PC PSU shutdown condition with bad mobo caps

Excessive ripple keeps the 'Power good signal' set to false.

But whether or not that signal is involved in the chain of events (normally it drives reset and maybe not all designs necessarily use it that way) what is sensed on the PSU outputs is what I'm thinking about. Would it be voltage ripple and not just current all the way back in the PSU even if the PSU caps are good?

And are PSU caps likely to be damaged too after use with a bad mobo?

No. If they are good, the PS will just shut down. The CPU PS generates its own control signals. They no longer run from 3.3V. The voltage is lower, and at a lot higher currents. That's why the quality of the electrolytic caps is so critical. I see people complain about the caps being too close to the CPU, but 1.8V @ 30A doesn't allow for long traces.

The purpose of the caps that are close to the CPU is to reduce the ripple on the processor power line to tolerable levels. The processor can go from zero to 40 amps at anywhere from zero (ground bounce) to GHz rates. The capacitors have to smooth all that out. Basically, they form an energy storage system to deliver power during the high current spikes to the power hungry CPU.

When the ESR (equivalent series resistance) of the caps increases due to electrolyte loss caused by overheating, less energy is available to the CPU during high current peaks. The result is lots of ripple and noise on the power supply line. Eventually, this gets the attention of the "power good" line to the power supply, which shuts down to protect the MB and CPU.

One irritating problem is the dominoes effect on multiple MB filter capacitors failing. When the ripple voltage is low across the capacitors, there is little ripple current through the capacitors. Capacitors only conduct during the time the voltage changes across them. When the voltage is pure DC, no current is drawn. They also only get warm when there's ripple current through the capacitor.

When one capacitor (usually nearest the CPU) fails, the ripple current trhough the others increases slightly. That means that the internal heat generated in the other caps increases. When the next capacitor fails, the ripple current and internal heating of the remaining capacitors again increases, result in even more heating. Eventually, the few remaining caps can't handle the self-heating, overheat, bulge, and die. This is why you want to replace ALL the capacitors in a parallel string when one of them fails. Measuring the ESR of such a parallel string is impossible.

Note that better MB's use polymer capacitors instead of electrolytics.

at

I'm going to expose my ignorance here: are there not also small value capacitors in parallel, to take care of the GHz spikes?

Yes. There are ceramic chip caps found on the bottom of the PCB, directly under the CPU. Short lead length required for high frequencies. None of the external CPU pins actually see GHz range signals as they are divided down before they hit the various buses. However, the rise times are well into the GHz (picosecond) ranges, which can have components in the GHz range, and must be bypassed lest they radiate RF and attract the attention of the FCC.

PG wire?

Jamie

That's the point of my question. So it sees increased ripple voltage and not just ripple current. That means the PSU caps are overwhelmed and might suffer some damage too.

I've known the power-good signal since the XT, but it used to be just for releasing RESET after the voltages came up. It isn't a very sensitive detector of excessive ripple current since the latter has to be worse than the load regulation spec to trigger it. In old designs it probably wouldn't have detected ripple at all since it had its own filtering to create a delay after power came up.

But you can't substitute them for electrolytics, can you?

The computer's power supply is on the other side of the CPU supply, but a ittle ripple can get into the 3.3V or 5 V rails and cause erratic operation. If the power supply is failing, it's usually just a coincidence unless it was a really crappy design.

They are also called solid polymer electrolytics. Substitution will depend on the max & min ESR allowed in the circuit.

Sorta. If any part of the noise or ripple on any of the power supply lines goes below the defined threshold, the power good line will drop.

See Fig 7 on Pg 25 and 3.3.1 on Pg 26.

Nope. There's quite a bit of DC resistance between the power supply and the ripple source, which is the CPU. The ripple is there not because the power supply is producing the ripple or noise, but because the CPU is NOT a constant DC load. The power consumption of the CPU varies radically what's being computed, power saving features, data transfer speeds, etc. The filter caps on the motherboard are actually protecting the power supply from too much ripple produced by the CPU.

