You are here

SPCR's Revised PSU Testing System

March 22, 2004 by Mike Chin

PC power supplies are poorly understood by typical computer users. Most PC hardware web site reviewers also seem to be "typical computer users" when it comes to power supplies. Web reviewers of PSUs generally show minimal understanding of fundamental electricity / electronics concepts, standards and practice, and use inadequate evaluation equipment for reviews. Even when data is collected, little attempt is made to understand what it really means and how relevant or accurate it is. Perhaps it comes down to the simple fact that PC hardware review sites are created and maintained by computer enthusiasts, who have little or no grounding in electronics.

SPCR's PSU reviews have always tried to do more, right from the beginning when we, too, had little in the way of appropriate testing equipment. I have the small advantage, perhaps, of having studied electronics briefly, and of having run an audio hifi store for some years; my partner, a talented electronics technician, taught me a lot.

LOAD TESTING: UNDER WHAT CONDITIONS?

After an angel presented SPCR with the gift of a DBS-2100 PSU load tester, we basked in the luxury of having a piece of test equipment rarely seen outside a power supply design engineering lab. Still, the PSU load tester is only one device, and it does not tell us everything we want to know.

One of the more challenging issues has been to observe or measure a PSU's performance in a realistic thermal environment. Testing by safety certification agencies such as UL, CSA and others is conducted in a controlled temperature chamber. As far as I can determine, the ambient temperature is controlled to remain constant throughout testing. This procedure is probably perfectly fine if all the validation testing by agencies are done at the same temperature, preferably a high enough temperature to reflect real conditions in a PC. But it appears that agencies test to validate the manufacturer's claims only, not in accordance with a universal standard.

In other words, if a manufacturer submits a power supply specified to produce 400W total power across its various voltage lines at 25°C with 115VAC input, then that claim is verified, and safety checks are made under precisely those conditions. But not at 45°C or with 90VAC input. This is because agency regulations are concerned chiefly with safety or EMI (electromagnetic interference), not with the performance aspects of power supplies.

The test ambient temperature often specified for many retail-level PSUs is 25°C . In some technical specification documents, usually not intended for consumer viewing, a typical qualifying phrase is "0ºC~25ºC for full rating of load, decrease to zero Watts O/P at 70ºC." The meaning of this phrase has been discussed here before in the PSU forum and in the Power Shmower text box in Recommended PSUs.

It means that power output capacity declines with temperature rise from maximum at 25°C to zero watts at 70°C. The gist of the issue is that a PSU in a typical computer would never see 25°C unless it was turned off. 35°C is much more realistic, and 40°C is not uncommon in a high performance low noise PC under full load. So a PSU rated for 400W output at 0-25°C and 0 watts at 70°C cannot produce its rated power in the thermal environment of a typical PC. If you assume that this power drop is linear, then max power capacity will drop by ~9W for every degree over 25ºC. At the relatively cool internal PC case temperature of 35ºC, this "400W" PSU could only produce 310W max. At 40ºC, max output would drop to 265W. At 45ºC, you're down to 220W.

Makes you want to run and check the fine print on your PSU, doesn't it?

In most specification documents released to consumers, the test ambient temperature is not even specified. So PSU power ratings on retail packages tell little about performance in real working conditions.

The fan in most higher quality PSUs is thermally controlled to speed up with increased temperature. Naturally, ambient temperature has a big influence on the noise of the PSU, not just its load. While most makers of quiet PSUs only claim low noise at minimal loads, as end users, it is useful for us to know at what load and temperature the fan begins to spin up and make more noise: All the more reason to test the PSU under realistic ambient temperatures.

OEM and system integrator markets are much tougher to satisfy than retail markets, because their buyers are often part of the engineering team. Obsfucation and snake oil marketing techniques don't work on them. These buyers usually seek PSUs rated for full power at 50°C. In mission critical corporate systems, in servers and in other tough applications, high temperature performance is a must. Companies and brands that sell successfully to this marketplace usually make much better PSUs than those who focus solely on the retail market.

SILENTPCREVIEW PSU TESTING PLATFORM

This article was originally begun to describe the latest addition to SPCR's PSU test platform, but it's necessary to understand what came before. The table below is a summary.

