THG Power supply stess test (Redux)

renegade44e

Limp Gawd
Joined
Aug 10, 2005
Messages
132
Score 6 failures, 5 Passing

Now that it's been running a while what is the general concensus? I know our resident ps expert jonnyguru thinks they need to jack up the load on 12v1 and 12v2 rails at the expense of the 3.3c and 5v rails. I don't disagee but are they being too tough? I would also agree the temeprature readings are overdone and the other thing i would do is put the power supply in a PC case and not out in the open

Looks like the manufacturers are lining up to get their products tested since it started
 
http://www.tomshardware.com/howto/200507111/stresstest-03.html

We keep ten power supplies permanently on the verge of overloading. Complete system data, readings and pictures are available live from the THG Laboratory in Munich

limited applicability, and isnt "stress testing" in my book
sure thay loaded the piss out of the supplies, but its a static load
you stress a supply by going from a high load to next to nothing and ask it to not produce an overshoot, you vary its VAC, increase its ambient, you throw it curveballs and ask it to perform.

a basic crossload chart would have had more utility value
THG is great at sensationalizing a single aspect of a supply and blowing it out of all context
a disservice IMO

this time its a sustained high load, last time it was what..efficiency?
its like running a 10 day burnin test for a CPU to test for latent defects
and has little to no value other than a makes quality control, and quality of component selection, but its not asking the "design" of the supply to perform

http://www.smpstech.com/

its like the old car "torture tests" where they where flat out for days, but stop and go traffic is the reality and harder on a car than flat out

this is just my personal opinion, and has no "official" overtones
 
Ice Czar said:
THG is great at sensationalizing a single aspect of a supply and blowing it out of all context
a disservice IMO

Well put.

Look, I don't want to be any kind of benchmark of PSU tests. All I do is make sure that the PSU can pull off different loads, correctly proportioned to reflect somewhat realistic scenarios, at different intervals, even if they exceed label specs, and not blow up.

This test is useless.

There are so many variables.

First: You're not going to sustain a load for this long of a period. Yes, I admin that I sustain a load for half an hour before I jot down numbers, but I do this so the PSU can heat up. As the PSU heats up, efficiencies change and your numbers will change. I make it a rule to wait half an hour just so I don't get two different numbers from the same power supply being tested twice! And fact of the matter is, even a PSU manufacturer is only going to run a PSU for one hour at a given load before it determines that the load is "max." Just ONE HOUR. IF your rig REQUIRED this "max load capability" for ANY sustained period of time, it would only need it for... 2 seconds? 30 seconds? 1 minute? If more, you need a bigger power supply. End of story.

Also, how much fluctuation am I going to see when going from one load to another? Think about it... Playing a game, 12V is under stress from a pair of 7800's, CDROM spins up to load up the next level. Hard drives going too at this point. 12V jumps from 20A to 26A. 12V drops from 12.2V to 11.9. Not bad. 12V drops from 12.2 to 11.5? Not good.

Turth be told, I could make a test that SEEMS reasonable, but manages to blow up 100% of the power supplies I test.... but what good does that do anyone?
 
Ice Czar is right on, that test was flawed from the get go, like most of THG tests.
 
there is one other aspect of loading a supply right to and over the edge
it provides a point in a derating curve for a given ambient temperature
had they plotted a whole derating curve that would have been a "single aspect" worth sensationalizing
say from 21C up to 55C

the whole point of buying a supply however as jonnyGURU pointed out is not to overload it
and thats done through education
 
Ice Czar said:
http://www.tomshardware.com/howto/200507111/stresstest-03.html
limited applicability, and isnt "stress testing" in my book
sure thay loaded the piss out of the supplies, but its a static load
you stress a supply by going from a high load to next to nothing and ask it to not produce an overshoot, you vary its VAC, increase its ambient, you throw it curveballs and ask it to perform.

