Question about 85c caps vs. 105c caps

[Wester547]

Re: Question about 85c caps vs. 105c caps

PCBONEZ, that's intriguing. But that doesn't mean bad PSU caps slowly kill HDDs or anything, do they? Because it was debunked on Tech ARP's HDD myth page that PSUs would slowly kill a HDD. It was stated they would "instantly" kill HDDs by sending a surge to them and frying them in an instant (or by just not being able to supply enough power, so the HDDs would fail to spin up). I don't think the PSU was the reason the WD800JB in that machine died, at least not in entirety.

[PCBONEZ]

Re: Question about 85c caps vs. 105c caps

Tech ARP's HDD myth page is full of shit then. [Whoever they are...]
-
With Ripple the data gets corrupted in the interface IC chips.
Those are in the controller on the mobo and it's counterpart on the drive's PCB.
It doesn't happen in the motor and it doesn't happen at the heads or disc.
The heads are recording what they were given to record, it gets corrupted on it's way there.
-
The IC chips run on +5v and/or +3.3v and the motor runs on +12v [unless it's a laptop drive].
-
AND ~ Ripple does not directly cause power surges - at least not like that.
If they said that they are clearly clueless.
If you have THAT much ripple the system wouldn't even boot.
.
.
Excessive ripple could cause the voltage regulation in the PSU to wander around and and over time that could burn-out the motor controller chip who's job it is to keep the discs spinning at a constant speed.
It wouldn't be some sudden failure though.
-
So yes, bad PSU caps can slowly corrupt or even kill a drive.
Usually what happens is a chip somewhere gives out though.
.
.
. FYI
Interface chips are like language translators.
~~~~
CPU sends data to chipset.
Chipset converts 'CPU talk' to 'PCI BUS talk' and sends to on-board drive interface controller.
~~~~
Controller converts 'PCI BUS Talk' to 'SATA Talk'. [or 'IDE Talk'] and sends to the on-drive controller.
~~~~
Drive controller converts 'SATA Talk' to storable 'Drive Talk' and sends to the heads/disc.
....
....
Excessive Ripple could garble any of those *translations* resulting in corrupted files.
.
.
[A Chipset is mostly just a whole bunch of translators packed into one or two chips.]
.

[Wester547]

Re: Question about 85c caps vs. 105c caps

It seemed informative to me, but I'm not incredulous about your argument either. If you wish to read:

http://www.techarp.com/showarticle.aspx?artno=84

So then, bad PSU caps can cause a drive to fail in every way? It can cause a drive to develop reallocated sectors, pending sectors, offline uncorrectable sectors, S.M.A.R.T. errors, etc? Because before it died, the WD800JB was going down completely. HDD Sentinel reported it to be at 7% health and it had 256+ pending and reallocated sectors, numerous spin up issues, CRC errors, offline uncorrectable sectors, etc. I'm certain it had over 15,000 errors in the S.M.A.R.T. log. I know S.M.A.R.T. doesn't mean everything for a drive and most certainly does not report much about mechanical failure of a HDD, just problems with the media itself. But I do not think it should be wholly ignored because it can foretell impending failure. Not only that, but the WD800JB would click lots before it died...

But it's definitely possible the PSU is at fault. Everything else seems fine. I have another question - I have read that the components of a system often works hardest when first it turns on (which is why computers tend to fail upon boot). Does this apply for capacitors too?

[mariushm]

Re: Question about 85c caps vs. 105c caps

No, the power supply wouldn't cause problems like your WD drive manifested. Those are simply caused most likely by physical faults in the disc platters.

No...think of it like this.

Most integrated processors, chips and so on have some tolerance and can work at a voltage a bit different than the exact one they were designed for, for example a 5v chip would probably work between 4.7 and 5.3v.
Most devices have capacitors and other components in them to smooth out the voltage and protect the chips against power fluctuations, spikes and so on.

Even a perfect power supply, it can't supply a device with perfect 5v, because this is how the power supplies are designed. The do however try to smooth the voltage it sends as much as possible using capacitors or feedback loops inside themselves. The devices you have don't trust that the power supply will smooth out the voltage as much as it should, so they do an extra step and use those extra components to smooth everything they receive further.