FUNCTION TOOL
Measure temperature of ambient air, case, and PSU exhaust
Digital readout thermometer. There are several in the lab that are used. (Like DigiDoc.) They measure within ~1°C of each other, which is good enough for our purposes. Powered by the PSU being tested. (1~2W power draw.)
Measure voltages across fans and DC output line
Heath / Zenith SM-2320 multimeter. This is an ordinary multimeter. It has been compared against a much more expensive lab instrument and comes very close (within 2%) on readings of 0~20VDC.

Load PSU to specific DC output power loads for each voltage line

DBS-2100 PSU load tester. Made specifically for testing computer power supplies, it consists of a large bank of high power precision resistors along with an extensive selection of switches on the front panel calibrated in Amps (current) and grouped into 6 voltage lines: +5, +12, -12V, +3.3, -5, +5SR. Leads from the PSU plug into the front panel, and there are taps for taking voltage readings for the 3.3V, 5V and 12V lines.

Measure AC power, power factor (PF), VA, AC line voltage
Kill-A-Watt Power Meter. An inexpensive consumer power meter with very good accuracy and a host of useful functions.

Measure noise in dBA from 1 meter distance B&K model 1613 sound level meter. This professional caliber SLM dates back to 1978, weighs over 10 pounds, and is completely analog in design. It has a dynamic range that spans over 140 dB. The unit's absolute sensitivity reaches below 0 dBA. A quiet environment is a prerequisite to low noise testing; the lab has been measured down to ~17 dBA at night, and a 12 dBA adjacent room is also available for any PSUs that are quieter.

Thermal environment directly related to the power delivered. Custom Test Jig. The NEW ADDITION to the lab that promoted the writing of this article. It is fully explained below.
Vary AC Voltage to consider effects of brownouts and other real-world conditions on PSU performance California Instruments 801RP Variable AC Power Supply: Coming Very Soon! 800VA capacity, 0 - 270VAC range, 16 - 500 Hz frequency range. More on this item and how it will be used when it arrives in the lab!

More About Individual Test Gear

The DBS-2100 load tester enables simplified, controlled PSU load testing from as little as a few watts to a rated maximum of 614W (in DC voltage output). It is equipped with 2 AC outlets (individually fused with 7A 250V fuses) and 4 exhaust fans on the back panel. A bypass switch toggles the fans on or off so that noise measurements on the PSU can be made. The resistors get very hot under high loads, so it is important not to leave the fans off for long.

The California Instruments 801RP Variable AC Power Supply is a professional lab-grade instrument with very tight (0.5~1%) output voltage tolerances. Full details about this product is available on their website, http://www.calinst.com/rpseries.html. This instrument will help us clearly differentiate PSUs that are honestly rated an from those that are high powered in claims only. More later after this expensive new acquisition arrives in the lab.

THANK YOU to all the sponsors, friends and patrons of SPCR whose contributions helped to make this equipment purchase possible!

REALISTIC OPERATING ENVIRONMENT TEMPERATURES

Anyone who plays around with computer hardware for any length of time quickly learns that the temperature of the air inside a typical PC easily reaches over 40°C especially near or above the CPU when the system is under load. This is precise the area in which a PSU is normally situated. For testing purposes, it would be ideal to use a real computer but the problem is that then the PSU load cannot be easily controlled; it must be done with CPU stress and system stress software, which is usually an all-or-nothing approach.

Two different strategies were used for previous PSU reviews to simulate a realistic thermal environment:

1) A hot light bulb placed at the intake vents of the PSU while it was being load tested on the open test bench, as shown below. This approach suffered from lack of thermal control, or any simulation of airflow in a case. Furthermore, the bulb needed to match the AC power drawn by the PSU. Above 150W or so, this was impractical to do, due to the absence of appropriate wattage bulbs and the tedium of changing hot bulbs.

2) A light bulb placed inside a mid-tower case in the same approximate position as the CPU. This approach simulated the rising heat aspect of the real PC environment but suffered the same problem as 1): No suitable high power bulbs.