Good point! Can you design a test that will apply itself equally to a potential of up to 40 different power supplies? Please include a budget and a schedule with both hardware and manpower in both man-hours and number of personnel required. The "variable stress test" must be the same for all units taking in account units may vary from 400W to 900W and cannot be stressed above their rated capacity. Using percentages may be acceptable to accomplish this. In order to be fair the "variable stress test” must be computer controlled to insure the same stress conditions are applied equally to all units. Please include the man-hours to write this software in your budget
 
Ice Czar said:
there is one other aspect of loading a supply right to and over the edge
it provides a point in a derating curve for a given ambient temperature
had they plotted a whole derating curve that would have been a "single aspect" worth sensationalizing
say from 21C up to 55C

the whole point of buying a supply however as jonnyGURU pointed out is not to overload it
and thats done through education

what is the average temp that a psu should be running at anyway? i mean, spproximate exhaust temp maybe? is there such a thing?
 
well it totally depends

first there is the room ambient, not everyone has AC,
and we have several members that are in the tropics
but lets assume its 18C (65F) you then dump the heat generated by the computers components into that with a variable airflow from one persons case to another,
so there is a large variable right there as to both the airflow volume, the slower the hotter, and the total heat being produced, more or less drives, the CPU and GPU doing more or less work, and the PSU is "typically" exhausting all that from above the mobo compartment

Id "guess" an average of 35>40C (95>104F) with a worse case near 55C (130F)
see heat transfer is all about temperature differentials, if half as much differential exists half as much heat ends up transfered, and to compensate the interface (heatsink, radiator ect) needs to increase or the transfer mass (air water refrigerant) needs to increase inorder to transfer the same amount of heat
those three elements are at the core of all thermal solutions somewhere
even with phase change

while "most" supplies out there are "rated" per the Intel certification spec with an openbench 21C
so you can see as much as 1\3rd the rated watts and amps fall off a supply
some supplies employ a more realistic rating temperature, such as Seasonic, OCZ (Topower) and PCP&C, while others actually publish a derating curve

FSP's is
Derate 2W / C° from 50C° to 25C°
for the AX400-PN

or deduct 2 watts for every degree above 25C

I highly recommend reading
Ryan Shrout's General Heat Transfer Guide @ PC Perspective, specifically oriented to computer cooling ;)_
 
tdg said:
Ice Czar is right on, that test was flawed from the get go, like most of THG tests.

[H] roolz, toms sucks!!! Is that what you are really trying to say? Yeah right!

I appreciate comments that argue the merits of the issue rather than just pure boisterous bullshit

I do have a great deal of respect for Kyle cuz even though it says noobie I've been here a while and I've seen him eat a big ole dish of crow in public when he's been wrong even when it's not his fault but one of his writers screwing the pooch. I think the dear doctor on the other side of the pond has done so on occasion if memory serves me right.

In the 8 years I've been reading web reviews I've learned to take all of them with a certain grain of salt especially since 99% of them rely on the generosity of the industry in providing items for review which in many cases the reviewer is then allowed to keep.

Sometimes I think we need a site like resellers.com except for reviewers
 
ice czar, thank you much for that explanation and the link. i'll read it in the morning when i'm not so tired.
 
jonnyGURU said:
Well put.

Look, I don't want to be any kind of benchmark of PSU tests. All I do is make sure that the PSU can pull off different loads, correctly proportioned to reflect somewhat realistic scenarios, at different intervals, even if they exceed label specs, and not blow up.

This test is useless.

There are so many variables.

First: You're not going to sustain a load for this long of a period. Yes, I admin that I sustain a load for half an hour before I jot down numbers, but I do this so the PSU can heat up. As the PSU heats up, efficiencies change and your numbers will change. I make it a rule to wait half an hour just so I don't get two different numbers from the same power supply being tested twice! And fact of the matter is, even a PSU manufacturer is only going to run a PSU for one hour at a given load before it determines that the load is "max." Just ONE HOUR. IF your rig REQUIRED this "max load capability" for ANY sustained period of time, it would only need it for... 2 seconds? 30 seconds? 1 minute? If more, you need a bigger power supply. End of story.

Also, how much fluctuation am I going to see when going from one load to another? Think about it... Playing a game, 12V is under stress from a pair of 7800's, CDROM spins up to load up the next level. Hard drives going too at this point. 12V jumps from 20A to 26A. 12V drops from 12.2V to 11.9. Not bad. 12V drops from 12.2 to 11.5? Not good.

Turth be told, I could make a test that SEEMS reasonable, but manages to blow up 100% of the power supplies I test.... but what good does that do anyone?

This was posted on The THG Power supply test website. I pointed out in a previous post to Ice Czar the difficulties in making a test exactly as you are calling for.

If what is below isn't the right way to do it please tell us how you would do it.



Our tests in our Munich labs continue and we thought it would be a good idea to recap where we are today. So far, we have tested several units and 8 have failed at least one of our tests.

