Article on PSU size & overkill; irresponsible journalism

[spookmineer]

jonnyGURU, your post was well balanced, credible and polite.

I'm glad that for some part you are "with us" and the PSU is getting kind of crazy these days.
The manufacturer's PR machine is running overtime now they see that uneducated people are better safe then sorry, and they can sell 1000W PSU's to people who don't really need them.

From all the comments on forums, myself included, people have a standard tendency to build in some headroom. If it starts with a reviewer, saying "this PSU will be enough to power a such and so", people still add a little something just to be safe. Even if the reviewer's statement was safe to begin with, and already calculated in some headroom. So we get headroom on headroom (see last paragraph of my post please).

I'm fairly sure, people will recognise that mcoleg's setup is on the edge, but it does wake them up (that was his only intention I think) and that is an eye opener for some.
I don't think for one minute, that someone with the same setup as his, will buy the same PSU.

I've got a PSU which is clearly overkill for my setup (and next build)... If only I had read some threads which were more clear on this. I read tons of threads, reviews and it's hard to get the right PSU if you're a beginner.
For me, it only matters 20, maybe 30 bucks, but people buying a 1200W PSU when they could easily buy a quality 650W are just wasting money...
[on the other hand, I have to say it will hardly ever ramp up the fan while getting somewhat loaded - still the same it's overkill ]

It's nice to know that some people are willing to spend some time and money to educate others. After regularly visiting a lot of hardware forums, I now know that there is a lot of BS going on. Especially on overclocker forums, 16 year old people with 1000+ posts are coming accross more knowing then older people with 10+ posts (some 16 year olds are smart though), influencing other people into the "wrong" direction.
I even read someone buying a $20 600W PSU, because 600W is enough... My reply got ignored I guess.

Keep up the good work, I'm sure many people reward it (like I did when I went to look for another PSU) - it's tricky business though, it's hard to recommend someone a certain PSU if there is even the slightest chance it may fail in their specific case.

[jessekopelman]

Since the lifespan of a PSU is directly correlated to its running temperature

I don't know that this is strictly true. Do you think a PSU with an operating temperature of 20C would last longer than one with 30C? Meanwhile, I'm sure one with a operating temperature of 200C would do no worse than 150C (ie they'd both fail instantly). What seems to be pretty well established is that there is a range of safe temperatures at which to run, but it is unclear whether differences within that range will have any effect and if so what it will be. Look at the recent Google study that showed HDD running at 30-35C had a lower failure rate than those running below 30C.

[jessekopelman]

actually, we can interpolate some of the things we need to know about a comparative lifespan - as that experiment i proposed with two psu's of the same series but with different power rating.

i already got me a thermocouple with pc interface; now i need a usb adapter for it and the whole thing should be ready for testing.

Explain your experiment in greater detail. Quickly looking back through this thread, all I could find is that you are going to do some load testing on two PSU and I guess measure some temperatures. What is the the hypothesis and how will you test it?

[gmat]

I don't know that this is strictly true. Do you think a PSU with an operating temperature of 20C would last longer than one with 30C? Meanwhile, I'm sure one with a operating temperature of 200C would do no worse than 150C (ie they'd both fail instantly). (...)

I get your point, still the temperatures that concern us here are more centered around 50°C.. Most of the MTBF diagrams for electronic components i've seen are based on temperature charts, and not directly power load or current. (i say most, as some components like relays etc. have other mechanical considerations). The military spec components have larger temperature ranges (usually around -30°C to +120°C) but again, there's a higher life expectancy at cold temperatures. The most obvious is capacitor evaporation (which i guess is the main point of failure in most cases).

Just remember though that a 350W power load produces more heat than a 150W one, even if the 350W one is at 82% efficiency (76W of heat) and the 150W one is at 70% efficiency (64W of heat). You have to consider the actual heat loss here, so yeah a substantially higher load on a PSU will reduce its life expectancy as more heat will be drawn. Still, that's inverse reasoning: one should adapt the right sized PSU to the power needs of the system ! And of course a more efficient PSU will produce less heat, and live longer than a less efficient one.
That's exactly why 1000W PSU's are a very bad idea, as they will certainly be far from their efficiency sweet spot (which may be around 800W) on all single systems, and dump more heat, and use more electricity than correctly sized ones. Of course you could use a 1000W to power several systems at once but that would mean servers, and the failure mode of several servers connected to a single PSU is just frightening (the server mode of operation call for multiple redundant PSU's in one system, not the opposite).
I may be repeating very obvious things here, which have been hammered through SPCR articles, still i felt it was needed.
What the power supply manufacturers should do today, is target the 300-500W segment with very high quality components, higher efficiency, maybe passive cooling (allowed by better efficiency), better regulation under variable loads, and, please, all modular cables. Corsair were the first with their 520W unit, i expect Seasonic to follow suit (as the recent interview from Taipei lets us think).

