A little over a year after our last major update, we’ve made changes extensive enough to consider this the fourth iteration of SPCR’s power supply testing system. PSU reviews have been popular bread-and-butter articles at SPCR for years, partly because of our close attention to detail, fundamentally sound test procedures, good test gear and thorough acoustic analysis. It is also because there were few serious PSU reviews done by other hardware web sites. Now, we’ve added crossloading, low AC input voltage, and ripple testing to complete our mix of PSU tests.
Oct 30, 2006 by Mike Chin
** POSTSCRIPT on page 7 added Dec 22, 2006: v4.1 or Bonefish edition **
A little over a year after our last major update, we’ve made a series of changes extensive enough to consider this the fourth iteration of SPCR’s power supply testing system. PSU reviews have been popular bread-and-butter articles at SPCR for years. This is partly because of our close attention to detail, fundamentally sound test procedures, good test gear and thorough acoustic analysis. It is also because there were very few serious PSU reviews done by other hardware web sites. In this context, SPCR PSU reviews really stand out.
Recently, some worthy competitors have emerged. These include extremeoverclocking, xbitlabs, pcperspective, and the latest, johnnyguru. We welcome these serious PSU reviewers to the field. There are still too few sane voices trying to cut through the raucous babble of hyper-marketing that surrounds power supplies for computers these days. Our worthy colleagues have also helped us to reexamine and improve our own test resources and techniques, and we thank them for the impetus.
The changes in our system add several new items to our long list of tested parameters:
Other changes include:
For the benefit of new readers, this article covers our entire PSU testing system as it stands today. You do not have to refer back to articles describing the earlier versions… unless you want to.
For the record, we first began testing power supplies because they are one of the major sources of noise in a PC, the others being hard drives and fans. The PSU is also a major source of heat; in converting AC voltage from the wall to DC for the computer components, the power supply generates heat as a "waste" byproduct. In most systems, fans are required to remove the heat produced by various computer parts, such as the power supply. The greater the heat, the faster the fans need to spin in order to remove the heat. The relationship between fan noise and fan speed is almost linear. Our PSU test procedure began by examining noise primarily, but quickly expanded to include efficiency, power delivery, voltage regulation, temperatures and many other parameters that describe the electrical, thermal and acoustic characteristics of a power supply.
The photo below shows all the components of our power supply testing system except for the silent PC that runs the audio and oscilloscope software, the microphone preamp and the external digital sound box. Unlike commercial power supply testing systems that can often be fully automated, our customized system is completely manual, with fans that can be completely turned off/on at will to serve our particular needs, one of which is very low ambient noise. There’s no way to measure a quiet power supply if the automated power supply tester is whirring, whooshing and whining away with multiple fans. As far as we know, all automated power supply testers have built in fans that are non-defeatable. This makes them useless for us.
From the left: Sennheiser ME 66 shotgun microphone, custom built wood PC thermal simulation box with Nexus 120 fan and internal 21A loading resistors for the 12V2 line, digital thermometer to monitor in/out temperatures of the power supply, a test power supply in the "hot seat", digital multimeter to monitor the PSU fan voltage, DBS-2100 PSU load tester integrated with PC thermal simulation box, B&K 2203 sound level meter in front, 120/240 VAC dedicated outlets, high accuracy Extech 560 digital multimeter to measure output voltage and current, Extech 380803 AC power analyzer/datalogger, 20A 0~140VAC variac, USB Instruments DS1M12 digital oscilloscope to measure / display AC ripple, LCD monitor for silent PC running sound and oscilloscope software, along with spreadsheet software to calculate PSU settings and tabulate measurements. For more details on all of our test gear, please check pages 5 and 6.
PSU TEST PLATFORM CONCEPT
We examine power supplies in a number of different ways, starting first with their electrical characteristics. The various electrical tests are detailed in the next section.
There are two aspects of our test platform that make it unique in the world of PSU testing. They include:
These particular aspects are important enough for us to present an overview about them.