THE NEW RIG

The total amount of heat inside any PC is equal to the total AC power it draws. That is, if 100W is the total power of a PC as measured by an AC power meter like the Kill-a-Watt, then 100W is the amount of heat that will be dissipated in the PC. Some of that heat will be generated by the power supply itself. During AC/DC conversion, there is always some loss, ranging from 40~20% of the total. The rest is all delivered as DC voltage to the components in the system and those components generate heat.

The objective was to allow the heat in the test environment of the PSU to be directly proportionate to the amount of power drawn by the PSU. In a real system, this is what happens.

With the DBS-2100 PSU load tester, the DC power from the PSU is dissipated by the large resistors inside. They get very hot under load, which is why the four 80mm fans are necessary.

It occurred to me that if the heat from the PSU load tester could be transferred into the operating environment of the PSU being tested, this would be the closest thermal simulation to actual PSU operating conditions that could be devised.

Eureka!

There seemed only two practical ways to achieve this:

1) Remove the components from inside the DBS-2100 PSU load tester and reassemble them inside a typical PC case. One look at the complex internal wiring convinced me this was not a good idea. Too much could go wrong. There were too many places where I could goof.

2) Build a PC case or other similar enclosure with ducts that the four exhaust fans on the DBS-2100 PSU load tester blow into, hence not only transferring the heat into the PSU's working environment, but also simulating the case cooling airflow that almost all PC systems depend on. This was the option chosen.

Here is the first result of a couple afternoons in the garage, working with scrap plywood.



A light in the bottom of the box: So you can see it better.

The internal dimensions of the box measure approximately 40 x 40 x 20 cm (16" x 16" x 8"). This is about 32 liters, slightly smaller than a typical mid-tower case, which is closer to 40 liters. However, this box is completely empty, while a typical mid-tower PC will be filled with a motherboard, CPU/HSF, optical and hard drives, VGA card... with all the components in place, the typical mid-tower is likely to have about the same volume of air.

There is a cutout on one side of the box that the protruding fans on the DBS-2100 PSU load tester fit into perfectly. The box has no bottom; none was required, and leaving the bottom open made it easier to get to the inside during assembly.

You can see that there is an opening on one top corner: This is where any ATX form factor PSU fits in perfectly. The bottom and back are open to accommodate multiple or 120mm fan power supplies. See the photo below.

Preliminary trials showed that the rig works pretty much as intended. There was only one problem: The tight fit of the PSU caused it to become mechanically coupled to the box, and the box resonated enough so that it added 3~10 dBA of additional noise to the PSU fan noise. This was unacceptable, so a rethink was necessary.

Back to the drawing board.

My solution in the end was to cut one side and the top away so that the PSU would be placed on the rig, rather than mounted in it.

One side is a block of foam, and all the surfaces the PSU touches are lined with soft damping materials or foam weather-stripping that keeps it mechanically decoupled from the box. You can see in the above photo the shadow of the thermal sensor that has been fixed about 1" below the back bottom edge of the PSU being tested.

The above photo
shows a Seasonic Super Tornado PSU in the rig. It's plugged into a Kill-a-Watt AC power meter. The output cables from the PSU are long enough to reach the input panel on the load tester; this may be an issue with some PSUs that have short leads. On the top right is a digital thermometer powered via a 4-pin Molex from the PSU. The long row of slots along the front edge of the PSU load tester are the intake vents for the four fan that draw air across the hot resistors. (They are equivalent to the front intake vent in a typical PC case.) All the parts of the box that make contact with other surfaces (the bottom rim, the contact area around the load tester fans) are lined with weather-stripping for good air sealing (though it does not have to be perfect) and minimal vibration transfer.

One item missing in the above photo is the thermal sensor for the PSU exhaust air. It is positioned near the center of the exhaust grill, about 1/3 of the way down from the top, within 1cm of the grill. This is usually the hottest spot, but the sensor is moved around a bit for each PSU to find the hottest spot before being fixed in place.

Some quick testing showed this setup to be perfectly neutral acoustically, at least when the PSU fan is spinning slowly. At very high PSU fan speeds, there is some added resonance from the air being vibrated in the case, and perhaps a bit from the PSU panels as well. This needs to be examined in more detail. But in the range that matters to PC silencers (PSU noise of under 30 dBA), there is no difference in noise level whether the PSU is in the rig or placed on a piece of foam on the bench.

