Aragorn wrote:

I dont mind the PSU failing as such, but if it fails in a manner which damages other components then thats no good.

The worst "working" PSU I found using my makeshift PSU tester so far has 2V p-p ripple on +5V, giving 5V with overlaid circa 50kHz 2V AC, not spikey but almost sinusoid. The worrying thing is that using normal DC voltmeters it showed a seemingly perfect 5.02V output, and the tester system (mobo, graphics card and drives) actually worked with that.

But this big ripple eventually will destroy all caps (and more in consequence) of the system.
Occasionally one sees mobos with all electrolytics bulged, no matter of size and manufacture.
This is the consequence of using such a seemingly "working" PSU.

retardware wrote:

The worst "working" PSU I found using my makeshift PSU tester so far has 2V p-p ripple on +5V, giving 5V with overlaid circa 50kHz 2V AC, not spikey but almost sinusoid. The worrying thing is that using normal DC voltmeters it showed a seemingly perfect 5.02V output, and the tester system (mobo, graphics card and drives) actually worked with that.

This is why it's good to have a true RMS meter with decent bandwidth (and do some testing in AC mode to pick up the ripple), or test using a scope.

Makes sense, i have a scope and can certainly check for ripple.

Still i'm unsure on the best approach to take. Do i test for ripple and if good just ignore? Or do i pre-emptively re-cap anyway? Or do i just buy some nice modern PSU's?

The modern PSU's seem tricky. This old 250w Dell PSU for instance manages 20A on the 5v rail, whereas a typical modern 350w might only do 15A on the 5v rail. Infact even some much higher rated supplies from some vendors still have less than 20A on the 5v rail!

Aragorn wrote:

Makes sense, i have a scope and can certainly check for ripple.

Still i'm unsure on the best approach to take. Do i test for ripple and if good just ignore? Or do i pre-emptively re-cap anyway? Or do i just buy some nice modern PSU's?

The modern PSU's seem tricky. This old 250w Dell PSU for instance manages 20A on the 5v rail, whereas a typical modern 350w might only do 15A on the 5v rail. Infact even some much higher rated supplies from some vendors still have less than 20A on the 5v rail!

Only a few ATX power supplies are still available new from factory with ATXv1.3 compliant load distributions. That's what you'll need for any beefy Athlon/Dual P3 builds. The Startech ATXPower300 is the only one I would have confidence in.

As for testing, I'd suggest going by the ATX spec. This is from v0.9, but the current spec retains the same maximums for the rails that still exist.

https://paginas.fe.up.pt/~asousa/pc-info/atxp … _pow_supply.pdf

A good PSU should be compliant across the full range of operation, so check it unloaded first, then with a system load hooked up.

It's always recommended to test the ripple on older PSU's. Elecrolytic capacitors usually have a good tolerance for 10-15 years if they are good quality. Beyond that, ripple starts to creep upward as the elecrolytics inside degrade, puke out, or dry up.

You can test the 3.3V, 5V, and 12V rails by loading the power supply to 50% or so (which is what you should run a power supply at) and put an o-scope or true RMS multimeter on each of those outputs to see if they meet AT or ATX specifications.

Does anyone know if these older supplies are stable with unloaded rails? IE if i hook up say just the 12v to a dummy load, will the PSU be happy that the other rails are unloaded? Or am i going to have to find an extra dummy load for testing?

Aragorn wrote:

Does anyone know if these older supplies are stable with unloaded rails? IE if i hook up say just the 12v to a dummy load, will the PSU be happy that the other rails are unloaded? Or am i going to have to find an extra dummy load for testing?

By standard, they shouldn't require any load to stabilize. AT to ATX adapters usually load 3.3V a bit because AT doesn't use that voltage. Without something on that rail, group regulated designs will face a cross-load.

Aragorn wrote:

... if i hook up say just the 12v to a dummy load, will the PSU be happy that the other rails are unloaded? Or am i going to have to find an extra dummy load for testing?

You need load to see if the electrolytics still have sufficient capacity on each rail that you are interested in.
Furthermore it is well-known that AT class (and many other) PSU don't work well without minimum load (say, ca. 1`amp).

Edit:
By the way, using the AC range of digital voltmeters to find out ripple is questionable.

As I found that "Levicom" PSU had that ripple I mentioned, I first threw it into my e-junk basket. But then I reconsidered, as its ripple was unusual, that sinusoid one instead of spikes. I re-hook it into the PSU tester and did multimeter AC tests.

I did test with five different multimeters. Two 7106-based ones didn't show anything. One showed 10.5V AC where the oscilloscope showed 1.2Vpp. A Chinese 3 3/4 digit multimeter with unknown specs showed 2V AC. On 2Vpp another multimeter, HP3455A (specs say up to 1MHz AC supported) said 0.6V AC.

Now I am curious to re-test to see what my multimeters show with another PSU that has a different ripple pattern, short but high spikes.

retardware wrote:

Edit:
By the way, using the AC range of digital voltmeters to find out ripple is questionable.

I agree, it's not the ideal tool, but sometimes you have to get on with what you've got.

retardware wrote:

Two 7106-based ones didn't show anything.

Garbage meters for this use.

retardware wrote:

One showed 10.5V AC where the oscilloscope showed 1.2Vpp. A Chinese 3 3/4 digit multimeter with unknown specs showed 2V AC.

Also garbage for this purpose.

retardware wrote:

On 2Vpp another multimeter, HP3455A (specs say up to 1MHz AC supported) said 0.6V AC.