When the capacitors or something else in the power supply start to go bad, the ripple will increase on the output and the quality of the power the psu delivers decreases. BUT, the extra components on your device will take over and smooth out the power a bit so it will be within acceptable range, and the device will still be happy. Eventually though, the quality of power from the power supply can become so bad this circuitry won't be able to smooth it enough to keep the device working (but if you're here, probably the motherboard and other components won't start)

The problem is integrated chips can have a certain degree of quality, some come out as worse quality but still within design specs and some come out as much better than design specs - just like you have processors that can be overclocked or not.

So some chips that were designed to work with 5v could work within a higher range of voltages and are less sensitive to ripple and spikes that go over the circuitry that tries to smooth out what comes from the power supply. On the other hand, some chips tolerate some ripple and spikes but in time, some transistors inside can get damaged or "weakened". In the worst case, the chip simply blows up.

When the chip is subjected to more ripple than it can handle, even after the circuitry near it tried to smooth out the voltage, it's possible that when it does some calculations inside, it returns a 1 instead 0 or the other way around, so this corrupt data gets sent to the disk heads which write it to disk. Next time when some application reads the data, expected 0 but was read 1, and you get blue screens or crashes.

It can also happen the other way around - the disk heads read the data correctly from platters, but when the chip translats the data to the IDE or SATA transmission protocol, errors can appear due to the chips having issues.

I have read that the components of a system often works hardest when first it turns on (which is why computers tend to fail upon boot). Does this apply for capacitors too?

Generally, motors needs much more power when they start, they have to overcome the friction inside and other things that are much too complicated to explain here. So for example, a 12v regular fan inside your computer is designed to use up to 0.2A (so 0.2x12 = 6 watts) but just when it starts up and reaches its normal rotation speed, it could use 3-6 times as much.
So, if you have a lot of fans and/or hard drives (motors inside the drives work the same), it's possible to demand more power than the power supply can deliver right away. If the power supply is going bad, it could even kill itself because of this.

Think about it this way. Let's say a 400w power supply can deliver 16A on 12v and you have six hard drives and 3 fans (cpu fan, back panel, hdd fan)
Each hard drive has a 12v @ 0.4A motor, but needs 1.2A to start (about 4s)
Each fan has a 12v @ 0.15A but needs about 0.7A to start (about 2s)
The processor will use about 30w when it starts, which is generated from 12v with about 85% efficiency, so that's about 3A

At boot, you then have 5 drives x 1.2A + 3 x 0.7A + 3A = 12A, which is pretty close to 16A. After boot, they all use only 5x0.4 + 3x0.15 + 2-5A = 4-7A.

That's why most RAID cards and some BIOSes have a tricked called "staggered start" or something like that, which turns drives one after another instead of all of the same time.


Another reason why computers often fail after a short time after boot is heat-cold cycles. As the solder on the pcb traces heats up and cools down, it's possible to get fractures or cracks in the solder, if the soldering wasn't done right or there's too little solder.
If some component is then not making proper contact, something can get damaged.

[PeteS in CA]

Re: Question about 85c caps vs. 105c caps

The HN/HM were supposedly overfilled so chemistry had nothing to do with it.
No room for expansion causes leaks and then they dry-out.

That is the word I've heard as well. Someone forgot to reset the machine for a new, slightly smaller, case size than the previous run of parts. And we're talking fairly tiny cases. The water-based electrolyte used in Nichicon Hx (UCC KZx and all Rubycon low impedance caps) expands considerably with heat - personal observation, can't cite an online source for that. So slight overfill + heat = caps that vent and spill their electrolyte (which is highly conductive electrically).

IC chips don't work well with noise in their DC power.

When your Vcc has excessive ripple/noise voltage on it, the chips will be unable to discern signal from noise, leading to errors and data corruption. Especially if a chip has data inputs that use hysteretic input circuits to square-up data signals that may have a little noise or slightly sloppy edges.

[PeteS in CA]

Re: Question about 85c caps vs. 105c caps

Most integrated processors, chips and so on have some tolerance and can work at a voltage a bit different than the exact one they were designed for, for example a 5v chip would probably work between 4.7 and 5.3v.

