>The two main caps seemed a little budged on the top, so I changed them, but it didn't remove the ripple.

the big expensive 200-400V caps almost never go bad, replacing them is just waste of money.

A unloaded PSU will not regulate properly, and will often show a lower output voltage with high ripple until is has a decent load.

Try placing a 30-50Ohm load across the 12v rail - while testing it. Keep mind that this will generate about 1/4 to 1/2 watt of heat.

If you can't find a suitable resistor, any old 12v incandescent bulb would do (like a taillight)

Also keep in mind that on some PSUs the 12v rail requires the 5v rail to be "proper" ... you might have to load it a well.

Usually the loads don't have to be "big" just enough to draw a few dozen ma is usually enough to get reasonable readings.

The main problem with the PSU’s AT is that it is an AT.
In other words, it was designed by the ‘hive mind’ and manufactured who knows where.
With power supplies of this type, the +12V line is very often not connected to the feedback loop for regulation at all.
Therefore, for the measurements like suggested above, it needs to be properly loaded – at least 20 watts on the +5V line and at least 2–3 watts on the +12V line.

fix_metal wrote on 2026-10-07, 23:34:

I picked up the infamous AT PSU and tried to service it. Looking at it does not have any suspicious leaky/smoked component.

Turning it on with a multimeter attached it's giving ~ 10.6VDC through the yellow line ( +12VDC).

Ripple looks quite small, all switching-mode PSUs have some ripple. You need to load it to fully test it and a single CD-ROM unit might not be enough.
Usually these AT PSUs regulate 5V line and 12V one is just scaled to that with resistor network. Did you check the 5V, is that in spec or also low?
- If low look for a small potentiometer on the PCB, it is used to adjust the 5V and 12V will follow. The negative 12V is not very precisely regulated, it mostly mirrors 12V, and -5V (if present) is quite nice due to being regulated with a standard linear 7905 or something of the sort.
- If 5V looks good it's also possible the scaling resistor network drifted a lot. Now these should be more precise 1-2% resistors that usually don't drift all that much but with cheap ATs you never know what was "optimized" to save some money and increase profit margins.

The pot might be bad too, poor connection or went open, sometimes turning it will fix it. As for loading the PSU for testing, automotive light bulbs for 12V and 6V are great for that. A rule of thumb is a cheap switching-mode PSU will not regulate properly below 10% of it rated output - but about 10W on both 5V and 12V should do the trick. In fact you can use 12V bulb for 5V line too, just keep it below some 20W or else it might not even light up and while cold it will draw way more current than you'd expect.

Deunan wrote on Yesterday, 07:27:

Usually these AT PSUs regulate 5V line and 12V one is just scaled to that with resistor network.

No, this appeared in the ATX power supply.
If regulation is derived from two power supply lines, it is determined by the currents through the resistors, rather than by the voltages themselves.
Consequently, ‘weighting factors’ are introduced for each voltage. For early ATX systems, these are 90–95 % for +5 V and 5–10 % for +12 V.
This is not acceptable for AT systems, because for most of their history, the entire AT computer was powered directly from +5V, including the x286 CPU itself.

Therefore, in AT power supplies, the weighting for the +5V voltage is most likely 100%, whilst that for +12V is 0.
In other words, the +12V is not regulated at all – this is irrelevant for powering the 5.25-inch floppy drive motor.

Deunan wrote on Yesterday, 07:27:

Ripple looks quite small, all switching-mode PSUs have some ripple.

Yeah, 20-30mV peak-peak ripple at some tens of kHz is nothing, and to be expected.

rasz_pl wrote on Yesterday, 01:08:

>The two main caps seemed a little budged on the top, so I changed them, but it didn't remove the ripple.

the big expensive 200-400V caps almost never go bad, replacing them is just waste of money.

