A [H]ard Look at Power Supplies

Paul_Johnson said:
Doesn't change the fact that it is the recommended operating temperature range not the required output temperature.
Click to expand...

I'm sorry, I didn't understand your point.

Anyways, I need to apology because it seems that I didn't make myself clear. (Still learning to write in english...)

I will try again.

1. There is no such thing as "rated at temperature X" in the ATX specification

2. As a matter of fact, there is no "25oC" string in the specs

3. Full load = maximum/100% continous load. A full load for a 500W PSU is 500W

4. Read very carefully the folowwing recommendation from the ATX specification:
6.1 - Temperature - RECOMMENDED
Operating ambient +10 °C to +50 °C (At full load, with a maximum temperature rate of change of 5 °C/10 minutes, but no more than 10 °C/hr.)
Click to expand...

5. Consider this example (a):

A 500W PSU, which can operate in an environment temperature of +10oC to +50oC, but will only supply 405W at 50oC. This PSU does not comply with ATX recommendation 6.1.​

6. Consider this other example (b):

A 500W PSU, which can operate in an environment temperature of +10oC to +50oC, and will supply 500W at 50oC. This PSU does comply with ATX recommendation 6.1.​

7. An engineer would state that PSU (b) have an environment temperature operating range of 10oC to 50oC at full load. PSU (a) have an environment temperature operating range of 10oC to 50oC, with a derating of 10% for each 10oC above 30oC.

8. We ;) would state that PSU (b) is 500W rated at 50oC, and (a) is 500W rated at 30oC.

9. To be absolutely clear: I have the opinion that rating a PSU below 50oC is simply not honest

Cordially,
Virgulino.
 
davidhammock200 said:
If you won't believe us, maybe you will believe PC P&C. :rolleyes:
Click to expand...

I belive in them and in Jerry Pournelle since 1989.

david, I think you missed my point.

My argument about this "rated at X" marketing lingo is that the ATX specification is clear about it. It recommends that a PSU must operate at 50oC at full load. Please, do read the specification.

By logical inference, the ATX specification is recommending that a PSU should be "rated at 50oC".

Cordially,
Virgulino.
 
davidhammock200 said:
If PSU must deliver rated output at 50C to meet spec, then 95%+ of all PSU's do not meet spec. :cool:
Click to expand...

ATX Specification terminology:

Requirement - The status given to items within this design guide, which are required to meet design guide and a large majority of system applications.

Recommendation - The status given to items within this design guide, which are not required to meet design guide, however, are required by many system applications. May be a required item in a future design guide.
Click to expand...

However, I do believe that it's possible that 95% of all PSUs do not meet the specs requirements. :mad: And that was the main point of my original post: The output transient response and capacitive load requirements.



Cordially,
Virgulino - Standards freak :D

PS: Speaking of PC P&C, I'm worried about them. Have you seen their latest Silencers? Rated at 40oC, and violating the ATX requirement of 17ms of hold-up time!

Yep, this cute shiny Silencer 750 is not ATX compliant, although, to be fair, PC P&C does not claim that it is. I wonder if they will put some neon lights in the next model... :rolleyes:
 
Virgulino Ferreira said:
I'm sorry, I didn't understand your point.

Anyways, I need to apology because it seems that I didn't make myself clear. (Still learning to write in english...)

I will try again.

1. There is no such thing as "rated at temperature X" in the ATX specification

2. As a matter of fact, there is no "25oC" string in the specs

3. Full load = maximum/100% continous load. A full load for a 500W PSU is 500W

4. Read very carefully the folowwing recommendation from the ATX specification:


5. Consider this example (a):

A 500W PSU, which can operate in an environment temperature of +10oC to +50oC, but will only supply 405W at 50oC. This PSU does not comply with ATX recommendation 6.1.​

6. Consider this other example (b):

A 500W PSU, which can operate in an environment temperature of +10oC to +50oC, and will supply 500W at 50oC. This PSU does comply with ATX recommendation 6.1.​

7. An engineer would state that PSU (b) have an environment temperature operating range of 10oC to 50oC at full load. PSU (a) have an environment temperature operating range of 10oC to 50oC, with a derating of 10% for each 10oC above 30oC.

