Meanwhile, I finally got my unit washed, cleaned, and dried. Actually, that was like 2 weeks ago, but I've just been slow as usual and trying to finish a few other projects. Anyways, I'll see if I can do the recap this weekend and then (eventually) do a load test on the PSU to see how it behaves. You were completely right when you said earlier above that these FSP units are just a slight step-up from the gutless wonder PSUs... and even that's a maybe. In particular, I really don't like the low capacity of the input electrolytic caps in my unit (330 uF) - this is what I'm used to seeing in sub-200 Watt PSUs. And there were more stuff, but I'll just leave that for another post here.

analog_programmer wrote on 2025-02-07, 14:49:

Finally I got those chinese TS (Taiwan Semiconductor??) SF1608G double diodes (I ordered 3 pcs just in case... and of course the chinese ali-baba-gipsy-seller sent only 2 pcs instead of 3). At least they look brand new. I measured their Vf with my trusty multimeter in diode mode and it is only 0.435 V for each diode in both pieces. I don't know if this Vf will rise to about 1.7 V under load at high temperatures. This weekend I'll desolder the original SF1603G diodes from the PSU's PCB, so I'll be able to compare their Vf values and change those with these SF1608G.

Yes, it's normal to see a low voltage drop on the diodes with no / low load. The voltage will indeed rise as the current does. Temperature does the inverse, though: lower semiconductor temperature = higher voltage drop and higher temperature = lower drop.

Let's hope the rectifiers are legit too and not just some re-labeled part with lower rating. It's already enough that they didn't send you the correct count. FWIW, the cheap Chinese BJTs on AE, Amazon, Ebay, and etc. are almost always either re-labeled parts or new parts with smaller dies, thus not able to withstand their ratings.
A few years ago, I bought 10x "13009" NPN BJTs in TO-220 case - they type that's often used in half-bridge ATX PSUs. Of course I tested them before installing, looking at the datasheet carefully to make sure I don't exceed the SOA in my tests. What I found is that they couldn't even do half of their rated current. They appeared to *almost* be as good as 13005's, but not quite (more like 13003's, but the fact that they are in TO-220 case probably helped them pull higher current than they would otherwise.)

analog_programmer wrote on 2025-01-29, 14:30:

Well, now you have the opportunity to personally see the problem of these PSU models (PNR = PNF minus PFC reactor) and investigate it 😀

Yup, I see that now.
Two weeks ago, there were two more FSP units at the flea market - one the same ATX-350PNF and the other the one with almost same model, but with active APFC. Unfortunately, I forgot about them by the end of the sale, so didn't buy them... and the reason I didn't buy them right away is because I didn't know the seller and wasn't sure she'd give me a good price. Should have asked, I suppose, but oh well. These are probably gone to scrap heaven now.

analog_programmer wrote on 2025-01-29, 14:30:

IR thermal camera is a costly purchase, but such a tool will be very handy in electronics repairs workshop.

Yeah, that's why I haven't got one yet... but sometimes you buy expensive tools only once and then never look back with regret. 😀
For the time being, I have a type-K meter, so I use that. It's just a nuisance to move the probe around when I want to test different parts, and airflow over the sensor
also affects the temperature readings.

analog_programmer wrote on 2025-01-29, 14:30:

I don't know if it would be appropriate to ask you for diagrams/schematics of two mostly disassembled old cr*ppy PSUs, for which I can't find anything matching their actual PCBs and components (fortunately I have kept their main parts like transformers, inductors, MOSFETs, power diodes, heatsinks etc.). I also have one problematic laptop power supply to repair, but I can't find any circuit diagram for it and the root of its problem.

Sure, not a problem.
I don't actually have too many schematic diagrams (barely any, really). But with good pictures or the device in front of me, I usually make my own. When it comes to older (and especially gutless) ATX PSUs, a lot of them are a copy of each other or nearly so. With laptop adapters, it's not so much the case. But still, even these aren't that different.

momaka wrote on 2025-02-26, 08:17:

Meanwhile, I finally got my unit washed, cleaned, and dried. Actually, that was like 2 weeks ago, but I've just been slow as usual and trying to finish a few other projects. Anyways, I'll see if I can do the recap this weekend and then (eventually) do a load test on the PSU to see how it behaves. You were completely right when you said earlier above that these FSP units are just a slight step-up from the gutless wonder PSUs... and even that's a maybe. In particular, I really don't like the low capacity of the input electrolytic caps in my unit (330 uF) - this is what I'm used to seeing in sub-200 Watt PSUs. And there were more stuff, but I'll just leave that for another post here.

You know, 20-25 years ago these FSP/Fortron PSUs were considered to be "good"... what to say about even cr*ppier chinese junk 😁

momaka wrote on 2025-02-26, 08:17:

Yes, it's normal to see a low voltage drop on the diodes with no / low load. The voltage will indeed rise as the current does. Temperature does the inverse, though: lower semiconductor temperature = higher voltage drop and higher temperature = lower drop.

Usually with rising of the temperature the electrical resistance also rises for most materials. Of course there are exceptions to this rule and since I'm not a specialist in semiconductor components, in this particular case I have to trust you.

momaka wrote on 2025-02-26, 08:17:

Let's hope the rectifiers are legit too and not just some re-labeled part with lower rating. It's already enough that they didn't send you the correct count. FWIW, the cheap Chinese BJTs on AE, Amazon, Ebay, and etc. are almost always either re-labeled parts or new parts with smaller dies, thus not able to withstand their ratings.
A few years ago, I bought 10x "13009" NPN BJTs in TO-220 case - they type that's often used in half-bridge ATX PSUs. Of course I tested them before installing, looking at the datasheet carefully to make sure I don't exceed the SOA in my tests. What I found is that they couldn't even do half of their rated current. They appeared to *almost* be as good as 13005's, but not quite (more like 13003's, but the fact that they are in TO-220 case probably helped them pull higher current than they would otherwise.)

I still can't find enough spare time to gut-out the PSU for a third (or fourth) time. So these new TS SF1608G double diodes are still at hand, but I don't know how can I test them more than just measure their V-drop at room temperature. I don't have any laboratory PSU with 10+ A DC line.

momaka wrote on 2025-02-26, 08:17:

Two weeks ago, there were two more FSP units at the flea market - one the same ATX-350PNF and the other the one with almost same model, but with active APFC. Unfortunately, I forgot about them by the end of the sale, so didn't buy them... and the reason I didn't buy them right away is because I didn't know the seller and wasn't sure she'd give me a good price. Should have asked, I suppose, but oh well. These are probably gone to scrap heaven now.

It's nice to have a trusted electronic "scrap" seller 😀

momaka wrote on 2025-02-26, 08:17:

Yeah, that's why I haven't got one yet... but sometimes you buy expensive tools only once and then never look back with regret. 😀
For the time being, I have a type-K meter, so I use that. It's just a nuisance to move the probe around when I want to test different parts, and airflow over the sensor also affects the temperature readings.

The expensive tools have to be used more often, than just a couple of times. So these IR cameras are not for amateurs like me.

momaka wrote on 2025-02-26, 08:17:

Sure, not a problem.
I don't actually have too many schematic diagrams (barely any, really). But with good pictures or the device in front of me, I usually make my own. When it comes to older (and especially gutless) ATX PSUs, a lot of them are a copy of each other or nearly so. With laptop adapters, it's not so much the case. But still, even these aren't that different.

Ok, I'll send you a PM with a brief explanation and attached pictures in zip-archive when I find time to take pictures of the PCBs and their components. If I don't have at least a basic schematic of a given device, most of the time I don't know what to do, being at amateur level with electronics.

analog_programmer wrote on 2025-02-26, 12:57:

You know, 20-25 years ago these FSP/Fortron PSUs were considered to be "good"... what to say about even cr*ppier chinese junk 😁

Well, they are still not terrible... in fact decently built over all. But in a few select areas, they are sloppy/lousy, which is a shame, because they could have been otherwise good PSUs.

analog_programmer wrote on 2025-02-26, 12:57:

Usually with rising of the temperature the electrical resistance also rises for most materials. Of course there are exceptions to this rule and since I'm not a specialist in semiconductor components, in this particular case I have to trust you.