1) Thorough Acoustic Testing & Analysis
We use three methods of assessing noise:
2) Thermally Realistic Conditions
A primary feature of the test system is a simulated mid-tower case with modest airflow. The heat generated in the load tester by the output of the tested PSU is forced directly into the simulated mid-tower case. The higher the output power, the greater the heat in the test box. It replicates the thermal conditions faced by a PSU in a real PC: The total power a computer draws at the AC outlet is precisely the amount of heat that it generates. Our test box is a close simulation of actual-use conditions for a PSU in a typical quiet mid-tower case. A very quiet 120mm fan fan performs the same role as a back panel exhaust in a mid-tower case. This fan is decouple-mounted in foam to minimize noise and is voltage limited to provide under 20 CFM of measured airflow. Noise from the PSU is measured from a meter away at every test load point, from 40W up to maximum rated power, with all other noise sources in the room turned off.
There are several advantages to this setup:
Photo shows wooden case used for thermal simulation of quiet, low-airflow mid tower case. Note exhaust fan decouple-mounted in foam. A thermal sensor is placed at the exhaust of the PSU, and its fan lead tapped to monitor voltage. The DBS-2100 PSU load tester pressed up against the wooden box actually feeds its internal heat into the box via four slow 80mm fans.
The four cooling fans of the DBS-2100 PSU load tester feeds the heat generated by the loading resistors into the thermal simulation case. NOTE: The wooden box as shown above was in an earlier incarnation.
It is important to note that the room in which the power supplies are tested is almost always very quiet, typically <20 dBA, more often <18 dBA. If it is a noisy day (due to rain, wind, or lawn mowers) we wait until it's quieter to conduct any recordings. listening tests or sound measurements.
THE TESTS: ELECTRICAL PERFORMANCE
These tests are centered around the DBS-2100 load tester. (Check the reference equipment list, pages 5 and 6.) The equipmment needed to run the tests include the DBS-2100 PSU load tester, Extech 380803 Power Analyzer / Data logger, Digital / Analog thermometers, USB Instruments DS1M12 digital oscilloscope, Extech 560 Digital Multimeter, and the Ling Bridge TDFC 2J-3 0~140VAC Variac.
1. Power Output
We check the ability of the PSU to deliver DC power from 40W all the way up to full rated power. At each power level, the balance of loads on each voltage line is kept proportional to the maximum ratings of each of the lines at full power. Both the voltages and the current in every line are manually measured to ensure accuracy.
The AC/DC conversion efficiency is the ratio of DC output power to AC input power, expressed in percentage, with 100% being perfect. It is calculated for each power point; typically it’s lowest at very low load, best around 50~80% of rated capacity, and a bit lower at maximum load. If a PSU requires an input of 400W in AC to deliver 300W in DC voltages, then it has an efficiency of 75%, at this point, and 25% of the power is lost as heat within the power supply.
3. Voltage Regulation
It is the ability of the power supply to hold each voltage rail (or line) at the required 12V, 5V or 3.3V under a wide variety of conditions. Standard requirements are ±5%, but many quality PSUs do much better. VR is checked at every power test we perform, from 40W all the way to full power, during crossload tests, and during low VAC input tests.
4. Power Factor
Power factor is the ratio of real power to apparent power. It is a difficult concept to summarize because power factor touches on many complex concepts such as alternating current, phase, reactive and resistive loads, etc. There are many detailed explanations of PF on the web, some of which will be linked in the reference pages of this article. PF has nothing to do with the DC output, and everything to do with the way the power supply handles the AC input. The higher the PF, the "easier" it is for the utility to deliver "real" power" demanded by the power supply. The best PF is 1.0. Power factor can be corrected passively, for typical values of 0.7~0.8. Active PF correction usually leads to near-perfect PF, typically >0.95. Better PF does mean lower energy consumption from the point of view of the utility, but not high AC/DC efficiency, which is an entirely different matter. We measure PF at every power output level.
