AnandTech Power Supply Test Methodology
by Christoph Katzer on July 12, 2007 12:00 AM EST- Posted in
- Cases/Cooling/PSUs
Why It is so Difficult to Test Power Supplies
Over the past couple of years few markets in the IT industry have grown as rapidly as the power supply sector. Faster and more powerful CPUs, GPUs, and other components have led to a need for better and more powerful PSUs. A few years ago, most would have laughed at the thought of a desktop PC with a 700W or greater PSU, and yet such designs are becoming increasingly common. Companies are sprouting out of the ground like mushrooms and everybody wants a share of this fast evolving market.
With growing interest in power supplies it is natural that websites will take an interest in reviewing and testing them. However, power supplies are not as easy to test as many other components. For example, unlike a graphics card, CPU, or motherboard it is not enough to simply plug it in and run several of the latest 3D shooters.
Unfortunately, most published reviews on the internet are done in this manner. The DC outputs are quickly measured separately with a multimeter - or even worse the motherboard's BIOS. The rails seem to be stable, it runs the test system without difficulty, and it is relatively quiet; good enough for a Gold Award! These kinds of reviews do more harm than good to the actual users.
Because of the simple test rigs that are employed, the power supplies are not even close to being fully loaded and thus the DC outputs are generally without problem. It is also difficult or impossible to find flaws in this manner which would otherwise prevent people from buying a dangerous product. A certain PSU might run in system A just fine only to break down from an overload in system B and cause damage to the whole system. In addition to this type of problem, cheaper power supplies lack security features which should have been there for a reason.
The bottom line is, a normal PC is just not enough to fully load a high-end power supply, and therefore it cannot be properly tested and surely not approved by such a configuration.
Over the past couple of years few markets in the IT industry have grown as rapidly as the power supply sector. Faster and more powerful CPUs, GPUs, and other components have led to a need for better and more powerful PSUs. A few years ago, most would have laughed at the thought of a desktop PC with a 700W or greater PSU, and yet such designs are becoming increasingly common. Companies are sprouting out of the ground like mushrooms and everybody wants a share of this fast evolving market.
With growing interest in power supplies it is natural that websites will take an interest in reviewing and testing them. However, power supplies are not as easy to test as many other components. For example, unlike a graphics card, CPU, or motherboard it is not enough to simply plug it in and run several of the latest 3D shooters.
Unfortunately, most published reviews on the internet are done in this manner. The DC outputs are quickly measured separately with a multimeter - or even worse the motherboard's BIOS. The rails seem to be stable, it runs the test system without difficulty, and it is relatively quiet; good enough for a Gold Award! These kinds of reviews do more harm than good to the actual users.
Because of the simple test rigs that are employed, the power supplies are not even close to being fully loaded and thus the DC outputs are generally without problem. It is also difficult or impossible to find flaws in this manner which would otherwise prevent people from buying a dangerous product. A certain PSU might run in system A just fine only to break down from an overload in system B and cause damage to the whole system. In addition to this type of problem, cheaper power supplies lack security features which should have been there for a reason.
The bottom line is, a normal PC is just not enough to fully load a high-end power supply, and therefore it cannot be properly tested and surely not approved by such a configuration.
Facebook
Twitter
RSS
49 Comments
View All Comments
tynopik - Tuesday, July 17, 2007 - link
please please please add brownout/sag/powerdip/flicker tests!this is a big killer of power supplies and personally caused one of mine to blow up and take out everything connected to it (mb, 3 hd, cd, etc)
test how it maintains power when the voltage drops down to 90v
test how it responds to a series of flickers (this is what killed mine)
gold standards in PSU review
jonnyguru.com
silentpcreview.com
mcvan - Friday, July 13, 2007 - link
This looks like an ambitious effort with lots of good test gear, but the thermal and acoustic aspects of the testing setup are seriously flawed.1) A 1 meter cube box is a *horrific* acoustic test environment. It's way too small to avoid boundary effects. All the acoustic stuffing in the world will not help. You can see in the photo that the mic has to be at one corner and the PSU in the other corner to have 1m distance. Ever stand in the corner to sing? Notice what happens to your voice?