Things have changed a bit. See the timing specs in the ATX12V Power Supply Design Guide in the first URL.

If it meets the ATX design specs, *ANY* voltage excursion below the allowable design tolerances on any of the voltage lines, should initiate a shutdown. In general, the voltage tolerances at +/-5%. See

Yep. The old designs also didn't sense excursions of the 117VAC power supply input.

Yes, you can, but only if the ESR is low enough. Many motherboards have a mix of solid polymer and electrolytic caps with the polymer handling the high ripple current and high temperature filtering, and the electrolytics handling the less critical filtering. The issue is primarily cost. Solid polymer caps are about 5-10 times as expensive as electrolytic.

I don't know if the Power_Good signal from the PSU is just a power-on reset= signal for the motherboard or if the mobo is designed to shut down if that= signal goes false. OTOH the mobo doesn't drive the Power_Good signal; the= only thing it does drive is the Power_On signal on the green wire.

Motherboards have several voltage regulators -- CPU, AGP or PCI-E, and memo= ry, and I'm guessing they're designed to shut down the power in case of ove= rload because it could be too dangerous not to (tons of amps for creating a= rcs and fires); I can't say for sure because I haven't looked for this prot= ection on many mobos, but my ancient PC-2007 had a "U" shaped bar to sense = CPU current.

I don't know if PSU caps are damaged by bad caps on the mobo, but I doubt i= t's due to voltage but only current, and some reviewers have seen PSU caps = and other PSU components fail when loaded to the PSU's full rated power wit= h a testing machine -- see JonnyGuru.com , HardOCP.com , HardwareSecrets.co= m, which disect what they review, failed or not.

En el artículo , Tom Del Rosso escribió:

Yes.

En el artículo , Tom Del Rosso escribió:

Yes. All motherboards I've seen with failed caps in the processor voltage regulator cct have also had failing caps in the PSU.

Replacing the PSU as well as recapping the board is what makes repair uneconomic.

I don't see that combination of failures in the PCs I rebuild. Power supplies are usually easier to recap than a motherboard. Good 105C low ESR caps are cheap in quantity. Jeff posted one link. Here are some others that I've bought from:

I buy some Panasonic FM series in bulk from

That's not what I've been seeing. I don't do too much recapping, but replacing the MB caps usually solves the problem. I have seen bad power supplies, but MB cap failures seem more common. One common characteristic I see in failing ATX power supplies is that they tend to have only a few physically small capacitors in the DC output filter section. The ATX supplies that are crammed full of big electrolytics seem to survive nicely.

Ummm... it's easier to recap a power supply than a mother board. The PS comes out easily, is easily disassembled, usually has a single sided low density PCB, and is easy to test after recapping. The MB requires removing lots of hardware, replacing many more small caps, is a multi-layer PCB, and takes more time to test.

What makes MB recapping uneconomical is that rather large number of caps that usually need replacement. Testing the caps on the MB is problematic because many are in parallel. The only sure way is to replace anything that might fail. That's usually about 12-15 caps on a typical MB. I can frequently buy a used MB on eBay for less than what I charge for my time. I still do recapping because I detest creating eWaste, but I don't make much money on recapping.

The OST brand caps in that voltage regulator in my ECS K7S5A Pro failed, as= did the PC-Tur Elite caps (CEC International?) in that regulator in my ECS= P4S5A, but the caps in their PSUs were fine, and all were Panasonic, Rubyc= on, or Nichicon.

I don't see any mention of current so it isn't 'sorta' triggered by ripple voltage.

That would matter if the supply had sense lines, but PC supplies don't. The voltage is sensed at the supply's output.

But I understand you to mean that although the caps in the PSU are working harder, the discharge is small enough that they won't be heated unless they were defective as well.

For the high end 130W Xeon over 200A: