jonnyGURU Forums

jonnyGURU Forums (http://www.jonnyguru.com/forums/index.php)
-   PC Power Supply Discussion (http://www.jonnyguru.com/forums/forumdisplay.php?f=3)
-   -   Hold-up time explained (http://www.jonnyguru.com/forums/showthread.php?t=13273)

-The_Mask- 05-05-2016 05:24 AM

Hold-up time explained
 
Yeah another topic about the hold-up time. Why? Because I got the feeling some really don't understand what it is, what it does and what the problem is. ;)

What is the hold-up time?
It's the time in milliseconds that the PSU can work without without getting any power, measured at full load.

How does it work?
The primary capacitor of the PSU can store enough energy for the PSU to work some milliseconds after the power is cut.

How long should it last?
According to the ATX specification the PSU should deliver at least 17ms full power with voltages still inside the ATX spec.

But at least 1ms before the output voltages of the PSU go out of spec the PSU should drop the Power OK signal.

Power OK signal?
Yes Power OK signal, that's a wire in the 24 pins ATX connector that tells the motherboard that the power of the PSU is stable and that the PC can start or should shutdown.

What's the problem then?
Some PSU don't drop the Power OK before the output voltages of the PSU go out of spec. This isn't good because it puts a lot of stress on the components of the PC that are powered by the PSU and now getting voltages that are outside of spec.

Is it really a problem?
Well that depends if you experience a lot of problems with power outage where you life, it could be a problem. But if that is the case a UPS is maybe something that you should buy.

But there is something else you should know!
Hold-up time is measured at full load! Almost no one loads his of her PSU till full load. And even if someone does, the chance that PSU is working at full load if the power fails, is almost zero.

So that means that hold-up time actually should last a lot longer in real life conditions. It even could mean that the power OK signal drops before the output voltages go out of spec.

So it actually isn't a problem?
For most people not no. ;)

So please don't make this a huge problem, because it isn't one. It simply isn't done right according to the ATX specification.

For more information and tests see the reviews from Aris on TechPowerUp and TomsHardware:
http://www.techpowerup.com/reviews/?...=25&order=date
http://www.tomshardware.com/articles...power-supplies

marcos669 05-05-2016 05:40 AM

Great and straight to the point info, I looked for something like this a few days ago but couldn´t find much

GI_Joe 05-05-2016 06:08 AM

Your PG stuff is wrong. Will explain tomorrow.

Philipus II 05-05-2016 06:11 AM

Here in Germany hold up time is only relevant for professional use. I don't remember any power loss in the last two years here in my flat. In other countries power losses are normal. In my 6 month in spain the power was out at least once per month.

sith'ari 05-05-2016 06:16 AM

Quote:

Originally Posted by -The_Mask- (Post 124716)
.
.
.
So that means that hold-up time actually could last a lot longer in real life conditions. It even could mean that the power OK signal drops before the output voltages go out of spec.

The magic word is "could".
"Could" doesn't specify certainty. "Could" means that either it will or it will not !!
And as i said multiple times in the past, i'm not a person that likes risks, and the 50%-50% chance of the "could" word, is a serious risk percentage in my opinion.
So, if it 1) "could mean that the power OK signal drops before the output voltages go out of spec.", it also 2) could mean that the Power_OK could drop much much later after the voltages get out of spec!!!
So the conclusion is that we need even more thorough testing and NOT to abandon them completely and pretend that the issue doesn't exist at all !!!

-The_Mask- 05-05-2016 06:18 AM

Quote:

Originally Posted by sith'ari (Post 124724)
The magic word is "could".

Changed the first could to should. ;)

Quote:

and the 50%-50% chance of the "could" word, is a serious risk percentage in my opinion.
As I was trying to explain, the chance is probably very small in real life.

quest for silence 05-05-2016 06:20 AM

Quote:

Originally Posted by -The_Mask- (Post 124716)
Some PSU don't drop the Power OK before the output voltages of the PSU go out of spec.

And why then UVP doesn't kick in, in that case? Is it too slow?


Quote:

Originally Posted by Philipus II (Post 124723)
In other countries power losses are normal. In my 6 month in spain the power was out at least once per month.

Italy is much more similar to Spain.

-The_Mask- 05-05-2016 06:23 AM

Quote:

Originally Posted by quest for silence (Post 124726)
And why then UVP doesn't kick in, in that case? Is it too slow?

UVP is mostly set to kick in by a much bigger drop.
Quote:

Italy is much more similar to Spain.
The Netherlands is more like Germany, it almost never happens.

sith'ari 05-05-2016 06:28 AM

Quote:

Originally Posted by -The_Mask- (Post 124725)
Changed the first could to should. ;)
As I was trying to explain, the chance is probably very small in real life.

Oh come on.....!!!
I can't resist with what you are writing!!! Again, probably doesn't specify certainty, and i'm not a person that likes risks and ....blah blah blah you know the rest :lol:!!!

quest for silence 05-05-2016 06:31 AM

Quote:

Originally Posted by -The_Mask- (Post 124727)
UVP is mostly set to kick in by a much bigger drop.

1.2V are not enough? So which is the (general) UVP usefulness/goal?


Quote:

Originally Posted by -The_Mask- (Post 124727)
The Netherlands is more like Germany, it almost never happens.

Pares cum paribus facillime congregantur. :D

-The_Mask- 05-05-2016 07:07 AM

Quote:

Originally Posted by sith'ari (Post 124729)
Oh come on.....!!!
I can't resist with what you are writing!!! Again, probably doesn't specify certainty, and i'm not a person that likes risks and ....blah blah blah you know the rest :lol:!!!

There is only one thing certain in life and that is that you die. ;)

Quote:

Originally Posted by quest for silence (Post 124730)
1.2V are not enough? So which is the (general) UVP usefulness/goal?

That should be more then low enough for a 3,3V, 5V and 12V UVP trigger point. You can normally find the UVP trigger point in the datasheet of the secondary protection IC. If it doesn't work it probably means it isn't implemented the right way.
Quote:

Pares cum paribus facillime congregantur. :D
Geen idee wat dat betekent. ;)

rafal_iB_PL 05-05-2016 07:43 AM

Quote:

But if that is the case a UPS is maybe something that you should buy.
It should be pointed out that UPS operation is related to hold-up time, at least for cheaper ones. Cheaper UPS units have non-negligible transfer time when switching to batteries, so hold-up time needs to be longer than that transfer time. If you have online UPS, it doesn't matter, as those run from batteries all the time.

-The_Mask- 05-05-2016 08:38 AM

Yes I could at that, tomorrow.
First :beer: :D

quest for silence 05-05-2016 09:01 AM

Quote:

Originally Posted by -The_Mask- (Post 124731)
There is only one thing certain in life and that is that you die. ;))

you = royal/generic you!?! :lol:


Quote:

Originally Posted by -The_Mask- (Post 124731)
That should be more then low enough for a 3,3V, 5V and 12V UVP trigger point. You can normally find the UVP trigger point in the datasheet of the secondary protection IC. If it doesn't work it probably means it isn't implemented the right way.

If it should be more than enough, should we think that the Leadex Gold 550W had not properly implemented UVP?


Quote:

Originally Posted by -The_Mask- (Post 124731)
Geen idee wat dat betekent. ;)

Sorry mate: it means Pares cum paribus facillime congregantur :D


Quote:

Originally Posted by -The_Mask- (Post 124736)
First :beer: :D

It's not friday: is there some party there? ;)

jonnyGURU 05-05-2016 11:58 AM

Stickied...

-The_Mask- 05-05-2016 01:40 PM

Quote:

Originally Posted by quest for silence (Post 124737)

It's not friday: is there some party there? ;)

Liberated from the Germans today. :D

quest for silence 05-05-2016 03:33 PM

Quote:

Originally Posted by -The_Mask- (Post 124745)
Liberated from the Germans today. :D

So you were no more congregantur. :beer: :D

GI_Joe 05-05-2016 10:00 PM

Quote:

Originally Posted by -The_Mask- (Post 124716)
But at least 1ms before the output voltages of the PSU go out of spec the PSU should drop the Power OK signal.

No, the PSU's eventually existing OVP and/or UVP protections would cause it to shut own into Latch mode. If no such protections, the PC's components would receive the out of spec voltages for as long as they can without shutting down/blowing up themselves. For the MBU and GPU, their VRMs would try to regulate the out of spec voltages for as long and much as they can.

Quote:

What's the problem then?
Some PSU don't drop the Power OK before the output voltages of the PSU go out of spec. This isn't good because it puts a lot of stress on the components of the PC that are powered by the PSU and now getting voltages that are outside of spec.
The PG signal is not a protection. It is a "not that simple" "HELLO" from the PSU to the MBU, which has to answer that with a "I am awake". THEN only, the PSU releases its output in PS_ON mode, no matter how good or bad that output is, that is monitored by protections.
Your explanation mixes too much the purposes of PG & PS_On and their causes and effects. There is a whole lot more to it than I have time to write here.

Quote:

But there is something else you should know!
Hold-up time is measured at full load!
:rolleyes: Reality is different. The important thing here is the overall AC capacitance in use and not some paper spec hold-up time claim.
Quote:

Almost no one loads his of her PSU till full load. And even if someone does, the chance that PSU is working at full load if the power fails, is almost zero.
Gamers can do and it simply depends. Most users do not have long-enough brown-outs, but some do. short brown-outs are common in the USA for example. And don't get me started on India's power grid :D

Quote:

So it actually isn't a problem?
For most people not no. ;)
Again, it depends if you want to risk a restart.

What happens in a brown-out is that the PSU must "power" the system from the AC power stored its main capacitor. IF that power is exhausted, the system restarts and THEN PG signal hello & back from PSU & MBU is started and the outputs released into PS_On.


Long story short, Don't mix signals and protections and their criss-cross causes and effects :D

sith'ari 05-06-2016 04:59 AM

What i want you to make us clear, is your opinion, if whether the "Power_OK signal-Hold UP time tests" that are being conducted by Aris & Pavel, are complete/thorough/useful as they are, or if they should probably measure something else as well ?? (*again, according to your opinion):confused:

GI_Joe 05-06-2016 05:19 AM

De-rating + UPS compatibility, if UPS are common in your region

sith'ari 05-06-2016 05:35 AM

Quote:

Originally Posted by GI_Joe (Post 124759)
De-rating + UPS compatibility, if UPS are common in your region

UPSs are very common in Greece, but what exactly do you mean by de-rating? can you specify a little?

GI_Joe 05-06-2016 07:51 AM

Hold-up time shrinks with heat. To test real-life hold-up time test at 35-45C to get a realistic case ambieent temp. HardOCP tests with 45C, no matter the derating, which is mean against lower rated PSUs. You could stagger that to be fairer: hold-up at 25C+30C+35C+40C+45C.
Users would get a clearer impression on the need for an additional UPS or not. This also varies by the countries grid reliability.
And manufacturers with a larger capacitance headroom would get rewarded by the results.
Plus the likely compatibility with stepped/simulated/true Sinewave UPS. Nobody tests all 3 types.

GI_Joe 05-06-2016 07:55 AM

Aris is coming to Computex? I did not see him scheduled by our EU marketing :D
Korinna is back from maternity leave next week.

ehume 05-06-2016 08:05 PM

Quote:

Originally Posted by -The_Mask- (Post 124745)
Liberated from the Germans today. :D

Your time must be very ahead of my time + you are celebrating VE Day on May 7 and not May 8. Regardless, the 6-year war in Europe concluded around now. Congratulations. If my father were alive, he would lift a glass with you.

turkey3_scratch 05-06-2016 09:01 PM

^^I was wondering the same thing. So UVP and PWR_OK drop voltage threshold are different values, and different circuits, correct?

It doesn't seem right that UVP would be set so low. One would think UVP would kick in around 11.4V. Does UVP kick in at a certain voltage, or does it measure a drop in voltage?

Also, realistically, even if the PWR_OK signal is dropped when the voltage is below 11.4V, let's say 10.8V, it's not quite the end of the world. If you look at page 23 of the ATX specification http://www.formfactors.org/developer...rm_Rev_1_2.pdf

It reads:
1. At +12V1DC peak loading, regulation at the +12V1DC and +12V2DC outputs can go to
±10%.
2. At +12V2DC peak loading, regulation at the +12V1DC and +12V2DC outputs can go to
±10%.


The spec also says in various locations:
12V2DC supports processor power requirements and must have a separate current limit and provide 13 A peak
current for 10 ms; minimum voltage during peak is > 10.8 VDC


So the ATX specifications states that the positive voltages can be +-10% during peak load for a small amount of milliseconds. So, taking that knowledge and applying it to AC_LOSS to PWR_OK holdup time, it should not be a problem if the 12V rail is dropped at 10.8V and if it's under 11.4V for a duration of 10ms (or even higher wouldn't really matter). Then again, if the PSU is under a lighter load, the PWR_OK signal gets dropped even later, so it could be under 11.4V for a longer duration then. Also, I think it's best if people avoid the word "holdup time" in general and specify exactly what holdup time they are talking about. Let's look at the different definitions/types of holdup times:

AC_LOSS to PWR_OK holdup time
AC_LOSS to DC_LOSS (VOLT_OUT_SPEC) holdup time
defined as duration of time PSU remains on during AC loss

All of them are not the same thing, yet they are all grouped under the term "holdup time". The first would be the duration of time when there is AC loss to when the PWR_OK signal is cut. The second would be the duration of time until the 12V rail goes out of spec (I'm not if 5V and 3.3V play a role). And the last would be the duration of time from AC loss until the PS_ON signal initiates and executes.

Stefan Payne 05-07-2016 09:59 AM

Quote:

Originally Posted by quest for silence (Post 124737)
It's not friday: is there some party there? ;)

Yeah, thursday was 'Vatertag' here and many people did not work...
Though there is some exeption for some areas (like the one I live in)...

cypherpunks 12-20-2016 06:14 PM

Quote:

Originally Posted by GI_Joe (Post 124754)
Quote:

Originally Posted by -The_Mask- (Post 124716)
But at least 1ms before the output voltages of the PSU go out of spec the PSU should drop the Power OK signal.

No, the PSU's eventually existing OVP and/or UVP protections would cause it to shut own into Latch mode. If no such protections, the PC's components would receive the out of spec voltages for as long as they can without shutting down/blowing up themselves. For the MBU and GPU, their VRMs would try to regulate the out of spec voltages for as long and much as they can.

The PG signal is not a protection. It is a "not that simple" "HELLO" from the PSU to the MBU, which has to answer that with a "I am awake". THEN only, the PSU releases its output in PS_ON mode, no matter how good or bad that output is, that is monitored by protections.
Your explanation mixes too much the purposes of PG & PS_On and their causes and effects. There is a whole lot more to it than I have time to write here.

No, in addition to its role during power-on, the PWR_OK signal is required by ATX specification to be pulled low a minimum of 1 ms before the DC rails go out of spec , i.e. fall to 95% of their nominal values. This is parameter T6 in Table 11 on page 18 of the ATX12V 1.31 specification. Figure 2 states the 95% figure twice.

1 ms seems like not much time, but it's enough time to finish a flash memory write cycle.

Parameter T5 in the same table says that PWR_OK must stay high for 16 ms after loss of AC power.

That said, this is a spec that almost nobody tests, so I don't expect good compliance to it. Since nobody can trust the falling edge of PWR_OK, nobody uses it, and because nobody uses it, nobody implements it properly...

Quote:

Originally Posted by turkey3_scratch (Post 124782)
  1. At +12V1DC peak loading, regulation at the +12V1DC and +12V2DC outputs can go to ±10%.
  2. At +12V2DC peak loading, regulation at the +12V1DC and +12V2DC outputs can go to ±10%.

In v1.31 this changes that to ±5%. (Below table 5 on p. 14.)

GI_Joe 12-20-2016 09:28 PM

Cypherphunks, still, the original discussion made it sound like the PG Signal is the protection itself, but it is just the signal of states of PSU and MBU.
And that signal does not always come as written as in the DG.
Also a lot is unspecified still by the DG. The sequence of the Rise time is the main reason for PSU incompatibilities. MBU A is ok if 3.3V & 5V shoot up at the same time, MBU B wants 3.3V 5ms earlier but not stronger than 5V, etc..
The DG is just a target to shoot at. It is not the Bible/Koran/Torah/etc.. Even Intel does not follow it itself fully. "REQUIRED" is for ... :D

cypherpunks 12-21-2016 02:24 AM

Quote:

Originally Posted by GI_Joe (Post 134096)
Cypherphunks, still, the original discussion made it sound like the PG Signal is the protection itself, but it is just the signal of states of PSU and MBU.

I may have misunderstood. Your English is somewhat stilted, and it's hard to be sure I understand. You are right that PWR_OK only applies in the case of AC power loss. In the case of malfunction or overload (such as OVP or OCP), it is not required to give warning (or, indeed, to go low at all).

(I also agree that what the specification requires is irrelevant to PSU manufacturers, who only care about what the market requires.)

crmaris 01-08-2017 04:32 AM

I think this belong here:

The worst case scenario for testing overshoot is to remove (and apply of course) power with the PSU under full load. The same goes of course for hold-up time with only exception that you need to find the right phase when the bulk cap won't be fully charged (worst case scenario that is, for the hold-up time). You don't just simply remove AC power when ever you want but you have to do it in the right phase of the incoming AC voltage. The duration of the AC power removal doesn't play a role in finding the actually hold-up time, what matters the most is WHEN you will remove power. This is if you want to find the REAL hold-up time.


The worst case scenario for a PSU with a APFC converter is when its bulk cap(s) has just delivered load and begin to charge. This moment is when the incoming power, from the APFC converter, starts to increase and becomes equal to the power level that the PWM stage requires.

Considering that P = E x I = Vrms x Irms = Vrms x Vrms/R(load) = Vrms^2 / R(load) and the power that the PFC delivers instantaneously equals P(t) = V(t)^2 / R(load). By using the Vrms values (in order to be able to use the DC power equation) you have V(t)= Vrms= Vpk / sqrt(2). This is close to 45 and 135 degrees on the phase of the incoming AC signal.


http://img.techpowerup.org/170101/capture1087.jpg

Personally I use my AC source to achieve the right phase since it provides me the option to start/stop the PSU in an exact degree of a phase (not any more because the super expensive AC source is highly stressed). However I should note that the above doesn't apply on all PSUs. So I have to try different phase angles in order to record the lowest possible hold-up time. Needless to say that this is a painful (for my AC source) and long procedure. I made an arduino device back in the day with programmable phase for power cut/apply which worked like a charm.


Another interesting fact about hold-up time. Does anybody ever wondered why it is 17ms and not for example 15 or 19ms?

Well according to the standard forums set by the Server System Infrastructure (SSI) Forum the minimum hold-up time at fully rated output power is one cycle. So for 50Hz it should be 20 ms while with 60 Hz it is 16.66 ms or 17 ms. This means that with 50 Hz normally it should be higher. Personally I use 230V and 50Hz in order to keep compatibility with my previous database, although at some point I should probably run two tests, one with 115V and one 230V (but the day only has 24h).

jonnyGURU 02-25-2017 03:36 PM

It is kind of silly that MOST of the world uses 230V, but standards are defined in the vacuum of a 115V country. :D

Surely cost savings can't be the only reason why PSUs don't come with caps that are capable of 17ms.. or even 20ms.. hold up time. There shouldn't be an effect on efficiency as well, should there?