Sometimes even they go bad. I recapped a few AT power supplies a couple of weeks ago, and in one of them a high-voltage cap had bulged a bit.

fix_metal wrote on 2026-10-07, 23:34:

At the start of this post I wrote "apparently": what I suspect is, in fact, that during my first testing round of the PSU I hooked up the molex the other way around.

I'm not sure that's possible. Those molex connectors are keyed with one set of corners slightly rounded. It would be extremely difficult to plug it backwards, if not impossible without breaking the connector.

As the others have mentioned, you need to load the PSU more properly in order to see the output voltages regulate.
Incandescent or halogen 12V auto (car) bulbs work well for this purpose. A 20W bulb will usually draw about 5W (1 Amp) if put on the 5V rail... which is usually enough of a load for the 5V rail to keep these old PSUs happy. The 12V rail can do with much less of a load - a large 100 Ohm 2 to 3 Watt resistor should be fine (beware it will run scorching hot after half a minute or so, but that's normal.)
I wouldn't go higher than 35 Watt bulbs, especially if halogen type, because anything more usually has a very low cold resistance and runs the risk of tripping the PSU's short-circuit protection.

rasz_pl wrote on Yesterday, 01:08:

the big expensive 200-400V caps almost never go bad, replacing them is just waste of money.

Agreed.
However, key word here is "almost never". I have seen them go bad on a few very rare occasions. Mostly, it's the extremely crappy no-name brands and when they are way too underspecced for the PSU. But if you run into a PSU like that, you're probably better off using a different PSU anyways.

As for 400V-rated caps - now those do go bad a lot more often, but only in PSUs with APFC... which wasn't really a thing one would ever see in the era of AT PSUs, so nothing to worry about. Mentioning this, because I have seen a few AT PSUs with a single 400V cap instead of 2x 200V caps in series. But those AT PSUs used flyback (discontinuous) design rather than the more traditional continuous designs.

shevalier wrote on Yesterday, 08:08:

Therefore, in AT power supplies, the weighting for the +5V voltage is most likely 100%, whilst that for +12V is 0.

I wouldn't go concluding that's the case here, though.
Plenty of AT PSUs do take feedback from the 12V rail. It's just not weighted as heavily as the 5V rail, like you said. But it's not at "0%" weight either.
In any case, this is the reason for the low voltage on the 12V rail with no load on the PSU.
It's not necessary to load the 5V rail with at least 20W as you stated. You're forgetting that many 286 and 386 computers' TDP hardly broke 10 Watts of power consumption with the optical and hard drives removed. And in fact, most competent PSU manufacturers usually recommended at least 1 Amp of load on the 5V rail and anywhere from 200 to 300 mA and up on the 12V rail for the voltage rails to be close to spec.

"Almost never" isn't applicable. It depends on the age, and the load, temperatures, etc it saw throughout its service life. Or the opposite, if it was never used and sat in storage - that is bad as well. Anytime an electrolytic capacitor is 10yrs+ old, it is suspect to begin with. When issues are encountered, they should all be checked at a minimum. Or if you're following a proper EPRI preventative maintenance plan, they get replaced every 15 years regardless, to stave off the inevitable problems with electrolyte chemistry changes (and evaporation) or dielectric breakdown.

Sometimes they last far longer than expected, sometimes they die far earlier. Generalizations really can't be made because it impacts all brands and years of production. (outliers like the capacitor plague era were especially bad of course) This is the reason we follow EPRI templates within operational technology environments, especially within critical infrastructure.

momaka wrote:

I'm not sure that's possible. Those molex connectors are keyed with one set of corners slightly rounded. It would be extremely difficult to plug it backwards, if not impossible without breaking the connector.

I know. But who knows what I did that night.

I wouldn't go higher than 35 Watt bulbs, especially if halogen type, because anything more usually has a very low cold resistance and runs the risk of tripping the PSU's short-circuit protection.