[Mr Evil]

I don't know that this is strictly true. Do you think a PSU with an operating temperature of 20C would last longer than one with 30C?..

Yes. The lifespan of electronic equipment is directly related to operating temperature.

...Meanwhile, I'm sure one with a operating temperature of 200C would do no worse than 150C (ie they'd both fail instantly)...

Yes, there's only a range of temperature where the above relationship holds true. Go above that and thigns start to melt or evaporate, go below that and it may also cause problems, through freezing or mechanical stress (semiconductors tend to be able to go well below 0C though).

... Look at the recent Google study that showed HDD running at 30-35C had a lower failure rate than those running below 30C.

Mechanical components are probably the culprit there - expansion/contraction of bearings, viscosity of lubricants etc, which is why the fan is often excluded in PSU MTBF calculations.

[Zinj]

Just a quick comment on MTBF.


C'DaleRider said,

"So, you have an average life span.......MTBF. This, of course, means that half the products produced will fail in less time than stated and half will fail in a time longer than stated."

and Jessekopelman responded,

"What C'DaleRider is talking about is the 68-95-99.7 rule. What that rule tells us is that for a normal distribution (any random sample can be assumed to be normally distributed if it is sufficiently large) 68% of the samples will be within a single standard deviation of the mean."


I think C'DaleRider was talking about means and medians, not means and averages (or std. deviations). Mean and "average" are the same thing: add up the observed values, divide by the number of cases, and that's the mean (or average). Medians are another measure of central tendency. In the case of the median, half of the observations fall above the median, half below. Where the data are symetrically distributed, the mean and median will be the same (for example, the normal distribution).

Problem is, many distributions aren't symmetric. For example, the MTBF is usually calculated based on an exponential distribution. The exponential is right skewed, so the median is actually about 70% of the mean. In a perfect world, if the average time to failure were 100,000 hours, half of the units would fail by 70,000 hours, and half after. To make matters worse, slightly different distributions can have similar mean times to failure even though though the point at which half of the units fail actually varies widely (http://www.reliasoft.com/newsletter/2Q2000/mttf.htm). Manufacturers also don't all calculate MTBF to the same standard.

A rough guide indeed!

Z-

[mcoleg]

sry for not replying earlier, guys.

jessekopelman - i don't think the length of the warranty directly correlated to the lifespan. it's more like - what they can get away with

gmat - "Since the lifespan of a PSU is directly correlated to its running temperature, it IS somewhat related to its power draw." it's only technically so. in reality, only solid-state capacitors are affected by current (assuming the load does not exceed nominal values; if it does, practically any capacitor goes boom anyway ) . electrolytic capacitors simply evaporate over time.

spookmineer - thanks for the nice words, not only to myself but to jonny. i do respect the man and appreciate his expertise, you know.

jessekopelman - "PSU with an operating temperature of 20C would last longer than one with 30C" - once again, technically, yes. the difference in life expectancy at those temperatures would be neglect able for our purposes, however. since most well-made consumer-grade psu's are rated at 40C, that's the temperature we should be looking at. above that and we have to apply a de-rating curve; bellow that we can't calculate so we assume it's equal in performance to operations at 40C (it's not but for our calculations it might just as well be).

gmat - i tend to look at the efficiency as a curve, not as a point. above 80% is what we should be aiming for, really - the "Goldilocks" zone. 80+ rated psu's promise 80+% efficiency from 20 to 80 % of the load. as long as the actual loads for the pc fall into that zone, we should be in the clear.

Mr Evil - you got to explaining all that before me, again

Zinj - that's why i am not so focused on MTBF any longer. without life expectancy numbers it's not exactly something we can freely operate with.

jessekopelman - there's no "greater" details, really. i want to keep it as simple as possible - which one of the psu's will produce more heat will die first. if the temperatures are close to each other, the life expectancy difference is insignificant, if it's far apart than we can make some guesses.

if you got any suggestions though, i'd like to hear them.