The box will impose its very significant acoustic signature on everything they measure or record in there. The most important effect will be boosts in all the freq below ~500Hz, along with peaks in certain bass frequencies. Any fan used in any PSU will excite such resonances.
Also, there's no indication of what ambient noise level you're getting in there with the Chroma testers going. You mentioned 15 dBA -- I say no way, not with the Chroma machines going. I actually bought a similar AC source device a few years ago and had to abandon it due to the noise problem -- it easily reached 60 dBA @1m. For their test box to achieve 15 dBA ambient with the Chroma machines going in the same room, the box would have to have something like >35 dBA noise isolation and/or the room would have to be gigantic (so that the Chroma machines could be far away.... which then brings up the problem of cable resistance). The former requires astonishing acoustic technology and knowhow; judging by the box construction details, I question whether you can get anywhere near that kind of noise isolation, and if the test room was that big, you'd have said so.
2) The PSU will be in the 1m test box more or less by itself. The box will be sealed during testing. This is unrealistic for 2 reasons --
A) a PSU in a real PC always see some fresh air intake, and
B) it also sees at least some of the heat produced by the components it is feeding.
To expand on this further, the DC output of the PSU goes to components that heat up. These components are in the same case as the PSU, and they affect the ambient temperature of the PSU. It's a simple fact that the total heat generated in the computer is equal to its AC power consumption. Case fans obviously affect the overall thermal situation for the PSU, but the PSU's cooling fan draws air from inside the case. Unless the PSU is in a separate chamber with its own intake vent, it has to deal with not only the heat generated within itself as a result of AC/DC conversion losses, but also at least some of the heat from the components it powers. Heat can affect a PSU's max power output capacity, its stability, and of course, its fan speed/noise.
It's because Anandtech is trying to do "serious" acoustic testing that all this becomes important. In http://www.silentpcreview.com/article683-page1.htm...">SPCR's PSU testing, we have a very clear operational model: The test rig is a close facsimile of a quiet enthusiast's low airflow, low airflow-impedance system. Your test setup doesn't resemble any real system -- there is no fresh air intake, and the exhaust heat goes nowhere. Can such a test setup tell anything practical and worthwhile about the noise/load or the heat/load behavior of any PSU? Certainly not if you intend to install that PSU in a case where there are case fans, intake vents, and other components that also get hot.
I also see another serious thermal issue: It seems to me that in your setup, once a certain amount of heat has been generated inside, continued testing can only lead to continually rising temperature for any PSU. In other words, the temperature/load relationship will be dictated almost entirely by how LONG the PSU is kept running. The heat is trapped in the box; it has nowhere to go, and the box is well insulated. Even at a modest 100W load, leave it running in that test box for a few hours, and I'd bet $$ almost any PSU will shut down due to overheating. So where in the rising curve (over time) could you say is the right point to look at temperature or fan speed or noise? This is a challenging problem.
This is completely different from SPCR's test rig. We generally try to do all the PSU testing at the same room temperature (20~22C), which we can maintain most of the year -- we don't do PSU testing on hot July and August days. This does not stop the internal temperature of the test rig from reaching upwards of 50C at high power load. Importantly, at almost any load level with almost any PSU, the system does come to an equilibrium where temperature (in and out), fan speed and noise stabilize. This is then the obvious point to stop and take measurements.
Furthermore, in the SPCR test rig, almost any well-built PSU can be run at >75% of rated power indefinitely without overheating, which I think cannot be true of your current setup in that sealed box. (I wanted to say 100%, but I suspect this is not true -- even some of the well-built PSUs may overheat at full load if left at full load long enough.) All this is to say that our test rig has predictable thermal characteristics that do closely resemble a real PC system, which is very important if you want to do accurate acoustic testing / analysis. It's also very important for thermal analysis.