In addition, logic chips have specified ranges for how data inputs will interpreted. Under a certain voltage threshold is guaranteed to be "seen" as a "0"; over a certain other voltage threshold is guaranteed to be "seen" as a "1". Between those thresholds is an area of uncertainty in which chip performance is unpredictable. If the noise on the Vcc causes data to go into the uncertainty range, or even the range where a "1" might be seen as a "0" (or vice versa), well, bleep happens. Maybe data gets corrupted; maybe error detection/correction recognizes the error and time gets wasted trying to correct the error.

[PCBONEZ]

Re: Question about 85c caps vs. 105c caps

It seemed informative to me, but I'm not incredulous about your argument either. If you wish to read:

http://www.techarp.com/showarticle.aspx?artno=84

As suspected,,, journalists, not techs...
They asked someone and repeated it but they don't really know what they are talking about...

So then, bad PSU caps can cause a drive to fail in every way?

What you are missing is that ripple causes corrupted files on drives that are perfectly good.
There aren't any bad sectors involved.
The files were bad -before- they got to the drive.
~ So then later...
You are doing something and windows tells you that you have corrupted files - or it won't boot to windows because some file is bad or missing.
~ The automatic assumption is the drive is going bad.
~ Not always true....

It can cause a drive to develop reallocated sectors, pending sectors, offline uncorrectable sectors, S.M.A.R.T. errors, etc? Because before it died, the WD800JB was going down completely. HDD Sentinel reported it to be at 7% health and it had 256+ pending and reallocated sectors, numerous spin up issues, CRC errors, offline uncorrectable sectors, etc. I'm certain it had over 15,000 errors in the S.M.A.R.T. log. I know S.M.A.R.T. doesn't mean everything for a drive and most certainly does not report much about mechanical failure of a HDD, just problems with the media itself. But I do not think it should be wholly ignored because it can foretell impending failure. Not only that, but the WD800JB would click lots before it died...

mariushm covered that pretty well.

But it's definitely possible the PSU is at fault.

You asked earlier if a bad PSU could damage a drive.
~ The answer is yes.
Did it damage THAT particular drive?
By what you described, probably not.
.

[PCBONEZ]

Re: Question about 85c caps vs. 105c caps

When your Vcc has excessive ripple/noise voltage on it, the chips will be unable to discern signal from noise, leading to errors and data corruption. Especially if a chip has data inputs that use hysteretic input circuits to square-up data signals that may have a little noise or slightly sloppy edges.

In addition, logic chips have specified ranges for how data inputs will interpreted. Under a certain voltage threshold is guaranteed to be "seen" as a "0"; over a certain other voltage threshold is guaranteed to be "seen" as a "1". Between those thresholds is an area of uncertainty in which chip performance is unpredictable. If the noise on the Vcc causes data to go into the uncertainty range, or even the range where a "1" might be seen as a "0" (or vice versa), well, bleep happens. Maybe data gets corrupted; maybe error detection/correction recognizes the error and time gets wasted trying to correct the error.

Yes to all of that.
I think without pics/drawings it might be a little over the OP's head at this point though.
.

[PCBONEZ]

Re: Question about 85c caps vs. 105c caps

Here's some quick-dirty oversimplified drawings that should help get the idea across.
The 'No Ripple' one shows the correct Output from some IC chip.
.
If Data is "0" then volts has to rise to threshold [1.25v] to become a Data "1".
It's where volts is on the clock pulse that matters.
.
If at time T1 the ripple keeps the voltage below the 1.25v the Data won't change to a "1" and you'd get a data error.
If the Data is a file going to the drive you get a corrupted file.
It's not a mis-write so the drive isn't going to try and fix it.
There is nothing wrong with the drive, it just got bad data to save.
.

.

.