[gmat]

mcoleg, exactly, although current generation of '80+' rated PSU's still show that "sweet spot" behavior, and still fall down on each side of this sweet spot - the efficiency curve still looks the same. So taking a 80+ rated PSU just to use it at 20% of its rated power load, is maybe not the best option...
It would help a lot if component mfgers would publish power draw figures in their documentation, so when assembling a system we could add up the power draw from each component and then (and only then) pick a PSU that's appropriate. And by "component" i mean the whole mobo, the graphics card, etc.. Interesting enough, most (all ?) HDD's i've seen have power draw figures directly printed on them, or well shown in documentation. Why is it not mandatory for other components ?

[mcoleg]

gmat - good point about the power consumption of the components - would have made life that much easier.

about the sweet point. while i am sure it would be nice to hit it each and every time, a typical multi-use pc would have several different power draws depending on use. lets count:

-hibernating
-c&q and whatever intel's equivalent is for power saving
-lite desktop load
-a couple of different heavy loads (say, movies and games)

at all those the load will be different. i am thinking that it would be nice if all of those fall into 20-80% range. i am not going to begrudge a psu a couple of percent of efficiency loss as long as it's trying its best to stay above 80% at all those loads.

what do you think?

[OCC_Yoda]

Since it seems everybody isa talking about the lifespan of a given PSU...

Lets point some things out...
First of all a PSU discussion is never as cut and dry as people wish them to be..

When discussing the lifespan of a given PSU it really honestly helps to know what brand of caps and other information concerning a given PSU!!

Caps are interesting in there own right..

Some middle of the road caps that are properly cooled or utilize proper ventilation will last a very long time!!

Then there are other caps that even with proper venilation will last who knows how long!!

Certain components of a given PSU have track records regardless of which PSU they are used in!!

Once again lets dispell another myth.
Just because a PSU is rated 80+ efficent does NOT mean it is built better than a PSU that is NOT on the 80+ website......

efficieny does not correlate to quality and workmanship....

I have seen many many stories of PSU`s rated 80+ and in 6 month dying....lol

[jessekopelman]

"PSU with an operating temperature of 20C would last longer than one with 30C" - once again, technically, yes.

I challenge you to find anything supporting this assertion. My claim is that until you hit certain thresholds raising or lowering operating temperature will have no effect at all on lifespan.Exactly what those temperatures are is unknown, but my guess would be around 10C on the low end and 35C on the high end.

[jessekopelman]

there's no "greater" details, really. i want to keep it as simple as possible - which one of the psu's will produce more heat will die first. if the temperatures are close to each other, the life expectancy difference is insignificant, if it's far apart than we can make some guesses.

if you got any suggestions though, i'd like to hear them.

AFAIC, the point in contention is whether it is safe to continuously operate a PSU between 50-80% utilization. Therefore, this is the only load range that should be put on the lower rated PSU (the higher rated PSU gets exactly the same Wattage loads). This aside, the other issue of concern is what type of ambient conditions are you trying to replicate? Ideally, you would want to duplicate something that could be considered a typical low-end gamer configuration (high-end gamer seems likely to defy the idea of typical configuration).

[Mr Evil]

I challenge you to find anything supporting this assertion. My claim is that until you hit certain thresholds raising or lowering operating temperature will have no effect at all on lifespan.Exactly what those temperatures are is unknown, but my guess would be around 10C on the low end and 35C on the high end.

The "10C increase in temp halves lifespan" rule of thumb that is often heard mentioned in various places is not just some old-wives-tale, but based on measurements on all sorts of components, all of which show a similar relationship between lifespan and temperature. There will be upper and lower limits, as I mentioned before, but you won't hit them as long as you keep within the components rated temperature range.

For instance, capacitors lose capacitance faster and transistors* acquire more defects in their crystal lattice as temperature increases.

*look at section 3.4, page 43

[MikeC]

The "10C increase in temp halves lifespan" rule of thumb that is often heard mentioned in various places is not just some old-wives-tale, but based on measurements on all sorts of components, all of which show a similar relationship between lifespan and temperature.

