jonnyGURU Forums
Home Site Search Reviews Articles Contest Links PSU FAQs  


Go Back   jonnyGURU Forums > Site and Forum Topics > Testing Methodology Discussion

Testing Methodology Discussion Questions and comments regarding the testing methodologies used on jonnyguru.com

Reply
 
Thread Tools Display Modes
  #1  
Old 3 Weeks Ago
quest for silence quest for silence is online now
Silencer
 
Join Date: Sep 2013
Location: Italy
Posts: 267
Thanks: 26
Thanked 7 Times in 7 Posts
Default Transient load test (explained to novices) in contemporary PSUs?

I would like to learn from you, how much and when transient load test response matters in contemporary PSUs: using the forum search feature I didn't find any "crash course for novices", set aside some extremely specific threads, which clearly cannot explain any basics of those responses, even if somehow related.

As a novice it's difficult to me to extrapolate some real meaning (what a "pass" mean, why & when a good pass matter) from those tests, either they are made by HardOCP, or by TechPowerUp!, or eventually by any other reputable tester.

First of all, I'm not aware of any reviewer who may offer a relevant global chart about, in order to have sort of "the whole picture" or "the big picture". Then because very often recent results look like worse than previous one, at least to an unexperienced observer.
I think there are several reasons for that, and I often read on HardOCP that different wattage units are not comparable in those tests, but ignorance can easily foolish me about.
Just for instance, if we look at HardOCP data, something like the 2008 Antec NeoEco 650 (Seasonic-based) seems better than about any 650W contemporary unit, regardless of the price. How could this happen?
But above all, I would like to understand how much a good pass in those tests matter: which are the penalties in real life when a unit exhibit a drop of 300-350mV on the 12V rail, with reference to a better 200-250mV drop? And what about even greater drops? And given that the 5V rail now should matter mostly memory, how the relevant transient load response may concretely affect a typical home computer? And a small server would be a different scenario?

I hope my questions are not too naive or ill-formed, or that some of you would be so kind to equally try to answer me: thanks in advance for your understandings.

Last edited by quest for silence; 3 Weeks Ago at 10:26 AM. Reason: added question mark in title
Reply With Quote
  #2  
Old 3 Weeks Ago
McSteel's Avatar
McSteel McSteel is offline
Naughty
 
Join Date: Nov 2010
Location: North Disturbia
Posts: 1,440
Thanks: 18
Thanked 89 Times in 69 Posts
Default

Quote:
Originally Posted by quest for silence View Post
First of all, I'm not aware of any reviewer who may offer a relevant global chart about, in order to have sort of "the whole picture" or "the big picture". Then because very often recent results look like worse than previous one, at least to an unexperienced observer.
I think there are several reasons for that, and I often read on HardOCP that different wattage units are not comparable in those tests, but ignorance can easily foolish me about.
An inexperienced observer would stick to the final score, and perhaps the judgement of the reviewer pertaining to a particular testing stage, such as transient load testing. A more experienced reader, or someone who wishes to delve deeper, would ask about the finer points of testing on a relevant forum (or better yet, multiple forums and other communities). Both stances are correct and natural.

Either way, transient load testing (or advanced transient testing as it's sometimes called) is possibly the most important one in terms of system stability and overclocking potential/resilience.

Though you may have already read about the methodology behind these tests, I feel it bears repeating. Intel's ATX specification and the associated testing guide call for two stages of this test. The benchmark for such testing is the PSU's ability to hold onto it's regulation. A pass means that the PSU has stayed within the basic boundaries for regulation (meaning +/- 5% from the nominal values) at all times. Falling out of specs is, naturally, a failure. PSU's reaction to the testing scenario is tracked via an oscilloscope (with the required bias capacitance present on the probes -> 0.1uF ceramic disk and 10uF wet electrolytic capacitor in parallel with the probes)

The first stage requires the PSU under test to be statically loaded to 20% of it's declared maximum power, while the second calls for a 50% load. In both stages, a certain load is abruptly added (at a maximum slew rate of 1 A/μs), as a max-load percentage.

The prescribed maximum value is 40% for 12V rails (60% for any 12V rails assigned to CPU exclusively), 30% for 5V and 3.3V rails, and as an exception, 0.5A for 5VSB and 0.1A for -12V.

Take into consideration a hypothetical PSU capable of a declared maximum 25A for 12V1 (ATX, PCI-E, Peripheral) and 20A for 12V2 (CPU) combined 35A, 25A for 5V and 20A for 3.3V. Such a PSU would be subjected to no more than 14A of dynamic loading on +12V (of which 8.4A would be assigned to 12V2 and 5.6A to 12V1), no more than 10A on 5V and no more than 8A on 3.3V.