The whole assembly sits on a sturdy cabinet with drawers. Equipped with castors, it can be easily wheeled into the very quiet room adjacent to the main test room for super quiet PSUs (should they appear): The minimum noise level in the lab is about 16~17 dBA; in the adjacent room, I've measured as low as 12~13 dBA.

TESTING PROCEDURE

After many trials and experiments, I found that the following procedures provide the most consistent results:

A. Record the ambient temperature and noise.

B. Warm up the PSU before testing by turning it on at 65W load, leaving it running for a couple minutes, then setting the load to 50% of the rated output power and alllowing it to run for 15 minutes. Like all electronic devices, PSUs work more efficiently when they are warm.

C. Start measuring with the lowest power load and move up to max power. The test load points will be:

  • 65W
  • 90W
  • 150W
  • 200W
  • 250W
  • 300W
  • 400W
  • Max power

D. Allow the PSU to run at each power level for 3~5 minutes to ensure stability before taking any measurements.

E. Check and record voltages across each output line several times at several power levels, monitoring it as small variations in load are introduced on the 3.3V, 5V and 12V lines in turn.

F. Leave the PSU load tester fans on until just before temperatures are recorded. At lower power levels, there's hardly any change in the internal temperature. At higher power levels, turning the fans on actually increases the internal test box temperature (and sometimes, the PSU exhaust temperature) up by a degree or two. Convection is certainly not enough to move the hot air from the PSU loader into the test box.

G. Record the noise level of the PSU at each power level at the same time that the temperature is recorded.

H. Monitor and record the voltage across the internal PSU fan(s) after 3~5 minutes at each power level. This is not always possible, as some PSUs are wired so tightly that tapping into a fan line requires the precision and patience of a surgeon, which cannot always be mustered up.

The above refers only to documentation using instrumentation. There is always much listening and taking of notes regarding the quality of the noise as I hear it.

That's pretty much covers the current PSU testing system... until the new variable AC power supply arrives. :)

Look for reviews of new Enermax, Zalman, Fortron, and Seasonic PSUs in the near future.

* * *

POSTSCRIPT ADDED April 13, 2004: Please check the next page.

POSTSCRIPT — April 13, 2004

Shortly after this article was originally published, a member of the SPCR forum posted a strong criticism about the new PSU Testing System, especially the Thermal Simulation Box. Rory B. wrote:

"I am going to raise a stink about SPCR's new thermal box. In every picture of that PSUtester that I have ever seen, there are four fans on the rear. Mike described the box as being not air-tight with air being able to escape around the cables. He went on to say that he could feel the air coming out the holes. The logic is that the power resistors in the PSUtester will supply the heat for the box.

"But as long as the heat for the box is being blown in by the four 80mm fans which also provide back-pressure, the testing methodology is flawed.

"In many systems, the power supply's fan will be operating against negative pressure in the box, and the fans will need to spin faster due to the axial fan's relative inefficiency at low speeds. The thermal box, however, is a POSITIVE PRESSURE testing situation, and the four fans will be pressurizing the box. Because the pressure inside the box is already forcing air out through the power supply, the fans don't NEED to spin up! There are already four 80mm fans working to cool the PSU!"

Rory B.'s comment forced me to reexamine all the various aspects of this thermal simulation box. It helped to set down all the salient points:

1) The PSU Load Tester absorbs all the power output of the PSU, and thus gets very hot. Generally, at least twice as hot as the PSU. (Because at least ~65% of the AC power drawn by a PSU is sent to the Load Tester, while only ~35% or less of the power remains as heat in the PSU.)

2) The design and shape of the PSU Load Tester does not allow convection alone to keep the big resistor banks cool when more than ~150W is being dissipated: They must be fan cooled or eventually suffer heat damage.

3) Each fan gets 12VDC across its terminals, and there is definitely positive pressure inside the box. It means that unless the PSU cooling fans are blowing huge amounts of air, turning the PSU Load Tester fans on increases the flow of air through the PSU. This effect can be felt as increased airflow at the PSU exhaust outlet.