[Wester547]

Re: Question about 85c caps vs. 105c caps

Thanks for the replies, everyone. Very, very interesting. I wouldn't say it's over my head - if anything else, I find all these very lucid explanations. And I figured that thermal expansion (or contracting and expanding due to heat cycles, and solder balls and joints cracking) wasn't ever good for hardware, which would explain why components tend to fail when first they start. Thanks for all your patience. I was also wondering - slightly off-topic - but I've read that the spinning up of a mechanical drive is the most arduous task for its spindle motor. Is spinning down also harder on it than just constantly spinning (which is said to be easier on the bearings) ? I mean spinning down alone, such as spinning down and shutting off, not spinning back up. But I also mean spinning down itself, just the task of doing that, not staying spun down. Also, well, since I have a small motherboard in there, there's a bit of a mess of IDE cables.... and one of them is literally touching part of the second RAM module and is kind of close to the two VR capacitors.... that's not at all apt for airflow, right? Also, a fan's ball bearing can go bad and still spin at full speed, correct?

[PeteS in CA]

Re: Question about 85c caps vs. 105c caps

Thanks for the diagrams, PCB. As the saying goes, "A picture is worth ...". The thresholds I would have used, had I gone to that level of detail, would have been the thresholds for old-school +5V TTL. One extra wrinkle to add to your diagram, some noise goes below the 0V level. And if it goes sufficiently negative such spikes can have their own interesting effects. And another wrinkle, if the noise goes sufficiently high in the positive direction, some protective components in the IC become active, protecting the IC, but possibly affecting the speed. Since digital ICs are synchronized by clocks, the ICs' changing state can generate quite a bit of noise. The state change is not instantaneous, and for some 10s or 100s of picoseconds, each data output line can draw quite a bit of current. Large pulses of current plus stray inductances and capacitances can be a very noisy combination.

Significantly different coefficients of thermal expansion can do unpleasant things when different materials touch or are near each other. I recently helped a coworker diagnose and find a glass package rectifier that was probably cracked by being in contact with some epoxy adhesive of with a different thermal expansion characteristic.

Cooling air "likes" a straight, unobstructed path. Nothing's perfect, but it may be worth optimizing the air path to minimize obstruction, especially to critical areas.

Maybe some one with better knowledge could set me straight, but during spin-up the stress is both electrical (high current) and speed change (bearings). During spin-down I would think the stress might be the speed change.

[Wester547]

Re: Question about 85c caps vs. 105c caps

^ So, I'm guessing mechanical disk drives like to spin up, spin at a constant rotation rate for a while, then spin down? Spinning it up then spinning it down a few minutes later and some hours later, shutting it down after it spins down (but doesn't spin back up until next the system is booted) isn't great for the drive? And well, as far as airflow goes, I was just wondering if having part of an IDE (or in this case, a PATA) cable touching part of the second RAM module would hinder airflow significantly, especially since the two capacitors are near it.

[PCBONEZ]

Re: Question about 85c caps vs. 105c caps

Thanks for the diagrams, PCB. As the saying goes, "A picture is worth ...". The thresholds I would have used, had I gone to that level of detail, would have been the thresholds for old-school +5V TTL. One extra wrinkle to add to your diagram, some noise goes below the 0V level. And if it goes sufficiently negative such spikes can have their own interesting effects. And another wrinkle, if the noise goes sufficiently high in the positive direction, some protective components in the IC become active, protecting the IC, but possibly affecting the speed. Since digital ICs are synchronized by clocks, the ICs' changing state can generate quite a bit of noise. The state change is not instantaneous, and for some 10s or 100s of picoseconds, each data output line can draw quite a bit of current. Large pulses of current plus stray inductances and capacitances can be a very noisy combination.

The rise time doesn't have much to do specifically with Ripple corrupting a signal which was the point of the drawing.

Like I said that was quick and dirty. I used MS Paint.
If you wanna do up a prettier more detailed one then go right ahead. {I'll look...}

120mV Ripple is the ATX 12v spec. That's why I chose it.
With 5v signaling 120mV ripple isn't that big a deal because there's a lot of room to the threshold.
- Which is why old Mobo/CPU's like 486 that used 5v signaling for everything didn't need scads of low ESR caps.
5v signaling would be harder to show the Ripple related problem and it isn't used that much on mobos anymore anyway.
AGP was down to 0.8v before it faded out.
PCI can be either 5v or 3.3v but, while add-in cards are often 5v, onboard chips on the PCI BUS are most always 3.3v.