Well semi-conductors can be semi-logical sometimes. 😁
I forgot what the physical explanation behind that was, but all I remember is that the forward voltage of a P-N junction (of a diode, transistor, etc.) always drops when it heats up. (That said, this doesn't apply to MOSFETs conducting forward, as they don't have a P-N junction in that path.)
As an example, I attached a screenshot of the graph of average forward current vs. forward voltage drop for a UG10DCT (common-Cathode rectifier.)
download/file.php?mode=view&id=213363
First note how at low conducting currents (on the left axis), the forward voltage drop, Vf (bottom axis) is quite low. As the forward current increases, so does Vf. This is why your multimeter reads quite low of a drop - it's not putting any significant forward current through the diode, so Vf is low.
Next, note how as the junction temperature, Tj, is changed from room temperature (25C) to 100C and 125C, Vf "goes back" for the same forward current. For 1 Amp of I_Fa, for example, Vf can be as low as 0.55V at 125C Tj or as high as 0.75V at 25C Tj... which is almost a 0.2V difference.

But here is how it gets even weirder: as you put more current through the diode, you expect to have bigger Vf (according to that graph). However, higher Vf = more heat the diode will be dissipating, which means it will run hotter. As it runs hotter, this does the opposite - it reduces the Vf. So what's the final Vf going to be? - That depends on the heatsink the diode is attached to (if any) and how well it looses heat to its environment.

analog_programmer wrote on 2025-02-26, 12:57:

I still can't find enough spare time to gut-out the PSU for a third (or fourth) time.

Yeah, I know that feeling and don't blame you.
FWIW, I repaired a "classic" beige PC speakers last week, only to find the volume pot is scratchy as hell, the line in jack contacts are extremely loose, and the bass pot is not making good contact at all, essentially making the speakers loose all of their bass. So gonna have to tare into these 2nd time.

analog_programmer wrote on 2025-02-26, 12:57:

So these new TS SF1608G double diodes are still at hand, but I don't know how can I test them more than just measure their V-drop at room temperature. I don't have any laboratory PSU with 10+ A DC line.

Old car battery + a few 12V 55W (or similar power) auto headlamps, and you should be set. Just power the bulb(s) through the diode and measure Vf with your meter. Just one important detail about that, though: make sure to mount the rectifier onto a decently-size heatsink, or you risk of overheating it and burning it out. Don't pick a heatsink that is too large, though, because then you will only get Vf for more or less "room temperature" conditions. If you want to mimic more closely what would happen inside the PSU, you have to let setup run for a few minutes (or longer) until the heatsink heats up to about 50-60C (i.e. to a point where you can't touch it because it's too hot, but not instantly burning off your skin - for most humans, the threshold is about 50C, with 55C being hard to hold, and 60C is considered scalding temp.) So when the heatsink gets to that temperature, then you can see what the actual Vf is probably going to look like inside the PSU, as that's what heatsinks tend to run at in many ATX PSUs when they are under heavy load.

analog_programmer wrote on 2025-02-26, 12:57:

It's nice to have a trusted electronic "scrap" seller 😀

It wouldn't call them trusted... but it works out in favor for both sides right now.

momaka wrote on 2025-03-02, 13:22:

Well semi-conductors can be semi-logical sometimes. :D I forgot what the physical explanation behind that was, but all I remembe […]
Show full quote

Well semi-conductors can be semi-logical sometimes. 😁
I forgot what the physical explanation behind that was, but all I remember is that the forward voltage of a P-N junction (of a diode, transistor, etc.) always drops when it heats up. (That said, this doesn't apply to MOSFETs conducting forward, as they don't have a P-N junction in that path.)
As an example, I attached a screenshot of the graph of average forward current vs. forward voltage drop for a UG10DCT (common-Cathode rectifier.)
download/file.php?mode=view&id=213363
First note how at low conducting currents (on the left axis), the forward voltage drop, Vf (bottom axis) is quite low. As the forward current increases, so does Vf. This is why your multimeter reads quite low of a drop - it's not putting any significant forward current through the diode, so Vf is low.
Next, note how as the junction temperature, Tj, is changed from room temperature (25C) to 100C and 125C, Vf "goes back" for the same forward current. For 1 Amp of I_Fa, for example, Vf can be as low as 0.55V at 125C Tj or as high as 0.75V at 25C Tj... which is almost a 0.2V difference.

But here is how it gets even weirder: as you put more current through the diode, you expect to have bigger Vf (according to that graph). However, higher Vf = more heat the diode will be dissipating, which means it will run hotter. As it runs hotter, this does the opposite - it reduces the Vf. So what's the final Vf going to be? - That depends on the heatsink the diode is attached to (if any) and how well it looses heat to its environment.

We have to go back to the fundamental physics for the explanation of this effects. With the raise of the temperature the electrons in every known material tend to "vibrate" stronger and can "jump" in foreign atomic "orbits" easier. So there is logic in the statement, that P-N junction become more conductive with increasing its temperature, thus dropping its Vf (for the P-N junction "spot"). Until the P-N breaks electrically beyond self-recovery 😀

momaka wrote on 2025-03-02, 13:22:

Old car battery + a few 12V 55W (or similar power) auto headlamps, and you should be set. Just power the bulb(s) through the diode and measure Vf with your meter. Just one important detail about that, though: make sure to mount the rectifier onto a decently-size heatsink, or you risk of overheating it and burning it out. Don't pick a heatsink that is too large, though, because then you will only get Vf for more or less "room temperature" conditions. If you want to mimic more closely what would happen inside the PSU, you have to let setup run for a few minutes (or longer) until the heatsink heats up to about 50-60C (i.e. to a point where you can't touch it because it's too hot, but not instantly burning off your skin - for most humans, the threshold is about 50C, with 55C being hard to hold, and 60C is considered scalding temp.) So when the heatsink gets to that temperature, then you can see what the actual Vf is probably going to look like inside the PSU, as that's what heatsinks tend to run at in many ATX PSUs when they are under heavy load.

Yep, I can recreate this DIY test setup easily, but I have to use the battery from my car as I don't have spare one. I think I'll just risk to replace those +12 V power lines double diodes, and then I'll test the "fixed" PSU with two or three 55 W headlamps for dummy load before any real use of the PSU, hoping not to see "the magic" white smoke 😁 Thanks for pointing me on using the right "tools" at hand!

analog_programmer wrote on 2025-03-02, 14:55:

We have to go back to the fundamental physics for the explanation of this effects. With the raise of the temperature the electrons in every known material tend to "vibrate" stronger and can "jump" in foreign atomic "orbits" easier. So there is logic in the statement, that P-N junction become more conductive with increasing its temperature, thus dropping its Vf (for the P-N junction "spot"). Until the P-N breaks electrically beyond self-recovery 😀

Yes, the reason was indeed something along those lines, but I didn't feel like looking through my EE books again. 😁
It's not practical knowledge I need everyday, so I pass.

analog_programmer wrote on 2025-03-02, 14:55:

Yep, I can recreate this DIY test setup easily, but I have to use the battery from my car as I don't have spare one.I think I'll just risk to replace those +12 V power lines double diodes, and then I'll test the "fixed" PSU with two or three 55 W headlamps for dummy load before any real use of the PSU, hoping not to see "the magic" white smoke 😁 Thanks for pointing me on using the right "tools" at hand!

You can also use a cheapo ATX PSU instead of the car battery. That way, you don't have to step outside to mess with the car. Also more fun if the cheapo ATX PSU lets out the magic smoke. 😉

To give some feedback on this topic from my side:

I finally got a replacement toroid core (MICROMETALS T130-52), dismanteled the PSU and removed the original toroid (L1 as marked in the schematic in the very first post in this thread).
I thought it'd be wise not to damage the original one, so I put it aside and just wired some other enameled copper wire from an old transformer to the toroid. The wire's diameter didn't fit the original (in fact it was quite thinner -> 0.65mm instead of the original 1.25 and 1.45mm) but in my brain this doesn't matter for how long as one doesn't pull too much current out of the PSU.
Of course I did copy the original turns ratio which is [5V : 12V : -12V] = 11 : 23 : 23. For 5V and 12V they used two wires each but again this should only matter for current capability.
(If there's anything wrong in my thinking please correct me.)

Now to the testing:
In short, nothing changed, 5V-rail is about 4.8V and 12V-rail stays at around 12.8V. Just like with the old toroid.
With VR1 both rails go up and down together with a slightly different gain which doesn't help anything as I think someone here has already written before.
So it does seem as we can pretty much rule out the toroid core as root cause for the voltage spread as momoka already suspected here posting.php?mode=quote&p=1321436. Can we?

Side note: the turns ratio on the toroid seems to me more of a ratio of change between the voltages rather than the absolute ratio between 5V and 12V because turning VR1 results in 5V-rail going from e.g. 4.8V -> 4.45V while 12V-rail going from 12.8V to 12.1V which is roughly 23 : 11 = 2,09.

Alright, where to go next?
I'm still trying to understand the concept of regulation within these PSUs. If 5V and 12V rails are regulated together, what key component defines the fundamental height of each one and what circumstance is liable for the spread from ideal values? Is it some ageing effect and if so, in what component does it take place, or is it more of a general design flaw and this PSU always ran out of spec like that?
Looking at the schematic from first post I only see T1 as a common point for energy transfer to the rails. That transfer must already be regulated enough to generate approximate output levels. Any change in the duty cycle on the primary side would reach both 5V and 12V rails in the same way, i.e. bringing both up or down together.
If now one rail gets heavy load then energy is transferred through toroid L1 from the other rail to prohibit any load regulation only affecting one rail while the other, no so much loaded, going up ungerulated. This leads me to think T1 could be part of the problem...

Further, in my understanding any change to the voltage summing/feedback network to +E/A (pin 4 of 3528 control circuit) (i.e. increasing/decreasing the "weight" of any rail to the feedback signal) should still result in an increase or decrease of both 5V and 12V together. So that doesn't seem to be that useful, too.
But that's already that, isn't it? I don't see any other possibility to up one rail while lowering the other with given circuitry...

Thanks for the detailed input ChrisK!

I didn't understand if you're using a new toroid core made from the very same material as the original one. If it's from same material and the windings are with same count of turns as the original ones, it seems pretty normal to get the very same voltages (regardless of thinner wires for the windings).

As for the +5 V and + 12 V "joint regulation" schematic - I share the same thoughts: an absolutely non-sensible schematic, but this is what it is.

P.S. I temporarily "forgot" about my modification idea with the diodes replacement as recently I'm mainly fiddling with some AT-motherboards using a trusty 30 years old AT-PSU with rock-solid voltages, but it's still in my "to do list"...

It seems to me that you are trying to repair a perfectly good ATX v2.2 spec power supply.
In general, it should not work with any Pentium 3/Athlon XP with maximum consumption on the +5V rail.
It was designed for new systems.
It usually works, of course.
But it should not show ideal voltage values.
https://theswissbay.ch/pdf/Datasheets/ATX12V_ … _public_br2.pdf

https://www.rosch-computer.de/manuals/product … /ATX-350PNF.pdf

shevalier wrote on 2025-06-02, 10:46:

It seems to me that you are trying to repair a perfectly good ATX v2.2 spec power supply.
In general, it should not work with any Pentium 3/Athlon XP with maximum consumption on the +5V rail.
It was designed for new systems.

Do you mean that this PSU should work normally only with Athlon 64 x2 or some "newer" P4 based systems? 👀

analog_programmer wrote on 2025-06-02, 10:59:

Do you mean that this PSU should work normally only with Athlon 64 x2 or some "newer" P4 based systems? 👀

I mean, it should work fine for loads inside a closed curve. Outside, it's undocumented behavior.
Yep,
with the main load of +12V PNF demonstrates excellent voltage stability.
Excellent power supplies, always loved them.

shevalier wrote on 2025-06-02, 11:07:

I mean, it should work fine for loads inside a closed curve. Outside, it's undocumented behavior. Yep, with the main load of +12 […]
Show full quote

I mean, it should work fine for loads inside a closed curve. Outside, it's undocumented behavior.
Yep,
with the main load of +12V PNF demonstrates excellent voltage stability.
Excellent power supplies, always loved them.

I see. Maybe I'd like this PSU, if it was less "specific" to certain generation ATX systems. Now I'm in doubt if I should "repair" it for usage with older AT/ATX systems, but I still don't have any usable ATX-PSU for P2/3 and Athlon (XP) systems 🙁

Ok, I have to try it with the Athlon 64 x2 system, before I change those diodes, just to be sure if it will give normal output voltages in this case.

shevalier wrote on 2025-06-02, 10:46:

https://theswissbay.ch/pdf/Datasheets/ATX12V_ … _public_br2.pdf

Hmmm, according to this diagram, maybe the PSU needs a lot more +12 V line load, if it's used in older system (one HDD, one FDD, and one CD-ROM obviously do not provide enough load for the PSU's +12 V line, even if there is just enough load on +5 V line).

I really have to check the PSU with Athlon 64 x2 system and if it works properly maybe I can do another modification for usage with older ATX and AT-systems: 15-20 W resistive "dummy" load with on-off switch on +12 V line.

analog_programmer:
I'm using a new core with "52" material from a specialized "non-cheapo-chinese" (at least I hope so) core manufacturer (https://www.micrometals.com/).
Same windings count etc. is intended since the initial assumption was that the core material of the original core could have degraded. That's why only one "parameter" (the core itself) was changed.
I think that this can be ruled out. When I understood momaka's post right between the lines he (she?) also suspected this. Curious about his/her findings by the way... 😉

shevalier:
I wouldn't call it "repair" since it is actually working. It's more or less some "optimization" and also some "learning by doing" since it is vastly running out of spec regarding voltage tolerances, not combined load ratings or anything.

ATX-specs demand 5% tolerance on all positive rails under all operationg conditions, that is even at no load and light load. This is what this PSU is failing on and makes it a bit awkward to use since you rely on the tolerance of the hardware you are using it with.
At least my unit is running at 4.8V and 12.8..12.9V which is barely within spec (4.75V min.) and quite a bit above spec (12.6V max.), respectively. Even under load in a newer system (C2D and up).
The hardware will run with these levels and even with higher and lower levels, too, no question. But how can you say the PSU will stay at these for sure?
So we are basically looking for a way to get these rails more into balance.

ChrisK wrote on 2025-06-02, 12:15:

analog_programmer: I'm using a new core with "52" material from a specialized "non-cheapo-chinese" (at least I hope so) core man […]
Show full quote

analog_programmer:
I'm using a new core with "52" material from a specialized "non-cheapo-chinese" (at least I hope so) core manufacturer (https://www.micrometals.com/).
Same windings count etc. is intended since the initial assumption was that the core material of the original core could have degraded. That's why only one "parameter" (the core itself) was changed.
I think that this can be ruled out. When I understood momaka's post right between the lines he (she?) also suspected this. Curious about his/her findings by the way... 😉

Ok. So now we can surely remove the inductor's core material out of the "equation". But there are more "variables" left in - for example changing windings turns count like in one of the YT-videos I've linked before.

And of course shavelier's ATX PSU ver. 2.2 specs revelation seems legit ('til I check this in a "+12 V hungry system" and eventually prove as "wrong").

ChrisK wrote on 2025-06-02, 12:15:

shevalier: I wouldn't call it "repair" since it is actually working. It's more or less some "optimization" and also some "learni […]
Show full quote

shevalier:
I wouldn't call it "repair" since it is actually working. It's more or less some "optimization" and also some "learning by doing" since it is vastly running out of spec regarding voltage tolerances, not combined load ratings or anything.

ATX-specs demand 5% tolerance on all positive rails under all operationg conditions, that is even at no load and light load. This is what this PSU is failing on and makes it a bit awkward to use since you rely on the tolerance of the hardware you are using it with.
At least my unit is running at 4.8V and 12.8..12.9V which is barely within spec (4.75V min.) and quite a bit above spec (12.6V max.), respectively. Even under load in a newer system (C2D and up).
The hardware will run with these levels and even with higher and lower levels, too, no question. But how can you say the PSU will stay at these for sure?
So we are basically looking for a way to get these rails more into balance.

I can once more confirm your words. My PSU behaves exactly the same and produces very similar "out of specs" +5 V and +12 V output lines voltage values, when runs without any load or in old "pentium class" system with very little load on +12 V line.

shevalier wrote on 2025-06-02, 11:07:

I mean, it should work fine for loads inside a closed curve. Outside, it's undocumented behavior. Yep, with the main load of +12 […]
Show full quote

analog_programmer wrote on 2025-06-02, 10:59:

Do you mean that this PSU should work normally only with Athlon 64 x2 or some "newer" P4 based systems? 👀

I mean, it should work fine for loads inside a closed curve. Outside, it's undocumented behavior.
Yep,
with the main load of +12V PNF demonstrates excellent voltage stability.
Excellent power supplies, always loved them.

You mean voltages are allowed to go out of spec when no defined load is connected?
Combined with the spec you linked this could be possible but still arguable...
From ATX 2.2 spec:

3.2.1. DC Voltage Regulation The DC output voltages shall remain within the regulation ranges shown in Table 2 when measured at […]
Show full quote

3.2.1. DC Voltage Regulation
The DC output voltages shall remain within the regulation ranges shown in Table 2 when
measured at the load end of the output connectors under all line, load, and environmental
conditions.

Back to testing...

Yep, seems like we have to test this Fortron/FSP ATX-350PNF model with +3.3 & +5 V combined load + some proper +12 V load.

Without load (short green&black wire), the output voltages will be determined by the ratio of load resistors R42/R41, which you have circled absolutely correctly.
They are located next to the AC power input (and OCP board if it`s present) and have a darkened pcb underneath. 🙁
(Bad advice)
Put a 10W ceramic resistor in position R41 at 51 Ohm, and the voltages will be fine without load.
Only the temperature inside will not be fine.

My PSU produces +12.8 without any load.

analog_programmer wrote on 2025-06-02, 13:06:

Yep, seems like we have to test this Fortron/FSP ATX-350PNF model with +3.3 & +5 V combined load + some proper +12 V load.

Athlon S754 ~65W + 2600ProAGP~ 30W at +12V rail

shevalier wrote on 2025-06-02, 13:30:

Without load (short green&black wire), the output voltages will be determined by the ratio of load resistors R42/R41, which you […]
Show full quote

Without load (short green&black wire), the output voltages will be determined by the ratio of load resistors R42/R41, which you have circled absolutely correctly.
They are located next to the AC power input (and OCP board if it`s present) and have a darkened pcb underneath. 🙁
(Bad advice)
Put a 10W ceramic resistor in position R41 at 51 Ohm, and the voltages will be fine without load.
Only the temperature inside will not be fine.

My PSU produces +12.8 without any load.

Athlon S754 ~65W + 2600ProAGP~ 30W at +12V rail

Ok. seem like your PNF400 works fine with proper load in your Athlon 64 system.

I'm thinking about some more flexible +12 V line loading modification: additional "pack" of big ceramic resistors outside the PSU casing connected through an on-off switch.

If you really want to experiment, you need to replace the the Schottky rectifier with the +5V rail.
I don't remember if there is place for a second one in parallel.
If there is, then put two at once.
On a modern one, with a very small voltage drop.
https://www.vishay.com/docs/96456/vs-40cpq-n3series.pdf
https://eu.mouser.com/datasheet/2/240/media-3322386.pdf
You will gain 0.1-0.2 Volts due to the redistribution of energy due to the change in the internal resistance of the source.
And remove the second diode from the +12V rail.

shevalier wrote on 2025-06-02, 15:09:

If you really want to experiment, you need to replace the the Schottky rectifier with the +5V rail. I don't remember if there is […]
Show full quote

If you really want to experiment, you need to replace the the Schottky rectifier with the +5V rail.
I don't remember if there is place for a second one in parallel.
If there is, then put two at once.
On a modern one, with a very small voltage drop.
https://www.vishay.com/docs/96456/vs-40cpq-n3series.pdf
https://eu.mouser.com/datasheet/2/240/media-3322386.pdf
You will gain 0.1-0.2 Volts due to the redistribution of energy due to the change in the internal resistance of the source.
And remove the second diode from the +12V rail.

Ok, but let me first try the PSU with the properly distributed load on +5 V and + 12 V lines and then (if needed) I can still change those +5 V line schottky diodes with the ones that I already have (with a bigger V-drop) or with the ones you're suggesting (with a lower V-dorp) - according to how I will need to regulate those two output lines via VR1 if it need all this fiddling after all 😉