8. We ;) would state that PSU (b) is 500W rated at 50oC, and (a) is 500W rated at 30oC.

9. To be absolutely clear: I have the opinion that rating a PSU below 50oC is simply not honest

Cordially,
Virgulino.
Click to expand...

The problem is you aren't reading very carefully. Please reread that entire passage not just the part you want to see.
 
Virgulino Ferreira, I think you're missing the point. Recommended means that this is a good idea to have happen but it is not necessary.

Required means that this must happen.

6.1 that you're pointing to recommends +10C to +50C at full load but it's not something that must be met to make the ATX specification, it does in no way state that it is required.
 
Virgulino Ferreira said:
These percentages are the required transient step sizes that the PSU must be able to handle. They are not the recommended load percentages.

In other words, the specification is saying that, if the PSU has a 12Vcpu rail capable of 20A, it must be able to handle a load change from 5A to 17A (12A = 20A * 0.6), at a rate of 1A per microsecond. And that load change can happen 10000 times a second, up or down. All this while maintaining the required 5% voltage regulation.
Click to expand...

I didn't say the ATX12V didn't require this load balance capability. I merely said it was unrealistic. If you do this with a modern PSU, you'll never get to load the +12V rail up to it's full (supposed) capability, since the +12V rail is the bulk of what a PSU can output.

Virgulino Ferreira said:
As for multiple 12V rails, I personally hate it. Unfortunately, the 240VA limit is a legal requirement for many PSUs.
Click to expand...

Actually, it's not. Only for a certain class of PSU. UL has another class that a PSU can fall into that doesn't have a 240VA limit. Intel has acknowledged that the 240VA per rail limit is an unrealistic constraint for modern day PC's, but has not re-written the standard in the past two years to reflect this. :(
 
madmat said:
Virgulino Ferreira, I think you're missing the point. Recommended means that this is a good idea to have happen but it is not necessary.
Click to expand...

When did I say otherwise??

What I said:

The original article has a factual error, and I quote:
http://enthusiast.hardocp.com/article.html?art=MTI4OSw1LCxoZW50aHVzaWFzdA==
The ATX specification calls for power supplies to be rated at 25c (77F)[...]
Click to expand...

This is false. Otherwise, somone please show me in which page of the specification it is!

The only related information in the ATX spec, as I pointed, is the now infamous item 6.1. But many readers argued that this item had nothing to do with "rating temperature".
 
I have to admit that at this moment I can not find it. :)
 
jonnyGURU said:
I didn't say the ATX12V didn't require this load balance capability. I merely said it was unrealistic. If you do this with a modern PSU, you'll never get to load the +12V rail up to it's full (supposed) capability, since the +12V rail is the bulk of what a PSU can output.
Click to expand...

I'm not sure I'm following you. What item 3.2.2 has to do with load balance/distribution?

Are you saying that a step size of 60% implies a 60% maximum load?
 
No. By the time you loaded up the 3.3V and 5V, there wouldn't be any way to also load the +12V up because the total capability of the PSU would be exceeded.

Because (3.3V + 5V) + (12V) /= total wattage.
 
so I guess Virgulino Ferreira ran out of steam....

I had my bbq so I just popped another cold one!! I am waiting for the 2nd matinee to start...

I love sequels...
 
OCC_Yoda said:
so I guess Virgulino Ferreira ran out of steam....

I had my barbque so I just popped another cold one!! I am waiting for the 2nd matinee to start...

I love sequels...
Click to expand...
Given the current state of affairs (GPU's) we should only recommend PSU's that DO NOT COMPLY with the evil Intel 240VA spec!

ONLY single rail PSU's!

Of course we can include the newest Seasonics & Seasonic built Antecs & Corsairs,
that advertise multi-rail (pretending to bow to the evil Intel), but are really single rail!

That should start round 2! :D
 
jonnyGURU said:
No. By the time you loaded up the 3.3V and 5V, there wouldn't be any way to also load the +12V up because the total capability of the PSU would be exceeded.

Because (3.3V + 5V) + (12V) /= total wattage.
Click to expand...