In order to be able to better track the changes in voltage regulation, the dynamic load may be applied and removed from 50 to 10000 times per second (while respecting the slew rate).

Quote:
Originally Posted by quest for silence View Post
Just for instance, if we look at HardOCP data, something like the 2008 Antec NeoEco 650 (Seasonic-based) seems better than about any 650W contemporary unit, regardless of the price. How could this happen?
In general, different PSU topologies (internal "architecture") yield different strengths and weaknesses. Since the NeoEco's primary is a traditional two-transistor forward (or double-forward), it's range of usable PWM frequencies is fairly broad, and the frequency variance is quite liberal. With a good controller (ChampionMicro's CM680x series are pretty good, actually) the PSU is able to respond to rapid and dramatic changes in load patterns relatively easily and efficiently.

With constrained control methods such as LLC resonant half-bridge, the range of frequencies and the maximum frequency delta is pretty narrow, so the PSU is relatively slow and imprecise when it needs to adjust to a significant load change. Of course, the regulation at secondary will offset this, but watt-for-watt, a double-forward PSU will in most (if not all) cases convincingly outperform a modern LLC resonant one.

Quote:
Originally Posted by quest for silence View Post
But above all, I would like to understand how much a good pass in those tests matter: which are the penalties in real life when a unit exhibit a drop of 300-350mV on the 12V rail, with reference to a better 200-250mV drop? And what about even greater drops? And given that the 5V rail now should matter mostly memory, how the relevant transient load response may concretely affect a typical home computer? And a small server would be a different scenario?
The final impact of any voltage drop is specific to each individual PC. Depending on the motherboard's and VGA's VRM quality, and the amount of stress they're put under; a transient drop may not affect stability at all - or it may knock the system down completely. If the PC is running at the edge of the highest attainable OC, it's bound to be more sensitive to transient V-droop.

These transients usually happen during transitions from idle to load (and back), e.g. when going from idle desktop or forum browsing to a demanding game or a 4k movie reproduction with heavy post-processing (like heavily tweaked madVR). Or perhaps to starting prime95/IBT and FurMark/Unigine simultaneously (to check for OC stability and thermals).

No exact benchmark is officially given, but a rule of thumb is that below 100 mV is awesome, up to 300 mV is fine, below 500 mV is tolerable, and above that is mostly crap.

A bit long-winded, but hopefully I managed to shed some light on the matter.
__________________
Careful what you wish for... You just might get it.
Reply With Quote
The Following 2 Users Say Thank You to McSteel For This Useful Post:
-The_Mask- (3 Weeks Ago), quest for silence (3 Weeks Ago)
  #3  
Old 3 Weeks Ago
quest for silence quest for silence is online now
Silencer
 
Join Date: Sep 2013
Location: Italy
Posts: 267
Thanks: 26
Thanked 7 Times in 7 Posts
Default

Quote:
Originally Posted by McSteel View Post
An inexperienced observer
Point taken, and I really appreciate very much, thank you.


Quote:
Originally Posted by McSteel View Post
These transients usually happen during transitions from idle to load (and back)
I guess that in a properly designed power supply there may be some form of "cushioning", something to somehow "isolate" the load from the power supply topology actual response, sort of "reservoir capacitors" to somehow smooth those transients and do not push back them down to the converter/input level (even if, in case the measured transients response should already take into account such a "filter", so that a better response could probably reflect a better "cushion"): whether I thought so, would I be really far from understanding how the PSU actually works with reference to transients? I'm sorry for the inevitable inaccuracy.


Quote:
Originally Posted by McSteel View Post
A bit long-winded, but hopefully I managed to shed some light on the matter.
For sure, so that I'm reading something more about right now.
Thanks a lot for your thoughtfulness.
Reply With Quote
  #4  
Old 3 Weeks Ago
quest for silence quest for silence is online now
Silencer
 
Join Date: Sep 2013
Location: Italy
Posts: 267
Thanks: 26
Thanked 7 Times in 7 Posts
Default

Quote:
Originally Posted by quest for silence View Post
I guess that in a properly designed power supply ...snipped out... I'm sorry for the inevitable inaccuracy.
Well, my thought was rather naive, but after some reading now I'm better understanding your explanation (or so I think: I've found lots of paper on how to design a resonant LLC converter, but few information about performance analysis and transient loads.)

However, one of my big mistake was thinking that the larger capacitors of the more powerful units was one of the primary reasons of their usually better behaviour: as probably you already said, McSteel, the answer actually was more straight and simple, a more powerful unit starts those tests at a higher power level, when the converter was already working "better".

