View Single Post
  #1 (permalink)  
Old October 25, 2007, 08:59 AM
SKYMTL's Avatar
SKYMTL SKYMTL is offline
HardwareCanuck Review Editor
Join Date: Feb 2007
Location: Montreal
Posts: 13,622
Default Hardware Canucks Power Supply Testing Methodology

Power Supply Testing Methodology


Power supplies are one of the most difficult computer components to test and because of this we have been very cautious to go about this properly. We have chosen to take a kind of middle road between the (few) websites that use load testers and those that use very basic setups to test power supplies. Here are the two categories of review websites and our take on their content.

#1: Websites with a load tester

There are a few websites on the internet that have the art of reviewing power supplies down to perfection. Sites like, PcPerspective, HardOCP and a few others have blazed a trail by showing what a power supply should look like. They use highly accurate (and expensive) test equipment to push power supplies to their peak and then some with the capability of putting a 100% load on an unfortunate test unit. With the capability to test ripple, efficiency and power factor, these sites go above and beyond the call of duty in their reviews and have set a new standard.

Unfortunately, there is a small downfall to their reviews. While they list how much load is being put on the power supply, a consumer has no way (other than the few VERY inaccurate power calculators on the internet) how much load their actual system will be putting on a power supply. So, while a load number is great, it doesn't tell a consumer what he or she really needs to know: what can the tester power supply actually run safely? Finally, there is the matter of the extreme cost involved with getting the proper equipment and the training involved with learning how to properly use it. We didn't want to take a "me too"; approach so we opted out of this type of review.

#2 Websites with a system...and not much else

We have seen all sorts of hodge-podge reviews out there and they all pretty much run the gamut beginning at mediocrity and ending at downright pointless. Some use multimeters which is a small point in their camps but others use completely inaccurate software measuring programs which warranty a quick toss into the garbage. To make matters even worse, we have seen time and again review sites testing 600W or higher power supplies with single core, single GPU systems and suddenly popping in an "Editor's Choice"; rating for no apparent reason. To us, reviews like these are like reviewing a brand new graphics card without including game benchmarks; it just shouldn't be done. Sites like Hardwaresecrets have realized that reviewing power supplies is not easy and instead of taking the easy way out, they have shown these other review sites the way; they analyze the review units from the inside out without testing them and not dolling out any needless awards. But, these other review sites should either spruce up their reviews or stop reviewing power supplies altogether because they are doing a grave injustice to their readers. There, I said it.

Now, the middle ground

After weeks and weeks of discussion, we decided to float nicely in the middle ground. What we are doing is testing all of the power supplies we get on systems that a regular consumer would try to power. In this case everyone who reads our PSU reviews can see how certain power supplies scale in relation to actual systems and applications. In addition, we incorporate efficiency tests, ripple readings and highly accurate voltage readings as well. It is for this reason that you will never see a 1000W power supply hooked up to a single GPU system here at Hardware Canucks. We test with systems that put increasing load on our review units in order for you the reader to better understand what a certain PSU is capable of and where it starts falling apart.

Thus, we have devised two systems to test certain ranges of power supplies:

500W-650W PSU Test System

Processor: Intel Core 2 Duo E8400 @ 3.6Ghz
Memory: 4GB G.Skill DDR2 PC2-1000
Motherboard: DFI Lanparty X38 Dark
Graphics Card: ASUS 9800GX2 TOP
Disk Drive: Pioneer DVD Writer
Hard Drive: Hitachi Deskstar 500GB
Fans: 2X Yate Loon 120mm
Monitor: Samsung 305T

Al in all, we find that this setup is able to put the best possible stress and anyone who would have higher specs than this would be nuts to buy a 700W or below PSU. Here we shy away from overclocked quad core processors and high-powered SLI setups since we feel that these things are best left to higher-end power supplies to handle.

700W-900W PSU Test System

Processor: Intel Core 2 Quad Q6600 @ 3.5Ghz (B3)
Memory: 4GB Corsair Dominator DDR3 @ 1446Mhz (Thanks to Corsair)
Motherboard: Asus Blitz Extreme
Graphics Cards: 2X Gigabyte HD2900XT 512MB (Thanks to Gigabyte)
Disk Drive: Pioneer DVD Writer
Hard Drive: Seagate Barracuda 320GB SATAII
Fans: 5X Yate Loon 120mm @ 1200RPM
Monitor: Samsung 305T

So this is the system we will use to test the higher wattage power supplies. After overclocking the GPUs in conjunction with the processor, you will be shocked when you see some of the power consumption numbers we are able to pull from the setup. The B3 quad core was chosen due to its increased power consumption over the newer G0 stepping as well as the motherboard's P35 chipset being more of a power hog than the revamped X38/X48. This system was build for the sole purpose of stressing some the best power supplies out there.

950W+ PSU Test System

Processor: Intel Core 2 Quad Q6600 @ 3.5Ghz (B3)
Processor #2: Intel Core 2 Duo E8400 @ 3.6hz
Memory: 4GB Corsair Dominator DDR3 @ 1446Mhz (Thanks to Corsair)
Motherboard: Asus Blitz Extreme
Graphics Cards #1 & #2: 2X Gigabyte HD2900XT 512MB @ 852Mhz / 1768Mhz
Graphics Cards #3 ASUS 9800GX2 @ 758Mhz / 2210Mhz
Disk Drive: Pioneer DVD Writer
Hard Drive #1: Seagate Barracuda 320GB SATAII
Hard Drive #2: Hitachi Deskstar 500GB SATAII
Fans: 5X Yate Loon 120mm @ 1200RPM
Monitor: Samsung 305T

While this test system may not seem realistic, since you are now able to run four graphics cards together on both ATI and Nvidia systems the amount of load the graphics cards seen here can generate is well within the realm of possibilities. The same goes for the CPU. We thought it prodent to add the second CPU from our 500W to 650W test platform since on air cooling our Q6600 can only make it to 3.5Ghz while maintaining the stability we need for the tests. That being said, there are people out there who will use more extreme cooling methods which will push power consumption even further. Thus, the minimal increase in power consumption (when compared to a Q6600) the E8400 produces suits its purpose quite well in boosting CPU power consumption.

With these three systems we hope to show you how component choice should influence your choice of power supplies. We are aiming to use actual computer systems to load the review units in order to show the readers how different power supplies behave when running certain setups. We are not out to push these PSUs until they fail. Rather, our goal is to show you which PSU would be best for your system.

So, what do we use to test these power supplies? Read on...

Load conditions

In these sections we will be detailing how we load the tested power supply in order to determine how it performs in certain conditions. We try to gradually increase load with realistic tests which a consumer may experience with their unit. Usually, the last test is reserved for a "worst case scenario" which is used to stress the system far beyond what a normal user would experience.

It has only been after extensive testing of dozens of programs that we came up with programs / combinations of programs that put the MOST stress on our components. It is only the programs which consumed the most amount of power that we are using for our load conditions.

*All load tests are run over the course of one hour. All overclock tests are run over the course of 30 minutes.

400W-650W Load Conditions


750W-1200W Load Conditions

*Note that all of these tests are done with dual HD2900XT cards in Crossfire

Off: Is a load value where the system is turned off but a small amount of power is still required.

Idle: Idle values are determined by a stable Windows Vista x64 desktop.

Here we are going with a little bit different approach to the load values. With these tests, we will load the system in different ways to show how the power supply reacts depending on what kind of program you are running.

CPU Load: This test is run with 4 instances of a custom Prime95 test which we have found uses the most non-GPU power. The test is run for 30 minutes.

GPU Load: For this test we load our Crossfire setup with the Half Life 2: Lost Coast GPU Stress Test using the highest possible in-game settings at 1920x1440 resolution. We have chosen this game because our of all of the games tested, it is the one which was able to put our Crossfire setup under the most load. The benchmark is run 10 times.

Full System Stress: This is the big one. For this test we use the Crossfire setup running 3DMark06 Batch Render Test at 2560x1600 2xAA 16xAF while running our custom Prime95 test in the background.

Extreme Load Test (950W+ PSUs only): In this test we run the Full System Stress Test plus load the E8400 with and OCCT Custom Mix Test, overclock the graphics cards and additionally run the 3DMark Batch Render Test at 1280x1024 with 4xAA (at the highest quality setting) and 16xAF.

Voltage Regulation Testing Methodology

Voltage regulation is a very important aspect of every power supply in the sense that it can affect the stability and safety of your system. If the voltages dip too low, you can experience everything from crashes to unstable overclocks. The ATX specifications dictate that the rails of a power supply should be regulated as follows:

+12V Rail(s)

Normal: 12.00V
Min: 11.40V
Max: 12.60V

+3.3V Rail

Normal: 3.30V
Min: 3.14V
Max: 3.47V

+5V Rail

Normal: 5.00V
Min: 4.75V
Max: 5.25V

Testing voltage regulation is a must in any power supply review. Testing with software is inaccurate at best so we have chosen to go with an industrial-grade multimeter which is calibrated every 3 months by an ISO-certified local electronics distributor. This one calibrated multimeter is used as a benchmark for all of the other multimeter in our lab; they all have to read within 0.02V of the calibrated multimeter or they are discarded and replaced by another.

