematic of a MOSFET-based current regulator: connecting the circuit to a PSU with an output voltage of +V and turning the lever of the variable resistor R1 we are changing the voltage on the gate of the transistor VT1 thus changing the current I that flows through it from zero to maximum (determined by the characteristics of the transistor and/or tested PSU).

This circuit is not perfect, though. When the transistor heats up, its characteristics change somewhat and the current I changes, too, although the control voltage on the gate remains the same. To solve the problem, we need to add a second resistor R2 and an operation amplifier DA1 into the circuit:

When the transistor is open, the current I flows through its drain-source circuit and through the resistor R2. The voltage on the latter equals U = R2*I, according to Ohm’s law. This voltage goes to the inverting input of the operation amplifier DA1. The non-inverting input of the same opamp receives the control voltage U1 from the variable resistor R1. The opamp has such properties that it tries to maintain the same voltage on both its inputs by changing its output voltage, which in our circuit goes to the MOSFET’s gate and regulates the current flowing through it.

Suppose the resistance R2 equals 1 Ohm, and we’ve set a voltage of 1V on the resistor R1. Then the opamp will change its output voltage in such a way that the voltage drop on the R2 is 1V, too. The current I now equals 1V / 1Ohm = 1A. If we set a voltage of 2V on the R1, the opamp will react by setting the current I at 2A, etc. If the current I and, accordingly, the voltage on the resistor R2 changes due to the heating-up of the transistor, the opamp will immediately adjust its output voltage to bring them back.

Thus, we have a perfectly controllable load that allows to smoothly change the current from zero to maximum with a turn of a lever. It maintains the value you set automatically for as long as you wish. It is also compact, being much easier to handle than a clumsy set of low-resistance resistors connected in groups to the tested PSU.

The maximum amount of power dissipated on the transistor is determined by its thermal resistance, the maximum allowable die temperature, and the temperature of the heatsink it is installed on. Our testbed employs International Rectifier IRFP264N transistors (a 168KB .PDF file) with an allowable die temperature of 175°C and a die-heatsink thermal resistance of 0.63°C/W. The cooling system of the testbed can keep the temperature of the heatsink under the transistor below 80°C (yes, the fans that do it are very noisy). Thus, the maximum amount of power dissipated on one transistor is (175-80)/0.63 = 150W. To reach the desired total load we use several parallel-connected loads designed like described above – they all receive a control signal from the same DAC. A parallel connection of two transistors with one opamp could be used, too, and the maximum dissipated power would be half as high as with one transistor.

There is one step left to a fully automated testbed: we can replace the resistor with a PC-controlled DAC to regulate the load from the PC. Connecting several such loads to a multi-channel DAC and installing a multi-channel ADC to measure the output voltages of the tested PSU in real time, we get a full-featured testbed for testing PC power supplies in the entire load range and at any variants of load distribution.

The photo above shows our tested in its current form. The two top blocks of heatsinks cooled with powerful 120x120x38mm fans carry the load transistors of the +12V channels. The more modest heatsink cools the transistors of the +5V and +3.3V channels. The gray block connected to the LPT port of the controlling PC contains the mentioned DAC, ADC and accompanying electronics. Measuring 290x270x200mm, this testbed allows testing PSUs with a wattage rating up to 1350W (up to 1100W on the +12V rail and up to 250W on the +5V and +3.3V rails).

We wrote a special program to control the testbed and automate certain tests. You can see its screenshot above. It allows to:

Especially valuable is the automatic construction of cross-load diagrams. To draw such a diagram, you need to measure the PSU’s output voltages at all possible combinations of loads, which involves a very large number of measurements. Performing such a test manually would require great assiduity and a lot of time from the human tester. Our program takes in the specified PSU parameters, builds a chart of allowable loads and goes along this chart, measuring the output voltages at each step. The results of the measurements are presented as a diagram. The whole process takes 15 to 30 minutes depending on the PSU wattage and measurement step, but does not require man’s intervention.


Efficiency and power factor measurements

We use additional equipment to measure the efficiency and power factor of a PSU: the tested PSU is connected to the 220V mains via a shunt. A Velleman PCSU1000 oscilloscope is connected to the shunt as well. Its screen shows an oscillogram of the current consumed by the PSU, so we can calculate the amount of power consumed from the mains. Knowing also the load on the PSU we set by ourselves, we can calculate the efficiency. The measurements are performed automatically: the above-described PSUCheck program can take all the necessary data from the oscilloscope’s software that is connected to the PC via USB.

To ensure maximum precision of the result, we count in the output voltage deflection when measuring the output power of the PSU. For example, if at a load of 10A the output voltage of the +12V rail sags to 11.7V, the corresponding item in the efficiency calculation will be 10A * 11.7V = 117W.


Velleman PCSU1000 oscilloscope

This very oscilloscope is also used to measure the output voltage ripple. The measurements are performed for the +5V, +12V and +3.3V rails at the maximum permissible load. The oscilloscope is connected via a differential setup with two shunting capacitors (this connection is recommended in ATX Power Supply Design Guide):


Output voltage ripple measurement

We’ve got a dual-channel oscilloscope, so we can measure the ripple on one rail at a time. We repeat the measurement three times and the three resulting oscillograms, one for each tracked rail, are combined into a single picture:

The oscilloscope’s settings are indicated in the bottom left corner of the picture: the vertical scale is 50 millivolts/division, and the horizontal scale, 10 microseconds/division. The vertical scale usually remains the same in our tests but the horizontal one can change: some PSUs have low-frequency pulsation at the output and we provide another oscillogram, with a horizontal scale of 2 milliseconds/division, for them.

The speed of the PSU fans, depending on its load, is measured in semi-automatic mode. Our optical tachometer Velleman DTO2234 doesn’t have a PC interface, so its readings have to be entered into the PC manually. During this process the load on the PSU is changing in steps from 50W to the permissible maximum. The fan speed is recorded after the PSU has worked for 20 minutes or more at each step.

We also measure the growth of temperature that is passing through the PSU using a dual-channel thermal-couple thermometer Fluke 54 II. One of its sensors measures the air temperature in the room and the other sensor measures the air temperature at the output of the PSU. To increase the repeatability of the results, we fasten the second sensor on a special fixed-height stand placing it at a definite distance from the PSU. So, the sensor is in the same position relative the PSU in every test, which ensures identical conditions for each tested product.

The resulting diagram shows both the fan speed and the difference in the air temperature. This helps evaluate the specifics of the particular cooling system better.

When it is necessary to check out the measurement precision of the testbed or calibrate it, we use a digital multimeter Uni-Trend UT70D. The testbed is calibrated using an arbitrary number of measurement points placed at different parts of the available range. In other words, to calibrate by voltage we connect a regulated PSU to the testbed. The output voltage of the PSU is changed in steps from 1-2V to the maximum level measured by the testbed on the appropriate channel. At each step we enter the precise voltage value, as reported by the multimeter, into the testbed control program and the program fills in the correction table. This method of calibration allows achieving a high precision of measurements through the entire range of values.