According to Tony Kordyban, an expert in electronics cooling, this is a myth:

Stupid Thing No. 9. Reducing temperature because "every 10°C drop doubles the life."

This is still the gospel of the land. It started with the U.S. Department of Defense Military Handbook 217, which became the standard for electronics reliability. The 10°C rule was part of it.

Too bad it's not true. Not even the military uses 217 anymore. But like your mom's rule about not swimming for one hour after eating, this rule lives on.

I discussed some implications on HDD reliability in the PS to the Antec MX1 article.

Here's another relevant article questioning the validity of steady-state temperature data alone to predict electronics failure and shows that it's a lot more complex than that: Use of thermal analysis information in avionics equipment development. I quote a key portion here, but the whole piece is well worth a read for those who wonder whether the original poster's experiments, despite best intentions, can even begin to provide any answers to the questions he is asking.

Steady-state temperature, temperature cycles, temperature gradients, and time-dependent temperature changes all have the potential to affect the reliability of modern avionics. However, because of the use of reliability prediction methods, such as Mil-Hdbk-217, steady-state temperature has often been considered the only stress parameter affecting reliability. At its core are the Arrhenius-based models, formulated to predict the influence of steady-state temperature on electronic device reliability....

One problem with the use of this relationship is illustrated by Figure 1. The lack of correlation between observed failure rate and junction temperature suggest the inadequacy of Arrhenius-based reliability prediction models. Also, the effects of temperature on electronic devices are often assessed by accelerated tests carried out at extremely high temperatures. For example, electromigration tests are often conducted at temperatures above 250oC and at current densities ten times those applied in actual operation; the test results are then extrapolated to operating conditions to obtain a value for the thermal acceleration of device failures.

[Mr Evil]

According to Tony Kordyban, an expert in electronics cooling, this is a myth:

I discussed some implications on HDD reliability in the PS to the Antec MX1 article.

Here's another relevant article questioning the validity of steady-state temperature data alone to predict electronics failure and shows that it's a lot more complex than that: Use of thermal analysis information in avionics equipment development. I quote a key portion here, but the whole piece is well worth a read for those who wonder whether the original poster's experiments, despite best intentions, can even begin to provide any answers to the questions he is asking.

They're right, the commonly used Arrhenius model only applies to failure modes like chemical processes etc. while there are many other modes that don't have such a simple temperature relationship, like the ones mentioned in that 1999 article: "interconnects and connectors, system design, excessive environments, and improper user handling". Thermal cycling can be more damaging than absolute temperature too.

However, I would argue that the last two failure modes can be mostly discounted. "improper user handling" because, well, it's the user's fault, and "excessive environments" might well apply to aerospace (one of the graphs in that article had data from -10 to +40C), automotive and mobile applications, but a PC just sits indoors in more-or-less constant environmental conditions. Most users probably wouldn't even touch it other than to flick the switch at the back.

"interconnects and connectors" I can see being important for a PSU, especially with recent complaints about lead-free solder reliability problems, but on the other hand their reliability is the same no matter what the power rating of a PSU and thus aren't relevant to the reliability under differing loads.

As for the HDD results: Like I said before, mechanical components are different. It's very likely that to test a PSU all the way to final failure of the internal circuitry, you would have to replace the fan at least once.

[mcoleg]

jessekopelman - you are correct, there would be no "appreciable" effect on the lifespan. that is because if ran in that range the lifespan would be so long that the psu in question will outlive it's usefulness before it has a chance to break down (providing the components don't have "birth" defects). therefore for our purposes it's pretty much the same.

on the other hand, we are talking about a liquid. if there is a negative thermal differential between liquid and the ambient, the liquid will evaporate. the higher the temperature of the liquid, the faster the rate of the evaporation will be. this is very simplistic of course but that's pretty much all there is to it.


"AFAIC, the point in contention is whether it is safe to continuously operate a PSU between 50-80% utilization." - while it does seem what most people disagree about, i want to take more measurements under different loads to have a more complete picture.

and how's my rig not a high end, exactly? not sure what you mean by: "high-end gamer seems likely to defy the idea of typical configuration" btw.

seriously though, i thought i'll use 3 different systems for the test (have 2 already, putting together the third now). one of them should be close to a mid-range system you mentioned.