In our test-platform, we use four active electrical loads. The platform is custom made for us by Bernd Willer, which he calls the PC-NT-TB2. He is a former AMD engineer and now works as a consultant. The electrical loads are manufactured by the German company Statron.
We stress the single power rails of each PSU with one electrical load, +3.3, +5, +12 V1/V2. Each 12V rail is being loaded independently by a separate electrical load. Each electrical load is able to deliver a maximum current of 50Amps @75V. We load the rails never higher as stated on the max combined wattage sticker on the unit. The -12V and +5VSB rails are always loaded with constant 18W (based on ATX-12V-spec), whereas the load is increased an the other rails during the test. We start at a level of 20 percent of the maximum possible wattage and increase the load within 5h20min to 100 percent of the max combined wattage. You can see a picture of the standard-setup on this page
In addition we measure the temperature at the airflow input and output points and at the ATX connector. We also measure the temperature of the ambient air as well as measure the noise of the fan of the unit.
In addition we measure the ripple of the 3.3V and 12V rails during the test in accordance with the latest ATX-specs.

We say that a PSU has failed our test battery if any of the following happen:

If the ripple of 3.3V and 12V is not within the limits of the ATX-spec
If any of the voltages +3.3, +5, +12 V1/V2 delivered is not within the limits of the ATX-spec
If a unit switches off during the test before delivering its maximum combined wattage
If a unit switches off during the test delivering its maximum combined wattage anytime up to 19 hours

Stay tuned to this page for updates as we put the power supplies to the test.
Here is a list of units tested so far:
 
renegade44e said:
This was posted on The THG Power supply test website. I pointed out in a previous post to Ice Czar the difficulties in making a test exactly as you are calling for.

If what is below isn't the right way to do it please tell us how you would do it.

well
Id start with providing some baseline information
PSU Methodology Oleg is doing the best reviews out there at this time IMO
and is able to communicate his findings very quickly with graphical interfaces

such as the crossload
Offering a far more "balanced view of the supply as a whole, without sensationalizing a single aspect like Tom's does

then Id consult with
Jerrold Foutz our own gee and others about designing a "stress test" senerio

likely a computer controlled load regimen where under adverse conditions (read a high heat and a repeatable variable VAC) you have a highly variable load part of which would be a sustained peak load that drops to next to nothing and look for overshoots but other senerios as well that generally gives both the transient response, load and line regulation and the circuit design itself a workout, the number of things that can and do go wrong is underestimated SMPS arent "simple" devices



read these ;)


Underestimating the Complexity of Power Supply Design
Switching-mode power supplies are:

* non-linear circuits
* difficult to stabilize, often having real and complex poles in the right-half plane
* subject to chaos
* magnetically complex -- requiring consideration of proximity and eddy current effects and often requiring planar construction
* analog, digital, and power circuits, requiring knowledge of each discipline
* sources of electromagnetic interference
* etc.

They are not a trivial design task.


Power Supply Instability
Why switching-mode power supplies are so difficult to stabilize and how to keep them from oscillating in factory and field.

Low-Line Overstress
"However, if you are interested in field reliability, you should carefully check the turn-on / turn-off transient stress and add additional protection circuitry if needed."

MOSFET Runaway
"The on-resistance of a power MOSFET increases with temperature. With the resistance increase, the power dissipated for a constant on-current increases, the junction temperature increases, and the resistance further increases until the device is in thermal equilibrium with the heat removal system. If the heat removal system is inadequate, thermal run away occurs"

Output Filter Overshoot
Some overcurrent protection circuits limit the maximum current that can flow in the output filter inductor. Others do not. For overshoot control purposes, the former are preferred.

Chaos
Chaos can exist in a wide variety of power electronics circuits including:

* PWM regulators,
* ripple regulators,
* current-mode controllers,
* overcurrent protection circuits,
* series resonant circuits,
* rectifier-filter circuits,
* ferro-resonant circuits,
* snubbers with nonlinear inductors,
* magnetic amplifiers,
* circuits with a combination high-gain and low-gain loop, etc.

Input Filter Interaction
Adding an input filter to a switching-mode power supply or installing it in a system with an EMI filter can degrade its performance or cause it to go unstable. Long input cables or added output capacitance may cause the same problems.

just a few of the articles at SMPS Technology Database
most PSUs, that arent overtaxed, in a reasonable environment are likely adequate for the task at hand, but the objective of a review, is to find out to what degree the differences are between power supplies. And further to be unreasonable, because as they say, there is no such thing as idiot proof :p
 
Back
Top