The 3455A specifies measurement on input signals up to 1mhz, so it's not enough to measure noise as per standard, but probably enough to give a ripple measurement with reasonable accuracy. The meter used needs to measure real RMS values at a decent bandwidth, as close to 20mhz as you can get.

retardware wrote:

Now I am curious to re-test to see what my multimeters show with another PSU that has a different ripple pattern, short but high spikes.

Faster ripple and noise needs more bandwidth, although some will likely be attenuated by reactance of the load and the power leads. The HP will probably be best, but I suspect it's accuracy will suffer compared to the scope.

IMO, RMS measurements only mean something if the ripple is at least somewhat sinusoidal. You're only going to get the real picture with a scope, and using a meter to measure something that it can't express with its limited display capabilities is worse than pointless -- it's potentially misleading. Better than nothing? Ehh... yes, until it gives you a false sense of confidence in something that isn't as healthy as it looks on a digital readout. If that's all you have, then it beats not measuring at all, but only somewhat.

I re-cap all my older PSUs. About P4-era and before is my cut-off today. But, that's partly because I am far too lazy to build an adequate test rig and run every PSU through the gauntlet. I would rather just get in there, clean it, re-cap it, re-fan it, and know that it's as good as it'll get. Combined with selective tastes in manufacturers with a reputation for quality, I feel that's a reasonable stance. It certainly doesn't guarantee none of the supplies I'm using are defective or uniquely awful, but you have to draw the line on risk somewhere, and that's where I've chosen to do it.

This is a choice everyone has to make for themselves, taking into account your confidence in your own ability to choose appropriate replacement parts and do the work without damaging anything. I'm more confident in my skill than in 20-year-old capacitors, so I re-cap.

SirNickity wrote:

I re-cap all my older PSUs. About P4-era and before is my cut-off today. But, that's partly because I am far too lazy to build an adequate test rig and run every PSU through the gauntlet. I would rather just get in there, clean it, re-cap it, re-fan it, and know that it's as good as it'll get. Combined with selective tastes in manufacturers with a reputation for quality, I feel that's a reasonable stance. It certainly doesn't guarantee none of the supplies I'm using are defective or uniquely awful, but you have to draw the line on risk somewhere, and that's where I've chosen to do it.

Fair enough, although equally important (you're doing it anyway when you recap) is to visually inspect for failed semiconductors and cooked transformers. I don't feel the need to re-cap everything (my Delta AT supply has outlived some countries). Much depends on initial quality and how conservative it's engineering margins are.

SirNickity wrote:

IMO, RMS measurements only mean something if the ripple is at least somewhat sinusoidal. You're only going to get the real picture with a scope, and using a meter to measure something that it can't express with its limited display capabilities is worse than pointless -- it's potentially misleading. Better than nothing? Ehh... yes, until it gives you a false sense of confidence in something that isn't as healthy as it looks on a digital readout. If that's all you have, then it beats not measuring at all, but only somewhat.

I re-cap all my older PSUs. About P4-era and before is my cut-off today. But, that's partly because I am far too lazy to build an adequate test rig and run every PSU through the gauntlet. I would rather just get in there, clean it, re-cap it, re-fan it, and know that it's as good as it'll get. Combined with selective tastes in manufacturers with a reputation for quality, I feel that's a reasonable stance. It certainly doesn't guarantee none of the supplies I'm using are defective or uniquely awful, but you have to draw the line on risk somewhere, and that's where I've chosen to do it.

This is a choice everyone has to make for themselves, taking into account your confidence in your own ability to choose appropriate replacement parts and do the work without damaging anything. I'm more confident in my skill than in 20-year-old capacitors, so I re-cap.

Do you swap EVERY electrolytic cap? Or just try to ascertain which are the DC bus caps and replace those?

Looking inside this AT supply for instance there are a pair of 200v 220uf caps on the mains side, and then a couple 2200uf and some 220uf around the DC output. but theres also a selection of smaller stuff littered around the middle of the board.

My personal opinion on this: I swap every cap when recapping, simply because leaving old caps in place makes it more likely that I would have to recap again in the
near future and just doing everything in one sweep seems to be more time conserving.

I replace all of them, big asterisk. As well as the rectangular film cap you usually find at the AC input, and the X/Y ceramic disc caps. I don't usually bother replacing the ceramics sprinkled throughout the circuit unless they look dodgy. If it's exposed and not a nightmare to replace, I'll do the filter caps on the AC inlet itself -- where applicable. If the inlet is an integrated can with the caps inside, I'll let it slide -- it's probably a little obsessive to worry about those.

Now here's the confessional part. I have a few supplies that use just a crap ton of 0.1 to 10uF electrolytics. The IBM PS/2 supply was one of those. I didn't get in there and replace ALL of them, because some were on vertical boards I probably would have had to remove to get access to them all. The little ones don't usually fail, at least not on supplies where the bigger ones are in reasonably good shape still -- indicating quality parts and good engineering margins. If they're relatively easy to replace, I will, just because why not -- if I'm already in there anyway, might as well get it done.

Here's the "before" reference shot from the Tandy 1000 RL HD PSU I did this last weekend. Circled parts are the ones I replaced. The inlet was in a can.

Caps on the primary side are usually not a failure point anyway. For designs that don't have it by default, a filter can is a great retrofit that's really easy to implement. Small bypass caps, likewise, don't usually see much ripple and hence don't usually fail.

Your priority for replacement should always be the output pie filters and any buck converters on the motherboard. They always have large amplitude/high frequency ripple in the face.