Christoph Katzer - Monday, July 16, 2007 - link
Hey Mike,You can be sure your concern will be taken into account. First of all the box isn't just 1m. It just looks like this on the pictures but an extension is already planned since we also want to fit chassis inside. The PSU and Mic have enough room to the sides and yes, we separated the Chroma and the PSU that much from each other that we can test just fine. As said currently we are still working on the audio part but the box like it is brought us far until now.
We will have both room-temp and 50°C from the next review on since we think it is important to show what a PSU can handle. We don't know where the PSU is running later thus we present as many results possible that our readers have enough information to form an own opinion.
Christoph Katzer - Friday, July 13, 2007 - link
Thanks for all your comments and suggestions. You can be sure that it will all be taken into account and we are working on it right now.giantpandaman2 - Friday, July 13, 2007 - link
Could you test what happens to the power supply when there's a sudden drop in voltage and then a surge back to normal? I figure that'd emulate someone turning on, say, a heater on the same circuit?Now, I know a lot of people will say only an idiot would ever plug a computer with other appliance that have high power draw...well...sometimes we have to build computers for people who are a bit, how should we say, technically challenged. WOuld be nice to know if there are any foolproof PSU's for those situations.
langenbacher - Thursday, July 12, 2007 - link
The way in which a power supply acts at power-up and power-down can be very important. These guys publish specifications for PS's that include timing specifications for this.http://ssiforum.org">http://ssiforum.org http://ssiforum.org/specifications.aspx#powerSuppl...">http://ssiforum.org/specifications.aspx#powerSuppl...
http://www.formfactors.org">http://www.formfactors.org http://www.formfactors.org/developer%5Cspecs%5CPSU...">http://www.formfactors.org/developer%5Cspecs%5CPSU...
I need power supplies that "boot up" and shut down cleanly, in order to test my company's products to see that they meet spec for their own behavior at power-up and power-down time. Some power supplies that have been highly rated in the press have ugly ramp-up voltage waveforms that make it impossible for me to make measurements on our PCI-X and PCI-E boards.
I have seen how bad things can happen on a PCI bus if power-up doesn't go right, most commonly failure for devices to properly initialize and enumerate on the bus.
In the case of the "bad" power supplies I got, it was easy to see with an oscilloscope that the voltage did not ramp up cleanly, and it wasn't even necessary to know the official specs to see that something was wrong. I think that simply capturing a trace showing all the voltages ramp up and ramp down should be able to weed out the worst offending power supplies.
erikpurne - Thursday, July 12, 2007 - link
I'm no expert on these things, but I can see that PSUs would be a very difficult component to test in a meaningful way, and though there are a few things your setup appears to overlook, you've done an awesome job of putting together a rig that can do it as well as can reasonably be expected (once you get the ripple measurements figured out). Must have cost a bundle, but there's little doubt that this will make AT the world's #1 source for PSU info. And that anechoic chamber/box/whatever will come in handy for getting accurate SPL measurements from a whole bunch of other stuff, too! I just hope a PSU doesn't decide to explode or catch fire inside your nifty box and ruin it...Also, I assume you'll only be using the box for relatively quick sound measurements and not for the actual load testing. If not, how will you deal with the rapidly and indefinitely - till something fails - rising temps inside the extremely well insulated box? I mean, hooray for stress testing, but wouldn't that get a little extreme?
Either way, props on doing it right or not at all.
erikejw - Thursday, July 12, 2007 - link
"To measure the produced noise we have build an anechoic room in which we can measure the noise down to just 15 dB(A)."Please use this when you test fans too, it would rock.
Looking forward to this series of tests too.
Conky - Thursday, July 12, 2007 - link
This is long overdue. No more relying on lesser sites for PSU info. :)markitect - Thursday, July 12, 2007 - link
You should add a few additional temps to determine the robustness of the power supply.First:
Check stable outputs during different levels of short and long term power brown outs and surges. (handles 1 sec at 250V things like that.
Second:
check stability as temp gets out of control.
Raise the ambient temp to see if the output changes as the temp increases. Power supplies tend to suck air from the case so it could see cool-side temps up to 50C