Yeah, I just noticed that. I put a 55W 12V halogen lamp, and it worked until the PSU stopped after some moments. 11.6V at the multimeter until circuit protection kicked in. The PSU is rated 8A on the 12V line.

Then I switched to 5W resistors I found spare somewhere. With these, and a proper load on the 5V line, I get almost a pure sin signal on the 12V line, with ~10.5V at the multimeter by then.
I then plugged the CD drive and it stayed at ~10.5V.

I pulled an HDD which I don't care much about with rating of 12V 0.5A and 5V 0.65A. Then, the big surprise. The voltage would simply fail to stick stable, it would flap between 10.9V and 11.3V. the oscilloscope was giving somewhat worse ripple readings as the HDD's heads couldn't properly set. It span, but I could hear the heads keeping trying over and over. I've made a video for reference [0].

I suppose it's not even standing some proper load on +12V? I'm not entirely sure I'm not falling down a rabbit hole here.

[0] https://youtube.com/shorts/dFgEdRYT0bg?is=EcoZkcuSV42HB4rR

momaka wrote on Yesterday, 20:17:

I'm not sure that's possible. Those molex connectors are keyed with one set of corners slightly rounded. It would be extremely difficult to plug it backwards, if not impossible without breaking the connector.

oh its possible all right, my gf fried a HDD back in ~1999 this way by trying hard enough until molex socket complied :]

wiretap wrote on Yesterday, 21:44:

Anytime an electrolytic capacitor is 10yrs+ old, it is suspect to begin with.

Depends on application, there are vintage tube radios/TVs, or computers from the seventies with perfectly fine Electrolytic caps. There are also procedures for reforming/reconditioning those old caps.

wiretap wrote on Yesterday, 21:44:

Or if you're following a proper EPRI preventative maintenance plan, they get replaced every 15 years regardless, to stave off the inevitable problems with electrolyte chemistry changes (and evaporation) or dielectric breakdown.

EPRI as in The Electric Power Research Institute? That would mean heavy duty industrial power electronics, not some tiny capacitors living their best life in zero ripple nicely ventilated environment.

fix_metal wrote on Yesterday, 23:40:

Yeah, I just noticed that. I put a 55W 12V halogen lamp, and it worked until the PSU stopped after some moments. 11.6V at the multimeter until circuit protection kicked in. The PSU is rated 8A on the 12V line.

it shouldnt shut down with that load as long as there was also load on 5V

fix_metal wrote on Yesterday, 23:40:

Then I switched to 5W resistors I found spare somewhere. With these, and a proper load on the 5V line, I get almost a pure sin signal on the 12V line, with ~10.5V at the multimeter by then.

its only 60mV, thats fine
how about you measure 5V line, is it sagging too? its very plausible output caps are dried - those actually DO see a ton of high ripple load unlike the big main line ones.

rasz_pl wrote on Today, 00:05:

Depends on application, there are vintage tube radios/TVs, or computers from the seventies with perfectly fine Electrolytic caps […]
Show full quote

wiretap wrote on Yesterday, 21:44:

Anytime an electrolytic capacitor is 10yrs+ old, it is suspect to begin with.

Depends on application, there are vintage tube radios/TVs, or computers from the seventies with perfectly fine Electrolytic caps. There are also procedures for reforming/reconditioning those old caps.

wiretap wrote on Yesterday, 21:44:

Or if you're following a proper EPRI preventative maintenance plan, they get replaced every 15 years regardless, to stave off the inevitable problems with electrolyte chemistry changes (and evaporation) or dielectric breakdown.

EPRI as in The Electric Power Research Institute? That would mean heavy duty industrial power electronics, not some tiny capacitors living their best life in zero ripple nicely ventilated environment.

Yes, reforming is sometimes possible if they have been sitting, however it should be done periodically.

But no, you're incorrect about EPRI.. their PM templates for electrolytic capacitors go down to board level small components. I implement those template recommendations on a daily basis at work, in fact it is required by law.