So, summarizing, if I'm not wrong again: that transient response is sort of checking the unit voltage regulation but with dealing with big spikes and not at constant loads. Some units seem to hold on similar values in both scenarios, some other else slip in transient loads testing, the first ones are the better.
Loosing regulation dynamically should interfere on how CPU/MB/GPU VRM works, so resulting somehow in a loss of their efficiency/effectiveness, and probably in some more heat wasted on those circuitries (if not in overall instability).

Still I'm not able to directly compare HardOCP and TPU findings because those look like very different (TPU values in mV seem always way lower, often falling under the above described "awesome" category), but that's something I can live with (at least, while I'm still trying to better my knowledge).
Reply With Quote
  #5  
Old 3 Weeks Ago
McSteel's Avatar
McSteel McSteel is offline
Naughty
 
Join Date: Nov 2010
Location: North Disturbia
Posts: 1,440
Thanks: 18
Thanked 89 Times in 69 Posts
Default

Quote:
Originally Posted by quest for silence View Post
I guess that in a properly designed power supply there may be some form of "cushioning", something to somehow "isolate" the load from the power supply topology actual response, sort of "reservoir capacitors" to somehow smooth those transients and do not push back them down to the converter/input level (even if, in case the measured transients response should already take into account such a "filter", so that a better response could probably reflect a better "cushion"): whether I thought so, would I be really far from understanding how the PSU actually works with reference to transients? I'm sorry for the inevitable inaccuracy.
Any real voltage source will drop it's output voltage when asked for a lot of current, this is because of it's internal resistance. There indeed is some (considerable, actually) "cushioning" in the secondary stage of any decent PC PSU, but it's energy storage capacity is very limited so it cannot pull the voltage up for too long.

The primary capacitors even though they can store large quantities of electrical energy, simply "don't care" what the output voltage is - they act like a burst current source (just like batteries do). This leaves only the primary as an active regulation provider in case of transient loads.

Quote:
Originally Posted by quest for silence View Post
Well, my thought was rather naive, but after some reading now I'm better understanding your explanation (or so I think: I've found lots of paper on how to design a resonant LLC converter, but few information about performance analysis and transient loads.)

However, one of my big mistake was thinking that the larger capacitors of the more powerful units was one of the primary reasons of their usually better behaviour: as probably you already said, McSteel, the answer actually was more straight and simple, a more powerful unit starts those tests at a higher power level, when the converter was already working "better".
Mostly correct, though the better active regulation (under dynamic loading) comes from lower internal resistance of higher-powered PSUs. This stems from the fact that you can't use (relatively) high resistance parts to deliver lots of power because their thermal dissipation will be hard (or impossible) to manage.

Quote:
Originally Posted by quest for silence View Post
So, summarizing, if I'm not wrong again: that transient response is sort of checking the unit voltage regulation but with dealing with big spikes and not at constant loads. Some units seem to hold on similar values in both scenarios, some other else slip in transient loads testing, the first ones are the better.

Loosing regulation dynamically should interfere on how CPU/MB/GPU VRM works, so resulting somehow in a loss of their efficiency/effectiveness, and probably in some more heat wasted on those circuitries (if not in overall instability).
Correct, that is pretty much it, in a nutshell.

Quote:
Originally Posted by quest for silence View Post
Still I'm not able to directly compare HardOCP and TPU findings because those look like very different (TPU values in mV seem always way lower, often falling under the above described "awesome" category), but that's something I can live with (at least, while I'm still trying to better my knowledge).
They don't use the same methodology, but the relative differences between their samples are directly comparable and related, so it's not like they're completely useless. One could possibly even determine the proportion ratio between [H] and TPU results, and I don't think it would deviate more than 10% between different pairs of samples (of the same PSU model).
__________________
Careful what you wish for... You just might get it.
Reply With Quote
  #6  
Old 3 Weeks Ago
quest for silence quest for silence is online now
Silencer
 
Join Date: Sep 2013
Location: Italy
Posts: 267
Thanks: 26
Thanked 7 Times in 7 Posts
Default

Quote:
Originally Posted by McSteel View Post
This leaves only the primary as an active regulation provider in case of transient loads.
I beg your pardon, but I haven't understood: did you mean "primary caps"?


Quote:
Originally Posted by McSteel View Post
This stems from the fact that you can't use (relatively) high resistance parts to deliver lots of power because their thermal dissipation will be hard (or impossible) to manage.
Indeed: nonetheless I guess that nothing but cost reasons may prevent the manufacturers to use the same low resistance parts into lower power units.
That also doesn't give me enough clues to let me understand a bit better the behaviour of some units, like the 850W Scythe Chouriki tested some years ago on TPU (yes, I always look to low noise units), which had a loose regulation on 3.3V on static loads, while surprisingly stronger transient response on the same rail.