Mr Evil, MikeC - "10C increase in temp halves lifespan" rule is often misused and applies to certain temperature ranges only. thus, there's some confusion about it. i am pretty certain it might still apply to electrolytic capacitors in specified temperatures just fine. i won't use that rule for mechanical devices though.


one more thing, everyone - the temperatures we should be concerned with are 40C and above. majority consumer-grade psu's are rated at 40C, roughly; an increase in temperature above that mark means the de-rating curve must be applied.

[MikeC]

The real issue is how to interpret the results you will get. You will have 3 samples (?) in 3 systems? or.... What does the failure of one sample in one condition tell us? Only that this sample failed in those conditions. That's about all.

You need much greater numbers to make your efforts have any meaning at all.

If you tested 30 PSU samples in those 3 systems and 10 of the same type failed consistently after a certain period in one of those 3 systems, then I'd say you have a result that's statistically meaningful, and a very close analysis would help reach some generalized conclusions, at least about that particular PSU model under those particular circumstances.

On the other hand, it might only mean that the very good buy on a huge batch of electrolytic capacitors (1500uF - 105C) made by the purchasing dept last spring wasn't such a good buy because they're failing at a much lower temperature than spec'd.

[jessekopelman]

not sure what you mean by: "high-end gamer seems likely to defy the idea of typical configuration" btw.

Just that high-end systems are bound to be very customized and I would be surprised if it were easy to get consensus on exactly which components should be used and how they should be configured. Meanwhile, I think saying a low-end system would have a < $200 CPU (mild or no OC), 1X 7200 RPM HDD, 1X <$200 video card, and 1X < $100 sound card in a mid-tower case would not be controversial.

one more thing, everyone - the temperatures we should be concerned with are 40C and above. majority consumer-grade psu's are rated at 40C, roughly; an increase in temperature above that mark means the de-rating curve must be applied.

This is pretty much what I'm getting. My concern is that someone will take PSU A running at 25C and PSU B running at 35C and extrapolate that PSU A will fail much sooner. My claim is that there would only be an issue if A is hotter than B and A is hotter than X (X being some safety threshold -- 40C according to you). Furthermore, if it turns out that both A and B are running hotter than X, I think we would best be served by rethinking our value for X before drawing any other conclusions.

[mcoleg]

Mike - you are right on one thing - i might not need to make it so complex. i don't really need 3 systems, only one, the system that provides the largest power draw. i wanted to test 3 of them at different loads to see how power consumption relates to the temperatures in psu's over wide range of loads but it's not necessary for the conditions of this experiment.

basically i need two numbers.

keeping things as simple as possible - 2 similar psu's, different power ratings, maximum load a system can master. the one that runs cooler will live longer. heat kill psu's, fire's hot, water's wet, etc, etc. . simple.

i do not think that the variance in random samples will be so great to justify an approach as you are proposing; it's not a load-to-failure experiment. same components and same configuration will give us similar results across the board as long as conditions are kept the same. a small variance in results is possible if different units are tested but it shouldn't be large enough to justify testing multiple units.

of course, if someone wants to disprove the results, all they have to do is to get two psu's of the same model and similar build with different power ratings and test them, replicating the experiment. nothing to it .




jessekopelman - the 40C limit and de-rating curves are not of my making. it's pretty much standard for consumer-grade psu's with some exceptions. check this seasonic page (scroll down):

http://www.seasonicusa.com/s12plus.htm

and also jonny's guide:

http://www.jonnyguru.com/forums/showthread.php?t=1036

so we can't really re-think our value for X unless it's in a manufacturer's specs.

we might see a scenario when A and B are hotter than X, at least for a short time. the cooling in a psu is a dynamic system that is designed to adjust to changing temperatures. how well it's calibrated, i don't know.

if A and B are constantly hotter than X, the de-rating curves will need to be used; also i won't be recommending to use those units for any builds .


as for mid-range vs. high-end systems... i can't make a mid-range system to load a 380w psu close to 80%. the system i've tested barely loads it to 70+%. unless you are saying that the test will still be valid without a border-line load, that is.

[MikeC]

i do not think that the variance in random samples will be so great to justify an approach as you are proposing; it's not a load-to-failure experiment. same components and same configuration will give us similar results across the board as long as conditions are kept the same. a small variance in results is possible if different units are tested but it shouldn't be large enough to justify testing multiple units.