5. Low Load
We check the power consumption of the power supply on standby (unit power switch on, plugged into AC). The results are of interest to anyone who cares about energy efficiency. We also check power demand with the unit turned on without any load, a condition that can cause problems for some high efficiency power supplies.
6. Ripple (new for V4 PSU test system)
It is the amount of alternating current (AC) that appears in the DC output lines. A switch-mode converter, the type used in all PC power supplies, tends to generate a significant amount of ripple voltage. Many PSU specs express ripple as a percentage; typically it is <1%. On the 12V line, this means <120mV; on 5V, it means <50mV. Very low ripple usually indicates a higher quality power supply. Higher than normal ripple can cause instability with some components. Ripple is measured and expressed in mV on all the voltage lines at every load. Screen capture images of the waveform may also be shown.
7. Low AC Input Voltage (new for V4 PSU test system)
Blackouts may be relatively rare in most areas of the developed world, but brownouts are much more common. Brownouts are periods of low voltage in utility lines that can cause lights to dim and equipment to fail. Also known as voltage sags, this is the most common AC power problem, accounting for up to 87% of all power disturbances. The severity, frequency and duration of voltage sags vary. PC power supplies are designed to continue working with some variance in AC voltage input, but just how well they perform under low VAC conditions is not well known. Certainly, intermittent, frequent voltage sags could affect PC stability; a PSU that does not handle low VAC well could be the source of mysterious computer instabilities that some of users face.
We now test PSUs with 110, 100, and 90 VAC input at 75% of rated power. The 75% load is about the highest power load that a PSU is likely to be see, as most builders factor in some overload headroom. It is a tough test, especially at <100VAC input, when the AC current will increase proportionately to maintain the same DC output power. A PSU that can handle 75% load well with low VAC will certainly do fine at lower loads. We monitor all AC and DC parameters as the VAC is lowered. A 0~140 VAC variac rated for 20A is used to control AC input voltage to the PSU in these tests.
8. Crossloading (new for V4 PSU test system)
Our power load testing follows standard industry protocol used by most organizations such as Intel, the 80 Plus program and the power supply manufacturers themselves. It is a proportionate load testing. In other words, the load placed on the various voltage lines at 90W or 200W total DC output is proportionate to the maximum ratings of each of the lines at full power. Crossloading describes a condition when the load is not proportionate. The most common occurrence of crossloading today occurs in a high power gaming rig where the 5V and 3.3V lines are at minimal (typically under 3A each, or 15W and 10W), but dual video cards and a power hungry CPU draw great amounts of power on the 12V line. The 12V load in an extreme gaming system could be as high as 350~400W at peaks, or 30~33A, while the power draw on the 5V and 3.3V lines amounts to well under 30W, less than 10% of the total DC output. We will examine the voltage regulation (stability) and AC ripple under crossload conditions.
THE TESTS: ACOUSTIC & THERMAL PERFORMANCE
Acoustics cannot really be separated from electrical or thermal performance because factors such as efficiency and cooling efficacy directly affect fan speed, which is the main cause of noise. However, the artificial separation helps to keep us organized about the tests themselves. We use thermometers, multimeters, a Bruel & Kjaer (B&K) model 2203 sound level meter (SLM), our digital audio recording system (see test equipment list, pages 5-6), and very careful listening to conduct the following tests.
9. Fan SPL and Fan Controller
The fan and the fan controller circuit are arguably the most critical factors in power supply noise. (With the exception of fanless PSUs, of course.) While the quality and speed rating of the fan sets the maximum and minimum noise limits, the thermal fan speed controller dictates how fast or slow the fans runs at various loads and temperatures. At every power level, we monitor the voltage fed to the fan by the fan controller l, measure the sound pressure level (SPL) of the fan with our SLM, and the in / out temperature of the airflow through the PSU. As the primary noise maker in the PSU, the fan naturally gets subject to a lot of scrutiny. We track down its origin whenever possible and report the manufacturer’s specifications (for better or worse), as well as any previous experience or information we may have about the fan.