DDR2 uses 1.8v
Chipsets use a sundry of signal voltages down to at least 1.5v
.
I was actually thinking of [a fairly modern] CPU when I picked 1.5v volts.
More or less a random pick of legit values that would show what I wanted to show.
.
.

[PeteS in CA]

Re: Question about 85c caps vs. 105c caps

I was actually thinking of [a fairly modern] CPU when I picked 1.5v volts.

Like I said, old school +5V TTL.

Granted that I went further afield, noise is noise, whether it comes from inadequately filtered P/S ripple, or inadequately decoupled ICs. In some ways the latter can be nastier, as it might not be visible with an oscilloscope.

[PCBONEZ]

Re: Question about 85c caps vs. 105c caps

^ So, I'm guessing mechanical disk drives like to spin up, spin at a constant rotation rate for a while, then spin down? Spinning it up then spinning it down a few minutes later and some hours later, shutting it down after it spins down (but doesn't spin back up until next the system is booted) isn't great for the drive? And well, as far as airflow goes, I was just wondering if having part of an IDE (or in this case, a PATA) cable touching part of the second RAM module would hinder airflow significantly, especially since the two capacitors are near it.

That's a lot of spins. Now my head is spinning.
.
The RPM is controlled by the firmware and there are various schemes in use.
Usually BIOS's have settings that can over-ride some of it.
.
The simplest scheme is to spin the drive up to full speed when it's needed.
Then there is a time-out that spins it down to an idle speed after x-many minutes of not being used.
[That time is what you can usually adjust in BIOS or sometimes in the OS.]
.
With drive bearings and wear you are really worrying about something that isn't worth worrying about much.
It's not like you can swap them for better bearings.
You get what you get.
And they are made of materials that last for eons.
- I think it's Stellite which is a cobalt-chromium alloy.
- Very hard, very heat resistant, used in 'cobalt' drill bits.
You are more likely to blow a chip than a bearing.
.

[PCBONEZ]

Re: Question about 85c caps vs. 105c caps

Like I said, old school +5V TTL.

Granted that I went further afield, noise is noise, whether it comes from inadequately filtered P/S ripple, or inadequately decoupled ICs. In some ways the latter can be nastier, as it might not be visible with an oscilloscope.

Agreed but it's better to explain one thing at a time and the core question was: how can a PSU issue screw up a hard drive.
.

[Wester547]

Re: Question about 85c caps vs. 105c caps

I guess my question more limpidly was, is it better to keep a hard drive constantly spinning, or is it any worse to spin it up for a few minutes (what turning on the system does) then keep it spun down until the system restarts? From what I garner, they're designed to do it either way without consequence. But...

Then there is a time-out that spins it down to an idle speed after x-many minutes of not being used.
[That time is what you can usually adjust in BIOS or sometimes in the OS.

Is this something like advanced power management levels? I thought a hard drive couldn't spin at anything less than its full rotation rate without spinning down. Is speed change more stressing to the hard drive than keeping it spinning at its rotation rate? That is, speed change (spin down) within a few minutes of starting the drive after a while (but again, without spinning it back up).

And I'm guessing hard drive heat doesn't have much on capacitors, right (hard drives that have no fan blowing over them)?

[PCBONEZ]

Re: Question about 85c caps vs. 105c caps

If you leave it spun up it eats more power and stays at a higher temp.
.
Spinning it up and down does cause more wear than constant speed.
And spinning fast is more wear than spinning slow.
And two hairs weigh more than one hair too.
- Point is 'more' doesn't necessarily mean 'significant'.

I usually set mine to spin down after 15-20 minutes.
If I'm not doing anything for that long it usually means I wandered off to do something else.
I do that for power use and heat though, not bearings.
.

[Wester547]

Re: Question about 85c caps vs. 105c caps

But is it bad to say.... start the system, then a few minutes after, spin the hard drive down and keep it spun down (instead of keeping it spinning for the whole duration of use, keep it spinning for a few minutes then spin it down until it spins up again during the next restart)? Is such an abrupt speed change any sort of issue for the bearings and spindle motor? ^^;

[PCBONEZ]

Re: Question about 85c caps vs. 105c caps

Read the previous post again.
.