I certainly do think sample variance can be significant. If your experiment is going to have any weight, you have to try and reduce the significance of sample variance.

But at this point I have to stop and ask -- what the heck are you going to do? I looked back in the thread and found this:

back to a simple premise, then - psu's are killed because of the high temperature, not the loads.

and here's a simple experiment - two psu's of the same model with different power ratings. one is 380w - most people will think twice before using it with my system; second is 500w - practically everyone is in agreement that it's plenty even in a long run.

run loads, measure temperatures with both. if the temperatures are close to each other, 380w is enough for a prolonged use; if they are significantly apart, it's not.

what do you think?

So then you're trying to determine whether one will fail sooner than the other? You goals are not very clear.

In any case, sampling size is critical in an experiment like this. To run the test with one sample each of two different models... is really not worth doing. Sample variance is a serious enough issue with fan performance at slow speeds, but when you're trying to assess product longevity under thermal stress, you absolutely need at least a handful of samples.

I may not be well versed enough in the language of statistical analysis and stress testing to get across the precise reasons why. Perhaps one of the engineers following this thread can see and better expound on the reasons.

[mcoleg]

Mike - let's say for now i am trying to determine the difference in temperatures under the same load between two power supplies of the same model with different power ratings. how's that?

we can draw conclusions after the test results are in.

as for the "significance of sample variance"; a couple of things i think i should say...

your approach will invalidate my initial experiment. i already thought about it, that's why i spent lots of time trying to hunt down some statistics on psu's longevity.

your approach will also invalidate any single hardware review made, including those made on this site. i don't see anyone testing a couple of dozen psu's to determine some median numbers just to be as precise as possible.

why it's not done?

first off - practical reasons - too expensive and time-consuming.

secondly - there's such concept as "good enough". it might sound as something lame but it's actually a valid approach used in science and engineering. one need to use such instruments and record data with enough precision only to answer questions in hand. anything above that is a realm of diminishing returns.
that is why reviewers assume that any other product same as the one they reviewed will have similar characteristics.


now, what i just said does not make their approach accurate. it also does not make my experiments very scientific either. however, an independent verification was made with my first test and that lent it some validity. if anyone would like to replicate the second test, they are welcome to it as well.

[jessekopelman]

the 40C limit and de-rating curves are not of my making. it's pretty much standard for consumer-grade psu's with some exceptions.

I'm not so much questioning 40C as a number but more questioning whether your method (yet undeclared) of measuring temperature will be the same as a vendor's. What I'm getting at is that your 40C might equal their 35C, because you are not measuring the same way/place.

as for mid-range vs. high-end systems... i can't make a mid-range system to load a 380w psu close to 80%. the system i've tested barely loads it to 70+%. unless you are saying that the test will still be valid without a border-line load, that is.

What I'm saying is that to keep things simple we should only try and obtain loads that are within the range of 50-80%. If you don't even think you can get up to 80%, fine. The key is that you get some loads that are within the range.

[jessekopelman]

Thinking about this more:

The inherent heat of the PSU (Required Draw/Efficiency - Required Draw) will be similar and maybe even favor the lower rated PSU at a given load (better efficiency). However, the PSU fan charged with dissipating this heat will also have to deal with the heat from other components, in most case configurations. So, one theory is that we must operate a PSU well below rating to preserve cooling headroom to deal with these other heat sources. This means that the key issues in any experimental test of this theory is what heat producing components are used, how they and the PSU are laid out in the case, and what non-PSU cooling methods are present.

One conclusion I'd draw from the above, for mcoleg's experiment, is that the test builds must have enough overall cooling that ambient case temperature is kept below 40C. Furthermore, the PSU fan should not be the only system exhaust -- that is just not a standard configuration for anything other than extreme SFF systems and those are generally not recommended for high wattage scenarios. What this amounts to is that both case ambient and PSU temperatures must be monitored for the experiment. After all, a result of PSU temp = 50C and case temp = 43C indicates inadequate system cooling more so than an underprovisioned PSU.

[peteamer]

Back to the PSU Wattage required front:

Looks like the price drops have already begun. Q6600 + ecs mobo for $299:

Edit: Bought the combo morning... initial tests show 2000PPD folding with a Windows SMP (project 2651). Power consumption figures aren't that bad either:

Idle (1.6ghz) = 66W
SMP Folding 100% ([email protected]) = 120W
SMP Folding 100% ([email protected]) = 140W

From Here.