Multimeters Used:
Fluke 187 (master)
Extech 570 x2
Extech 110 (smaller probes used for tight-fit areas)

*Note: All voltage readings indicated in the review are the minimum voltages seen over the period of our tests

Most review sites take their voltages from an unused Molex connector but we do not. Rather, we always take voltage readings from a loaded connector in order to more accurately see the voltage fluctuations our components are experiencing. Thus, this is how voltages are measured:

+12V: In the CPU Load test the voltages are taken directly from the CPU connector of the power supply. In the GPU Load test the voltages are taken from a PCI-E connector which is plugged in to one of the graphics cards. All voltages are the lowest recorded over the test period.

+3.3V / +5V: From the main ATX connector. All voltages are the lowest recorded over the test period.

AC Ripple Testing Methodology

Here we begin delving into things which are a little more complicated than are seen in your "normal" power supply review. AC Ripple (or noise as it is also called) is one of the most under-measured aspects of a power supply's performance that we have come across. Among the few sites we have come upon that actually measure it, only a bare minimum measure it with any accuracy. Yet, this one measurement is as important as voltage regulation.

A high amount of ripple coming from the power supply's rails can be detrimental to the stability and longevity of your system. While many motherboards, graphics cards and other components may not see any immediate effects of excess ripple, it can contribute to wear down your components' lifespan significantly. Some may wonder why their ram suddenly stops working while others may scratch their heads at why a previously stable overclock has become unstable after a few months. Some of these cases can be traced back to excess ripple. So, these are the ATX specifications regarding ripple:

+12V : 120mV Max
+3.3V: 50mV Max
+5V: 50mV Max

We believe that the ATX specifications for the +12V rail are set far too high and thus will consider anything over 100mV a failure. All ripple measurements given are the highest recorded over the test period. The values were the highest peak ripple measurement across all of the +12V rails. So, if the +12V1 rail shows a ripple of 20mV and the +12V2 rail shows a ripple of 40mV, the highest value will be graphed.

Instruments Used:
USB Instruments Stingray Digital Oscilloscope
USB Instruments Differential Oscilloscope Probe

Here is where we have spent the majority of the money when it comes to testing equipment and time learning how to use it. Basically, the Stingray is plugged into a laptop and is used to determine AC ripple (among other capabilities) emanating for the power supply's rails. Since we do not have a load tester with a BNC connector for the standard o-scope probe, we needed a Differential probe in order to give us the proper capacitance to accurately determine ripple. In addition, the differential probe has a pair of connectors which are very much akin to a multimeter's probes which makes them idea for use on SMPS designs.

Efficiency Testing Methodology

In the last year or two, efficiency has become more and more important in the realm of power supplies as consumers have become more attuned to the environmental impact of consuming copious amounts of electricity. Not only has the cost of electricity gone up dramatically but the power consumption of modern computers is hitting all-time highs as well. Lower efficiency means more power being wasted as heat inside of a power supply and because of this the overall heat signature of a computer will increase as efficiency decreases. In addition, because of this increased heat, the fan on the power supply has to ramp up as well in order to keep the sensitive components cool. It is all a vicious circle with only one thing being painfully obvious: the higher the efficiency the better.

The few websites that measure efficiency are able to give you a percentage (ie: 80% efficiency) based on the difference between actual DC load (the power consumption of your components) and AC power consumption (the amount of electricity being drawn from the wall). Unfortunately, we have no accurate way of knowing how much load our components actually consume so there is no way of coming up with an accurate percentage figure. So, we have much more realistic goals: to compare the power supply being reviewed to the best we have tested in the past in the same category in terms of AC power consumption.

Instruments Used:
UPM Power Meter
Tripp Lite LC1800 Line Conditioner

What the UPM power meter does is give us how much power the entire system is drawing from the wall at any given point. The data points you see in our charts show the AVERAGE PEAK AC power consumption over all of the tests conducted.

A few other tidbits

- AC Input Voltage: 120V constant
- All tests done in a closed-case environment
- Noise and heat are subjectively tested


So this wraps up our whirlwind tour of the Hardware Canucks power supply testing methodology. In the end, we want these reviews to appeal to everyone from seasoned power supply reviewers (you know who you are) to beginners building their first system. Thus, we include all of the necessary tests while trying to keep the complicated things like o-scope screenshots to a minimum. We hope to strike a good balance which everyone will be happy with and we hope that you end up enjoying reading the reviews as much as we have by writing them. There will of course be revisions to this testing methodology but as it stands now, we are convinced that we will be able to give you some of the most accurate and beginner-friendly power supply reviews around.

Last edited by SKYMTL; May 11, 2008 at 08:08 PM.
Reply With Quote