Jonny, I'm very sorry, but I'm having trouble to understand you. Could you please help me out here?

Since my first message, all I'm talking about is item 3.2.2. It's about transient loads ( delta watts / delta time ).

You seem to disagree with this item (and me), which is ok. My real problem is that your argument is about power distribution. As if you were talking about item 12 of the ATX spec.

There is absolutely nothing in item 3.2.2 that remotely relates to how many watts each rail (and combination of) must be capable of.

I kindly ask you to forgive my difficuly to express myself, and help me figure what our disagreement is about.

Cordially,
Virgulino.
 
All I am saying is that if you increase in the transient steps specified by table 8, you'll end up with a load with unrealistically high 3.3V and 12V and you will never test the maximum of the +12V because you will have maxed out the 3.3V + 5V combined before reaching the max of the +12V.

Take the Turbo-Cool 1kW load test sample as an example (find it online.) They use this method and you'll find that they never get to the point where the +12V is loaded to the maximum because the 3.3V and 5V top out first.
 
Virgulino Ferreira said:
When did I say otherwise??

What I said:

The original article has a factual error, and I quote:
http://enthusiast.hardocp.com/article.html?art=MTI4OSw1LCxoZW50aHVzaWFzdA==


This is false. Otherwise, somone please show me in which page of the specification it is!

The only related information in the ATX spec, as I pointed, is the now infamous item 6.1. But many readers argued that this item had nothing to do with "rating temperature".
Click to expand...

You said, and I quote:
Virgulino Ferreira said:
A 500W PSU, which can operate in an environment temperature of +10oC to +50oC, but will only supply 405W at 50oC. This PSU does not comply with ATX recommendation 6.1.
Click to expand...

And there you are mistaken since it does indeed put out 500W at a point below 50C it is ATX compliant. The 10C-50C is a recommendation not a requirement so even if a PSU de-rates at 30C it's still in compliance as long as it meets the requirements of the ATX spec. Hence my post about the difference in what recommended means versus required.
 
And going back to the point about 3.2.2, you could ALSO do those tests, but then you're just doing an additional barrage of tests for really no reason except to prove that you are testing with adherence to an antiquated standard.

Sure, you could say to NOT do them would be out of laziness, but I would certainly rather have the more realistic high +12V loads than not! :D
 
madmat said:
The 10C-50C is a recommendation not a requirement so even if a PSU de-rates at 30C it's still in compliance as long as it meets the requirements of the ATX spec. Hence my post about the difference in what recommended means versus required.
Click to expand...

Of course, I agree with you. And I quoted the ATX terminology for required x recommended, just to make clear that I was using the same definitions.

BUT, this is what I said, as you quoted: "does not comply with the RECOMMENDATION number 6.1".

What i did NOT said: "does not comply with the requirement" or "does not comply with the ATX specification".

I apologise if that wasn't clear enough.
 
Bbq said:
The only thing I see, is the assumption that a computer will max out a power supply for continuous periods of time. While reasonable, computers really don't take as much power as most people like to think: this SPCR article lists the actual power consumption of some of the more power-consuming components. Actual power draw is only slightly above 200w. In my case, a highly oc'd opteron, 5 hard drives, and a 7800gt, draws just over 100w full load. Not a particularly huge stress.

Just something to look out for.
Click to expand...

I find that IMPOSSIBLE to believe.
 
jonnyGURU said:
All I am saying is that if you increase in the transient steps specified by table 8, you'll end up with a load with unrealistically high 3.3V and 12V and you will never test the maximum of the +12V because you will have maxed out the 3.3V + 5V combined before reaching the max of the +12V.
Click to expand...

Thanks for you patience, jonny. I understand your point now.

It seems that we have a different understanding of item 3.2.2.

I will try to explain my reasoning:

1. It's not written that the only load change allowed is the maximum step size. In other words, if the maximum step size for 5V is 30%, a 5% step is allowed.

2. It's not written that, if the load on 5V changes 10%, 12V must change by X%.

3. It's not written that for every step up, the next step must be down. In other words, it's not written that a load change of +10% cannot be folowed by a load change of +20%. The only restriction is that there must be a 10ms delay between them.

4. If I really understood you correctly, your understanding of 3.2.2 implies that the ATX specification is self-conflicting: The power distribution tables and graphics in item 12 would be in conflict with 3.2.2.

To summarize, I believe that the following example is perfectly valid according to 3.2.2, and we reach 100% load on 12V:

..............0ms....15ms....30ms......45ms
----------------------------------------------------------------------
12Vcpu....5%.....65%......100%.....100%
12V.........5%.....45%.......85%......100%
5V...........2%.......3%........4%..........5%
3.3V........2%.......3%........4%..........5%

ms = miliseconds
% = percentage of maximum rated amperes for that rail

jonnyGURU said:
Take the Turbo-Cool 1kW load test sample as an example (find it online.) They use this method and you'll find that they never get to the point where the +12V is loaded to the maximum because the 3.3V and 5V top out first.
Click to expand...

I'm sorry, but I will disagree again. I don't think that they used this method. There is no reference to any methodology, or any timing, and the load percentages don't match those in 3.2.2.


Cordially,
Virgulino.
 
jonnyGURU said:
And going back to the point about 3.2.2, you could ALSO do those tests, but then you're just doing an additional barrage of tests for really no reason except to prove that you are testing with adherence to an antiquated standard.
Click to expand...

This is why I think that 3.2.2 and 3.2.7 are utterly critical:

To maintain voltage regulation with a static load is easy. No matter how big the load is.

To maintain voltage regulation with huge, brutal, lightning fast load changes, that is really hard.

At least that is what I read in a number of electrical engineering papers. Particularly about voltage regulation modules for cpus.

Nowadays computers are brutal with PSUs: we go out of idle, do a few million instructions while updating some 3D window, and then back to idle. That's easily a 100W increase and a few miliseconds later 100W decrease in the load.

jonnyGURU said:
Sure, you could say to NOT do them would be out of laziness, but I would certainly rather have the more realistic high +12V loads than not! :D
Click to expand...

I regret that I didn't write this in my original message, but better late then never:

Of all the major items in the ATX specification, 3.2.2 and 3.2.7 are, in my opinion, the hardest and most complex to test.

So hard that it's almost unfair to request this tests. But, at the same time, so relevant that it is impossible to do a conclusive PSU test without testing this items.


Cordially,
Virgulino.
 
I disagree. And at this point it as if you're just trolling.

A "crossload" where there's an exceptionally high or exceptionally low 3.3%+5V or +12V load is harder on a PSU.

Period.
 
jonnyGURU said:
I disagree. And at this point it as if you're just trolling.

A "crossload" where there's an exceptionally high or exceptionally low 3.3%+5V or +12V load is harder on a PSU.

Period.
Click to expand...

When I pulled up my lawn chair is when I thoughgt it was heading or had already been going in this direction...

Or shall we say having hands on experience and also book knowledge is much
preferable to having book knowledge and misinterpreting it......

A good example is what you said Jonny.....it seems as of Vitgulino is just reading into whats there on paper...never mind the ATX standards at best are antiquated!

Then MadMat hit the donkey in the ass when he said-- Virgulino Ferreira, I think you're missing the point. Recommended means that this is a good idea to have happen but it is not necessary.

Required means that this must happen.

:D
 
Spazilton said:
I find that IMPOSSIBLE to believe.
Click to expand...

Believe it. Mine's drawing 300W at full load and I have an X-2 3800+ @1.465V, 7950GT's in SLI, 2 optical drives, 1X 160gig HDD, 1X 500 gig HDD, Ultra ChillTEC air cooling, 2X 1gig of ram and an X-Meridian 7.1 and my system is drawing an uncharacteristically high wattage for what I've got. Typically people running the same wattage I am are running dual dualcore CPUs.
 
Virgulino Ferreira seems to be among the few who understand the issues inherant in dynamic load regulation. Finding PSU that go POP running a static load is a good measure for sorting out the weaklings, but it tells us quite little about fitness for powering a PC. No offense meant to anyone but I would much rather own a PSU that meets Virgulino's interpretation of spec than someone else's, and would want reviews to do as much as reasonably possible to differentiate these factors instead of assuming they don't matter.

The overshoot or depression from changing loads exceeds the ripple from a static load, by quite a lot in some cases. It is an extremely important factor, but the remaining question is whether it is feasible to test. We cannot expect extraordinary additional work by Paul, it is already quite an undertaking and there has to be a line drawn, how much work can be put into each review and how much time is spent on each part of testing.

If it's too hard to test this, then PC PSU testing of some units may be a lost cause beyond finding certain kinds of faults or grossly overrated capacity. These two are still important issues, it is not as though testing will be useless to only discriminate these factors and avoiding PSU failure and fraudulent ratings are perhaps priority #1 for many owners, but iIn the end, testing must necessarily aim to differentiate in the intended application. It's not just a generic switching power supply, it's one targeted for specific applications in a modern computer.

It does not matter if the spec is "antiquated" as someone wrote, because at the very least, the PSU is spec'd for this and has to meet it's own spec no matter what the system is like. When we start making excuses for why a PSU doesn't even have to meet it's own or ATX specs, it is just chaos that ensues, the whole point of specifications exists for a reason. If you don't like the specs, contact Intel - but until then that's the measure of a PSU, anything else you'd like to require for powering a modern system as a selection criteria may go beyond these specs (which we're arguing here), but should never ignore them.
 
Then by your own words you're stating that because MOST of the enthuisiast power supplies available today do NOT adhere to the ATX12V specification in regards to having the ability to put out more than 240VA on a single rail or not having the capability of having it's +12V loaded to it's supposed maximum capability and still be able to have the equally appropriate (appropriate as per the ATX12V specification) load on the 3.3V and 5V as well, that we should ignore their very usefulness, capabilities, what have you.

This would include all currently manufactured Seasonic units (including Antec and Corsair) and all of the current PC Power and Cooling power supplies.

Because outside of having 24 wires in the appropriate 24 places of the main "ATX" connector and a few other connectors that happen to be wired correctly, these power supplies DO NOT adhere to the ATX12V spec.
 
Actually with the ATE employed by Paul and Jon you could easily do dynamic loading going from one load to another as quickly as the human hand can push a button but the logging of the results would be nearly impossible since you have to monitor the voltages and log them manually.
 
jonnyGURU said:
Then by your own words you're stating that because MOST of the enthuisiast power supplies available today do NOT adhere to the ATX12V specification in regards to having the ability to put out more than 240VA on a single rail or not having the capability of having it's +12V loaded to it's supposed maximum capability and still be able to have the equally appropriate (appropriate as per the ATX12V specification) load on the 3.3V and 5V as well, that we should ignore their very usefulness, capabilities, what have you.

This would include all currently manufactured Seasonic units (including Antec and Corsair) and all of the current PC Power and Cooling power supplies.

Because outside of having 24 wires in the appropriate 24 places of the main "ATX" connector and a few other connectors that happen to be wired correctly, these power supplies DO NOT adhere to the ATX12V spec.
Click to expand...
QFT.

Any PSU that is fully spec compliant is usless for a highend SLI system! :p :rolleyes:
 
jonnyGURU said:
Then by your own words you're stating that
Click to expand...

No, by your extension of what I wrote, which is not by my own words, it's by your own words. Even so, you do make a valid point.

... because MOST of the enthuisiast power supplies available today do NOT adhere to the ATX12V specification in regards to having the ability to put out more than 240VA on a single rail
Click to expand...

True, but how does this justify ignoring the specs we were talking about instead of this side-tracked issue? We can also see that Intel themselves has recommended PSU with the post 240VA rails and make inference from it. It also goes back to what I wrote about "going beyond" the spec, that is to mean ALL aspects of performanace must be at, or above the spec.

... or not having the capability of having it's +12V loaded to it's supposed maximum capability and still be able to have the equally appropriate (appropriate as per the ATX12V specification) load on the 3.3V and 5V as well, that we should ignore their very usefulness, capabilities, what have you.
Click to expand...

Not ignore, but certainly derate if they can't maintain their 3V/5V current as well as another make/model with similar specs can, and since it is assumed to be ATX spec compliant, the valid 12V current rating should be that in which the other test parameters still hold true. If a side-note were to mention how much more 12V current it can supply when it's operating outside of ATX specs, some may find that useful as well.

Such matters may depend on the labeled specs, a lot can be forgiven if it is clearly stated on product or packaging rather than found out accidentally. So if they spec a lower rating or combined 3+5+12, that is instead the value to use for meeting the spec.


This would include all currently manufactured Seasonic units (including Antec and Corsair) and all of the current PC Power and Cooling power supplies.

Because outside of having 24 wires in the appropriate 24 places of the main "ATX" connector and a few other connectors that happen to be wired correctly, these power supplies DO NOT adhere to the ATX12V spec.
Click to expand...

I think we can all see quite a bit more similarity, and I wonder why you are putting forth the effort to avoid the only spec. Specs are a cornerstone of PSU selection, and supplies that can't meet them but claim to, should be siezed as fradulent.

Many of the supplies could in fact meet the spec, only with a lower 12V current. As mentioned above all the manufacturer has to do is mention the actual combined rating for the 3 major rails. If they only vaguely imply otherwise, it becomes the same circus as in past years where manufacturers just start making up lofty numbers and we will not have a way to discriminate them with only an 8 hour test.

Did you know you can take some 300W ATX2.0 PSU rated for 15A @ 12V, and get 20A out of that rail? All that's required is low load on 3V+5V, and a very minor change in the PSU feedback loop to put more bias on 12V rail. How would you rate such a supply compared to one claiming it's a 400W model with 20A @ 12V, but the same limitation of not being able to supply the labeled 3V+5V?

If we took the two above supplies which were actually identical in capacity, test both at 80% load which is a higher load on the latter 400W labeled specimen, it makes the one overrated as a 400W supply look like it's better when it is actually same or arguably worse for being less conservatively rated, misleading for it's claimed higher 3V & 5V capacity (as it usually the case when as you suggested, the 3V+5V has to be lower.

Your ideal goes directly against fair comparisons with some (not all) PSU, particularly those with less conservatively rated 12V rails. Some PSU we cannot realistically expect to output 80% for their lifetime even if they can manage for 8 hours. While we may not expect to use 80% continuously, it is still useful to differentiate.

On a related note, I vaguely recall reading a few PSu reviews where the unit might run at high load and seemingly pass, only to fail on subsequent power-ups. Hopefully the 8 hour test will be one of the first rather than the last so as to get a better idea of how that stress has effected it.
 
davidhammock200 said:
QFT.

Any PSU that is fully spec compliant is usless for a highend SLI system! :p :rolleyes:
Click to expand...

Yes some will need higher current 12V rails, but that does not excuse any other departures from the spec. I had thought this issue had been conceded already, since calling a 400-500W PSU the lower end assumes there must be a whole lot of people running SLI'd video cards (which I doubt, it seems more likely they are only one of the more vocal segments of PSU advocates).
 
J.C. said:
Yes some will need higher current 12V rails, but that does not excuse any other departures from the spec. I had thought this issue had been conceded already, since calling a 400-500W PSU the lower end assumes there must be a whole lot of people running SLI'd video cards (which I doubt, it seems more likely they are only one of the more vocal segments of PSU advocates).
Click to expand...
The most vocal by far, as they are the ones pushing for MORE POWER! ;)
 
Glad you guys @ [H] are stepping up and elevating yourselves with your reviews and the way you conduct them.

Impresses me that you get people on the project, that know the fields and varying degrees of electrical engineering.

That way, at least, for me, I know what is being put down on paper is not a bunch of hocus pocus BS.

I want to recommend that you look into the Silverstone PSU Product Line.
 
Caliche said:
Glad you guys @ [H] are stepping up and elevating yourselves with your reviews and the way you conduct them.

Impresses me that you get people on the project, that know the fields and varying degrees of electrical engineering.

That way, at least, for me, I know what is being put down on paper is not a bunch of hocus pocus BS.

I want to recommend that you look into the Silverstone PSU Product Line.
Click to expand...
See www.jonnyguru.com for excellent Silverstone reviews & discussion. ;)
 
(Posted due to a little miscalculation of mine.)
 
Last edited:
You must log in or register to reply here.
Back
Top