This was alluded to in point 9, because the internal temperature of the power supply is a key factor in determining the output voltage of the fan control circuit. We monitor the temperature of the air flowing into the PSU and exhausting out of it. The temperature rise tells us something about the PSU’s cooling efficicacy, which is affected by AC/DC conversion efficiency, cooling fin design and fan speed. We monitor the air temperature in the immediate vicinity of the PSU test rig and ensure that it falls within 21~23°C. From now on, no PSU tests will be conducted outside of this temperature range; the fan curves can be affected by higher or lower temperatures.
11. Electronic Component Buzzing
Most people know what we mean, and surprisingly to some, the most effective way of detecting such noise is to listen. Often, it is not measurable with a SLM. Such noise most often comes from capacitors or inductors, and can range from a simple buzzing to a complex mix of several components buzzing, whining, screaming and humming. Once we know what we are listening for, we can isolate it so that it can be recorded, and the waveform shown in a 3D FFT display. Most of the time, this is unecessary. We report on any direct electronic component noise, and the conditions in which the sound is apparent. Often it is only evident under specific loads.
FAQ ABOUT CHANGES AND NEW TESTS
Some questions will be on some readers’ minds:
How will the new 120mm exhaust fan be used?
At up to 400W output load, the fan is set to 7V, which gives us about 20cfm in free air and 17 dBA@1m. It’s run this slowly to best simulate a quiet PC. Its noise is irrelevant, as the fan its turned off during recording or SPL measurement. At 400W load, the fan voltage will be turned up to 9V, where we obtain about 28cfm. This switch is perfectly legitimate and relevant. Anyone with a real system drawing 400W or more would want to turn the exhaust fan up to this speed at least. The noise from the fan still remains at just 20 dBA@1m. Once we reach 600W load, the fan is turned up to full speed, which provides about 40cfm in free air. The measured SPL is about 22 dBA@1m. Again, this is to replicate the most realistic real-use scenario; no one would have a >600W DC power draw system with just one slow 120mm fan blowing at 9V. Any fan-cooled power supply would completely mask the miniscule noise contribution of the Nexus 120 fan even at 12V. (More typically in such a high power rig, there would be many more fans, including very high speed video card cooling fans, with a total rated airflow of well over 100cfm.)
What will be the effect of changing the 8~10cfm 80mm exhaust fan to a 120mm fan with twice the airflow?
The most likely effect is that some power supplies may fare slightly better for noise in the new setup compared to the old. The fan-cooled power supplies with good fan controllers have generally tended to start speeding up the fan above ~150W output load. With the same power supplies, it is possible that the fan ramp-up speed might go up a bit higher, perhaps to 200W. The overall fan speed rise curve may be less steep than with the previous fan setup. We will go back and check a few of our most popular models to see if there’s any significant change. In general, however, we do not believe that any change will be dramatic, nor will it change current "rankings" in the Recommended Quiet PSU tables.
What are the potential effects of brownouts or voltage sags?
Here is a worst case scenario: Power supplies in some electronic equipment will fall out of regulation. Errors, due to erratic power supply performance, may creep into computer operations. Marginally performing electronic or electric devices will cease operation. Electrical interference increases and may affect computer and communication operations. Spikes generated by electrical machinery also greatly increase. Air conditioners, refrigerators and other motorized devices will generate local spikes. Industrial machinery (often miles away) may create additional spikes which may find their way on to the electrical distribution network, and into your equipment.
The vast majority of low cost UPS units will continuously cycle between power line and internal battery operation. UPS batteries will soon discharge, unable to generate additional back up power. If the system has not been shut-down, the UPS and computer may snap back on when power rises slightly, only to shut down again when the power line voltage falls back into the brown-out zone. Such "ON-OFF" operational cycling is unhealthy for UPS, batteries and connected electronics.
More typically, milder cases of brownouts or voltage sags are often the cause of otherwise unexplained computer instabilities, and the premature failure of components.
For those who are interested in the details of our test gear, please check the remaining two pages.
* * *
COMPLETE TEST EQUIPMENT LIST, PART 1
Here’s a list of all the gear we use currently for PSU testing:
USB Instruments DS1M12 digital oscilloscope and high resolution Extech 560 digital multimeter.
Onscreen oscilloscope display.
We’ve also enhanced its accuracy by installing 0.01 ohm shunt resistors across the three main voltages to monitor current accurately. This is discussed in detail on page 4 of SPCR’s PSU Test Platform V.3.
Front panel of DBS-2100
PSU load tester nestled into PSU thermal simulation box.
Note digital displays on both; they show current across the various voltage lines.
A rotary switch selects the current display between 12V, 5V and 3.3V lines on the DBS-2100.
As the resistors would have to dissipate up to 230W, the network was divided into two banks. Six resistors are clamped between a pair of heavy aluminum plates that act as heatsinks. There is enough space between the resistors to allow airflow between them.
For Version 4 of the PSU test rig, we’re adding 8 more 4.7 ohm, 50W
power resistors to obtain an additional 348W in four 82W increments. This is to accommodate the highest loads for the highest rated power supplies (in excess of 1,000W total). The photos and captions below explain better than words alone.
The finished resistor banks are at the bottom of the PSU simulation box, in the flow of air from the fans in the DBS-2100 load tester. There is an inch of space behind each bank so the airflow can pass through between the resistors. All wiring connections are soldered.
(NOTE: New 50W resistor banks not yet installed; to be done next week when resistor parts come in.)
Note 5-switch “front panel” with 2x12V and 4x12V connectors. Unfortunately, an error was made with the latter — it’s a “male” 4x12V plug that was laboriously soldered in place, but what’s required is a “female”, which I have not been able to locate since discovering this error. Ah well… at least it does no harm.
How the DBS-2100 and PSU thermal simulation box go together.
The end result is that the PSU can now be loaded up to ~23A on +12V1 using the DBS-2100 (with the main ATX cable and 4-pin Molex plugs), ~19A on +12V2 using this one of the additional resistor banks (via the 2x12V AUX connector), and ~30A on the 50W resistor banks (via EPS12V and PCIe x16 6-pin power connectors) for a total of about 72A, or 864W total on the 12V lines. The three resistive load banks are entirely independent, so the current and power to each can be reported separately.
COMPLETE TEST EQUIPMENT LIST, PART 2
Extech AC power analyzer, ATX PSU and 0~140VAC 3KVA variac.
The last item weighs 16 kg, by the way.
B&K model 2203 sound level meter
measures down to ~16 dBA.
Audio recording system: Modified Shuttle Zen PC running a P4-2.53 and suspended Samsung 40G 2.5" notebook hard drive. This PC is virtually inaudible and measures less than 17 dBA@1m SPL. A single channel M-Audio Tampa mic preamp with 96-kHz / 24-bit A/D converter and M-Audio Firewire 410 external digital sound interface feed the signal from the microphone.
ME 66 shotgun microphone has extremely low internal noise, which allows recordings to be made from a meter away even with very quiet sounds. The 1m recordings capture "nominal" volume, audibility, and sound character; the 30cm recordings capture all the details from even the quietest noise sources.
The dedicated heavy-duty 240VAC appliance
outlet and plug. A standard Canada/US 120VAC plug is dwarfed in comparison.
The dedicated 240VAC and 120VAC outlets are only for PSU testing; this circuit does not feed any other electrical devices. The Extech 380803 Power Analyzer / Data logger in the background has no trouble with 240VAC, which is run through the unit. The LED display shows that 244VAC is at the input. Extech MultiMaster 560 True RMS Multimeter and DBS-2100
PSU load tester can be seen in the foreground.
* * *
SPCR Articles of Related Interest
SPCR’s PSU Test Platform V.3 (Sept 2005)
Corrected Efficiency Results for Recommended PSUs (Oct 2005)
Audio Recording Methods Revised (July 2006)
SPCR’s Revised PSU Testing System (March 2004)
Power Supply Fundamentals and Recommendations
* * *
* * *
Version 4.1: BONEFISH EDITION
Dec 22, 2006 by Mike Chin
It’s less than three months since our last update to the power supply test
rig; what would require yet another revision? Oh, nothing that serious. Just
the need to increase the load test capability to 1,000 watts in preparation
for a monster PSU from Enermax that’s been awaiting our attention.
It was mentioned earlier that the 50W resistors had not arrived, and thus,
our rig was still not capable of reaching much higher than about 650W. Once
the resistors got here, it turned out that adding another >350W load capability
to the setup required a few more changes.
The cables and connectors on the test rig already get mighty hot at high loads.
It’s not wise to send any more power through the existing connectors. The 2x12V
connector is shown signs of heat damage, and the DBS 2100 is far to packed to
hack into. The natural step was to add 6-pin PCIe power connectors to the load
tester; this would help to distribute the load over more wires and connectors.
The big challenge was to find 6-pin PCIe power connectors. In Vancouver, neither
female nor male versions could be found in any of the usual electronic supply
stores I frequent. In the end, a DIY hack/mod was the answer, as it has been
so often in the SPCR lab.
The Antec NeoHE power supply happens to use a bank of 6-pin power connectors
as the output sockets for their modular cable system. We happened to have a
NeoHE that Devon inadvertently killed some time ago. Its carcass was stripped,
and the PCB with 6-pin connectors retrieved. Wiring this PCB and connector assembly
for a simple 2-conductor pathway proved to be a royal pain because it was originally
a 4-conductor circuit handling 12V, 5V and 3.3V outputs.
Once done, the banks of resistors were wired up to the 6-pin connectors and
installed within the PSU test box. The connectors / PCB assembly was mounted
on the side of the box.
Originally, the wiring and backside teminations were left exposed, but Devon
expressed some concern over potential damage due to shorting or the possibility
of testers burning or electrocuting themselves by touching the things accidentally.
An insulating cover was devised by folding up a promotional plastic card from
a restaurant and applying a bit of glue from a hot glue gun. That’s how this
PSU tester version became the Bonefish edition.
Recognize the PSU test rig?
The slogan on the card reads, "We get fish, you get fresh".
Here’s a closeup of the 6-pin connectors.
Getting back to some semblance of business, that little switch
on the side selects between one bank of three 3.3 ohm 50W resistors and two
banks. Originally, I started with four resistors in the sandwiched aluminum
plates, but after a quick test run, I felt just how hot these things became
and decided that just three resistors per sandwich was wiser. With 12V input,
the three resistors represent a load of 131W; 262W with both banks. After some
pondering about how to integrate this into the PSU box, I decided to hang the
two banks of heatsinked resistors directly behind the 120mm exhaust fan. This
would ensure decent cooling.
Hanging resistor banks.
The total load capability is now in excess of 900W. However, it
is still not enough. We need a kilowatt capability. Since testing at such high
power will be rare, I decided not to load up the PSU test box any further, and
make a detachable load bank instead. This one has four 3.3 ohm 50W resistors
and will pose a load of 175W with 12V input. It will be sandwiched between two
aluminum plates and a multifinned heatsink. It will rest atop the PSU
box, and when in use, connect in parallel to the hanging resistor banks via
one of the 6-pin connectors. Devon and the rest of the SPCR lab crew have been
warned not to touch.
The final assemly will be clamped together with bolts.
And that’s our PSU tester, Bonefish edition. The end result is
load capability of nearly 1.1 kilowatt. Hopefully, we won’t be testing such
loads often. No computer PSU could be quiet at that kind of output.
* * *