Has anyone here experimented with replacing a bulk cap with one of larger capacitance and tested to see what the effects are?

I might have to do that some day.

crmaris 02-25-2017 04:34 PM

longer hold-up time. I did this in a Corsair PSU :) Don't remember which though.. However I didn't run the whole efficiency tests. Normally efficiency should be a little lower because higher capacity needs higher amperage to charge the cap.

jonnyGURU 02-25-2017 04:48 PM

Yeah. That's what I was thinking. Same problem with adding caps to the secondary to reduce ripple. You also kill your efficiency.

ehume 02-25-2017 05:23 PM

In shopping for PSU's (admittedly something I won't be doing for a while) I'll gladly sacrifice some efficiency for adequate holdup time and lower ripple. I just wish manufacturers were a little more transparent in their tradeoffs for (efficiency vs. holdup time vs. ripple) vs. cost (therefore price). If the customer is seeing that a higher price gets a better PSU then he/she can make a reasoned decision.

jonnyGURU 02-25-2017 05:59 PM

Quote:

Originally Posted by ehume (Post 136750)
In shopping for PSU's (admittedly something I won't be doing for a while) I'll gladly sacrifice some efficiency for adequate holdup time and lower ripple. I just wish manufacturers were a little more transparent in their tradeoffs for (efficiency vs. holdup time vs. ripple) vs. cost (therefore price). If the customer is seeing that a higher price gets a better PSU then he/she can make a reasoned decision.

Unfortunately, efficiency is easily quantifiable to the end user due to 80 PLUS, their independent testing and the logos they give out. Sure, reviewers will expose if a PSU has poor hold up time and/or ripple suppression. The majority of buyers don't read reviews, but do know what the 80 PLUS logo is and assumes it's a marque of quality.

Stefan Payne 02-25-2017 09:04 PM

Quote:

Originally Posted by jonnyGURU (Post 136742)
Has anyone here experimented with replacing a bulk cap with one of larger capacitance and tested to see what the effects are?

Don't have the equipment for that, sadly :(
But that seems to be an interesting question.

Nuckles_56 02-25-2017 10:59 PM

It obviously can be done, just look at the seasonic primes with their long holdup times and their high efficiency but it is just expensive

Internet 02-26-2017 12:37 AM

Quote:

Originally Posted by jonnyGURU (Post 136742)

Has anyone here experimented with replacing a bulk cap with one of larger capacitance and tested to see what the effects are?

Behemot did this in his review of the Corsair CX550M. However, he also used caps with lower ESR than the original one. As the result, he got slightly less ripple and more efficiency.

http://www.hardwareinsights.com/wp/c...esign/7/#Bonus

On a sidenote, I personally prefer to pay more attention to inrush current.

jonnyGURU 02-26-2017 12:43 AM

Quote:

Originally Posted by Internet (Post 136763)
Behemot did this in his review of the Corsair CX550M. However, he also used caps with lower ESR than the original one. Seems this also was responsible for slightly less ripple and more efficiency.

http://www.hardwareinsights.com/wp/c...esign/7/#Bonus

I see that now.

Also, our engineers told me that cost of the cap is about the only gating factor.

crmaris 02-26-2017 03:38 PM

using caps with lower ESR won't necessarily do good. In some cases it might also bring worse results because even ESR plays a role in a PSU by helping in preventing unwanted oscillations. I am referring to the secondary side's filtering caps.


All times are GMT -4. The time now is 01:24 PM.

Powered by vBulletin® Version 3.8.7
Copyright ©2000 - 2018, vBulletin Solutions, Inc.
Copyright © JonnyGURU.com 2000 -