Quote:
Originally Posted by McSteel View Post
They don't use the same methodology, but the relative differences between their samples are directly comparable and related, so it's not like they're completely useless. One could possibly even determine the proportion ratio between [H] and TPU results, and I don't think it would deviate more than 10% between different pairs of samples (of the same PSU model).
I can't talk about those things advisedly, at all, but as said it doesn't look that simple to me.

I will set aside the different assessments made by the reviewers on the following figures.

Take for instance the P1200 Platinum, reviewed by crmaris on last 28th july. The transient response of the 12V rail in the 20% test exhibits a very modest 63mV drop, bettering in the 50% down to 48mV. The very same unit, tested on 1st july (so I think that's the same revision) by [H] (did Spectre do that?), showed a drop in Test 1 (so the 20% one) of about 310mV, and even greater on Test 2, 380mV.
On the contrary if we give a look to their reviews of the Silverstone NightJar NJ520, on TPU the values I read are respectively 86mV/101mV, while on [H] the values are 280mV / 310mV.
So, granted my lack of knowledge and just about the loaded 12V rail figures, I register these "facts" : the smaller unit did better than the larger on [H], but the opposite on TPU, so that results look like diverging in magnitude and ratio (on TPU: +37%/+110% while on [H]: -10%/-18%). Results are diverging also in another way: in the P1200 review, response improve in 50% test for TPU, while it worse for [H], but the same things didn't happen for the NJ520 (as transient response in the "easier" test "worse" in both reviews).
If I take another review done by both, the Antec TPC 750W shows these values: TPU 78mV/90mV - [H] 350mV / 330mV. Comparing those figures with the previous one, I am not able to find a somehow evident correlation: TPU values are similar and slightly better, about 10%, than the NJ520, [H] ones are less similar, up to 25%, and of the opposite sign, worse than the fanless Seasonic/Silverstone PSU. Moreover, while on TPU on Test2 the TPC750 performed slightly worse in the "easier" 50% test, [H] logged a more "canonical" behaviour, with the 50% Test 2 response slightly better than Test 1 response. That's just due to sample variance? I can't help.

More probably that not three tests do not necessarily indicate a trend, but definitely I'm a bit puzzled while looking for any correlation (I mean, it's not that simple).
Reply With Quote
  #7  
Old 3 Weeks Ago
McSteel's Avatar
McSteel McSteel is offline
Naughty
 
Join Date: Nov 2010
Location: North Disturbia
Posts: 1,440
Thanks: 18
Thanked 89 Times in 69 Posts
Default

Quote:
Originally Posted by quest for silence View Post
I beg your pardon, but I haven't understood: did you mean "primary caps"?
No, I meant the whole primary stage of the PSU, meaning the part after APFC (the primary caps are actually APFC caps) and before the main transformer.

Quote:
Originally Posted by quest for silence View Post
More probably that not three tests do not necessarily indicate a trend, but definitely I'm a bit puzzled while looking for any correlation (I mean, it's not that simple).
To be honest, the correlation I mentioned was more of a guess than a factual statement...

However, the differences might lie in something even simpler than sample variance: oscilloscope settings and resolution. Since transient V-droop is measured peak-to-peak like ripple/noise, a 'scope with better resolution and lower time divisor could "catch" a short-lived intense spike, and thus show a much larger drop. Specifying and perhaps normalizing the duration of the measured drop might be interesting and even beneficial as a usable universal benchmark in this regard...
__________________
Careful what you wish for... You just might get it.
Reply With Quote
The Following User Says Thank You to McSteel For This Useful Post:
quest for silence (3 Weeks Ago)
  #8  
Old 3 Weeks Ago
quest for silence quest for silence is online now
Silencer
 
Join Date: Sep 2013
Location: Italy
Posts: 267
Thanks: 26
Thanked 7 Times in 7 Posts
Default

Quote:
Originally Posted by McSteel View Post
No, I meant the whole primary stage of the PSU, meaning the part after APFC (the primary caps are actually APFC caps) and before the main transformer.
So mainly the SMPS controller and mosfets, that is where the topology actually matters: eventually everything in its right place, or so it seems. Thanks.


Quote:
Originally Posted by McSteel View Post
the differences might lie in something even simpler than sample variance: oscilloscope settings and resolution. Since transient V-droop is measured peak-to-peak like ripple/noise, a 'scope with better resolution and lower time divisor could "catch" a short-lived intense spike, and thus show a much larger drop.
As said, even if I don't understand the reasons (is there any quoted reviewer out there? Is there anything you would like to add about your relevant methodology and setup?), it's something that I could live with for some time, at least as long as the relative differences will be somehow consistent.
Reply With Quote
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT -4. The time now is 01:30 PM.


Powered by vBulletin® Version 3.8.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.