[mcoleg]

jessekopelman - not sure how manufacturers measure it. i plan on placing a thermocouple at the exhaust grill using a multimeter connected to a pc to record the temperatures. i got this one:

http://www.amazon.com/gp/product/B00064 ... 9BZKUXE7W2

i've been using a cheap thermometer with a thermocouple for a couple of weeks now, seem to work fine:

http://www.performance-pcs.com/catalog/ ... s_id=21106

"try and obtain loads that are within the range of 50-80%"

- the 380 watter will be loaded above 70%, the 500 watter only above 50%.

"However, the PSU fan charged with dissipating this heat will also have to deal with the heat from other components, in most case configurations. So, one theory is that we must operate a PSU well below rating to preserve cooling headroom to deal with these other heat sources. This means that the key issues in any experimental test of this theory is what heat producing components are used, how they and the PSU are laid out in the case, and what non-PSU cooling methods are present.

One conclusion I'd draw from the above, for mcoleg's experiment, is that the test builds must have enough overall cooling that ambient case temperature is kept below 40C. Furthermore, the PSU fan should not be the only system exhaust -- that is just not a standard configuration for anything other than extreme SFF systems and those are generally not recommended for high wattage scenarios. What this amounts to is that both case ambient and PSU temperatures must be monitored for the experiment. After all, a result of PSU temp = 50C and case temp = 43C indicates inadequate system cooling more so than an underprovisioned PSU."

- not even a consideration at this point. it will be on a test stand. the idea is to record the difference in temperatures, the ambient doesn't matter.

it will matter for practical applications though so i might introduce an artificial heat load later on just to see how it will hold out.

[jessekopelman]

- not even a consideration at this point. it will be on a test stand. the idea is to record the difference in temperatures, the ambient doesn't matter.

I think you will learn nothing from this setup. Unless your 500 Watter has a very strange fan controller, there is no reason for it to be any cooler than a 380 Watter with the same draw. The only thing to produce heat is the inefficiency of the PSU and it is likely that the smaller PSU will actually be more efficient at this load.

Look at how SPCR measures both intake and exhaust temps in their PSU testing and rightly notes that it is the temperature delta and not the actual temperatures that are import. Why? Because the fan controller goes by temperature and not draw and temperature is highly dependent on ambient conditions -- not just the waste heat created by the PSU.

[mcoleg]

and i want to record the temperature delta as well; so, how's that a problem?

[mcoleg]

peteamer, here's some more:

http://www.hardforum.com/showthread.php?t=1178181

http://www.ocforums.com/showthread.php?t=517875

http://www.ocforums.com/showthread.php?t=518137

there was no mentioning of any other components though in the link you posted.

[MikeC]

Mike - let's say for now i am trying to determine the difference in temperatures under the same load between two power supplies of the same model with different power ratings. how's that?

And why should there be any difference at all? While I am all for people experimenting to find things out for themselves, I fail to see the value of experimentally verifying this particular matter. It's a no-brainer: There is no appreciable difference that will have any impact on any aspects of PC performance or longevity.

If they really are the same circuitry, and they have the same fan and fan controller and the same heatsinks (which I believe holds true for the Earthwatts 380 and 500), and you're loading them to around the middle of their range -- 250~300W -- there should be no appreciable difference in temperature rise. The same amount of heat will be generated within and dissipated by the PSU's cooling system! Maybe a degree or two, but this is too small to consider significant when you have only one sample of each.

Besides, you will not be able to generalize in any way about the thermal characteristics of higher vs lower rated PSUs -- only about this particular higher rated one vs that particular lower rated one -- because what holds true about this line may not hold true for all PSU lines. For example, if you compare a Seasonic S12-330 vs a S12-380, the latter may come off as cooler at higher loads simply because its heatsinks are larger. Also as you go higher up the line in PSUs, fans generally become higher speed models, so that as output is increased, more cooling is available for the higher rated PSU.

[jessekopelman]

and i want to record the temperature delta as well; so, how's that a problem?

The delta I'm talking about is between ambient and exhaust . . . if that's what you're talking about too then why say the ambient doesn't matter? Did you just mean that, since it's on a test stand, ambient will not get high enough to make the test unfair?