4) The fans themselves are Max Flow brand, model 8025D1-HSPL, 12VDC, 0.18A. The brand does not have its own website, but I found a site that lists some of its products. However, this particular model was not listed. Their AC version of the same fan (also rated for 0.18A) is said to push 22/25 cfm.

Checking my fan database for 80mm 12VDC fans rated close to 0.18A, there is a fairly broad range of airflow specs, from under 20 cfm to over 30 cfm. On the basis of simple feel and ballpack averaging, I will chose 25 cfm as a guesstimate for this model's rate airflow (free and unimpeded).

There are four fans in parallel, so one could say they represent 100 cfm, but we know this is not true, because the airflow is always measured in free air without resistance. The air intake for the load tester is a row of slots on the front edge that represents a much smaller area than that of the fan blades. Then the air has to pass through a huge jumble of cables as well as the large cylindrical resistors themselves.

My guess is that the total real airflow into the Thermal Box is somewhere in the vicinity of ~60 cfm.

What does this ~60 cfm airflow do?

1. It helps to cool the big hot resistors. The cooling is not 100% effective. In other words, the cooling airflow does not drop the temperature of the resistors down to ambient; some of the heat is retained by the resistors.

2. It forces a lot of the heat from the hot resistors into the Thermal Box and increases the air temperature inside the box.

3. From a case simulation point of view, it's kind of like running 2~3 Panaflo 80L fans at 12V for front intake. This is definitely more airflow than any serious PC silencer would use (because of the noise). However, the absence of any large exhaust vent or vent makes it a strange case.

What happens if the airflow is reduced?

Without getting into thermodynamics or fluid dynamics, it is safe to say that just how much heat the PSU gets exposed to in the Thermal Box depends on not only the power that's fed into the resistors, but also the amount of airflow across those resistors. The following was verified experimentally:

  • Without any airflow, the PSU actually "sees" less of the heat because it tends to stay in and around the resistors.
  • With a lot of airflow, the amount of heat is spread across a bigger volume of air, and so the average temperature of the air in the box is also not that high.
  • The trick is to get a level of airflow that allows the Load Tester resistors to be cooled well enough, and transfer enough of the heat into the Thermal Box to get it hot, yet not allow too much in the way of positive pressure so that the PSU is unduly helped in keeping itself cool.

A DUAL-PRONGED SOLUTION

1. PSU Load Tester Fans set to 5V —

Some exploration in the maze of wiring inside the PSU Load Tester revealed a 5V terminal that is easily accessed for the fans. Making the switch means all the fans in the Load Tester are driven off the 5V line of the PSU being tested. (This represents a load of ~2W, which is compensated for in the PSU review data.)

The result is much reduced but still steady airflow. The pressure is low enough that it is difficult to discern any increase in the PSU exhaust with a PSU that uses a low airflow Panaflo 80L fan. If we guesstimate by assuming a linear relationship between airflow and voltage, we're looking at 25~30 CFM for the Load Tester fans at 5V. (It is also far quieter, registering just ~26 dBA/1m.)

2. 80mm Exhaust Fan in the Thermal Simulation Box —

Virtually every PC has one, even quiet PCs: Why not this simulation box? A Panaflo 80mm Medium speed fan was chosen. This fan is rated for 32 CFM at 12V, and produces about 13~14 CFM at 5V, which is the intended setting. Why? Because 5V is about the level that SPCR readers would find this fan acceptably quiet, and it still provides a useful amount of airflow. This exhaust fan also ensures that the PSU does not benefit unfairly from positive airflow.

As you can see in the photo above, a rough 4" square hole was made on the "back" panel of the Thermal Box, about where an exhaust fan would go in a real case. A hole was cut out for the fan in a 4" square piece of 1" thick foam. Then the fan was inserted into the foam, and the foam+fan then wedged into the hole. The foam insulates the fan's vibrations from getting into the structure of the box and maintains a good seal.

The end result of these changes is a PSU Thermal Simulation box that closely resembles a typical SPCR PC case for airflow, and still uses the DC output of the PSU to heat the case as in a real PC.

For a practical illustration of what this change means for testing results, please check the Postscript to the Enermax NoiseTaker 475 PSU review.

* * *

Discuss this article in our Forums.

Sections: 

Google

www SPCR