shevalier wrote on 2026-05-08, 06:13:
About 15 years ago, I tinkered with the same OEM Hipro unit—without a clear schematic—and I really didn't like it. Overall, for this power range, I prefer the classic half-bridge topology.
I'm the other way around - I always prefer forward converter topologies due to the fact that they react much quicker in case of a short-circuit on the output (due to more precise OPP on the primary). This is actually really important when I test motherboards or video cards where I had to do MOSFET replacement. If there is anything wrong with the driver or PWM controller circuit on the video card / motherboard, it's important for the PSU to shut down quickly to prevent possibly further damage. Also, when testing a motherboard that I had to do MOSFET replacement on, I usually pick a lower wattage PSU so that, again, it will shut down quicker in case something isn't right.
Half-bridge not only has imprecise OPP, but some designs often won't even latch off and continue to try to power-cycle / turn On after shutting down. This can cause constant power-up power-down cycling and do damage to other components.
And lastly, half-bridge can easily blow its primary side BJTs during extended brownout conditions if the OPP is not set very well (and for many PSUs, especially the cheapo units, it's not). For forward topologies, this is much harder to do, unless the manufacturer goofed up and really messed up on the calculations for the OPP.
shevalier wrote on 2026-05-08, 06:13:
It produces the same high-voltage spikes as a flyback topology, but lacks “built-in short-circuit protection via power limiting.”
Not quite.
STF almost never lacks protection via power limiting, as the PWM IC is on the primary side and provides pulse-by-pulse current limiting.
Also, STF does not have the same "very quick" and high voltage back-EMF spikes (i.e. high amplitude, short duration impulse) that flyback topology does. That's why the switching transistor for flyback usually has to be rated for very high voltage (typically 600+V), whereas for STF, it can be lower.
In any case, the snubber still needs to grow really quickly for designs past 300-350 Watts... hence why it's not so practical for higher powers.
shevalier wrote on 2026-05-08, 06:13:
If the new power supply burned out, you went and replaced it under warranty.
Yeah, well, not always the case.
I mean if it's a local store and they can just take it back and exchange for another unit, then fine. But if it has to be shipped out, I've seen manufacturers make the buyer pay for the shipping to the warranty center. If that's the case, forget it - it will probably be cheaper to buy another PSU.
shevalier wrote on 2026-05-16, 04:55:
I’ve had to deal with a Codegen like that before.
It all ended up with me having to wind an extra winding for +3.3V on the transformer, fit diodes instead of the MOSFET, and route part of the group stabilisation choke windings to the +3.3V channel.
It actually worked, and not badly at all. For a P4-based system where there is no direct use of the +3.3V voltage from the PSU.
???
So you created a 3.3V rail with no regulation directly from the main transformer? If yes, that's really the least optimal way to do it. I've actually seen only a couple of very cheap/nasty PSUs use such design, and they did absolutely terrible on any cross-loading tests.
Here's one I looked at some years ago:
https://www.badcaps.net/forum/troubleshooting … g-inline-images
Actually on that one, the 3.3V rail didn't do too badly, so long as the 5V rail was loaded down nicely and consistently (like from a low-power Athlon or PII/3 system.)
But with a more modern system, even with low power draw (but majority of it from the 12V rail nonetheless), the 3.3V rail went way too high (above 3.5V). Likewise, with a very heavy 5V power draw, the 3.3V rail again became high. And if loading down the 3.3V rail by itself, it sagged really badly (barely 3V.)
So yeah, having an ATX PSU where all of the outputs are group-regulated together is not a good idea - not unless you really build the PSU to power a specific system where you know what all of the power draws will be.
tehsiggi wrote on 2026-05-18, 18:06:
So the PSU arrived. Good news? It's in one piece and is not using a linear regulator to get from 5V to 3.3V by the looks of it.
Inherently, there is nothing wrong with a linear 3.3V rail. Its only downside is that it's more (very) inefficient, especially if down-regulating from the 5V rail... and for the majority of PSUs out there that do/did have a linear 3.3V rail, this is how it's done indeed. So for losses, you have:
P_loss = (5V - 3.3V ) x I_3v3load
If for example, the load on 3.3V rail is 10 Amps, then the losses (as heat) in the MOSFET are 1.7V x 10 Amps = 17 Watts... which is quite a bit!
Compare that to a mag-amp, where for the same load, you just have a (Schottky) rectifier loss plus maybe a 1-3 Watts in the saturation toroid core... so about 0.5V x 10 Amps, or 5 Watts in the rectifier + 2 to 3 Watts in the saturation coil, totaling about 7-8 Watts - about half of that of the linear regulator.
But on the positive side, those linear 3.3V regulator circuits are super quiet on the output, regardless of their load. Even with terribly degraded caps, they will still regulate OK. Only if the 5V rail becomes unstable, that's when the 3.3V linear reg circuit can also refuse to work or become really noisy.
Now, HiPro did do it a little better in their PSUs - they generated a separate "high 3.3V" rail that was under 4V generally and then used a MOSFET to down-regulate to 3.3V. As such, the losses in the MOSFET were much lower since it now has to drop only 0.7V max, if even that. So for the same 10 Amp load, it would have a Schottky rectifier power loss (about 5 Watts for 10 Amps) + a much reduced MOSFET power loss (0.7V x 10 Amp, at most)... totalling about 12 Watts. Those losses are still obviously worse than a mag-amp regulator circuit, but they are also much improved compared to a "straight 5V to 3.3V) linear regulation implementation.
The other advantage of a linear 3.3V rail circuit is that it needs no load to regulate properly... if done right, of course (I did encounter a Sun Pro -built Raidmax PSU before where this wasn't the case and the 3.3V rail was usually too high most of the time... unless it was loaded - then it sagged badly 🤣 .) With mag-amp 3.3V, a constant minimum load may be needed to keep it stable/operating within range. It all depends for what loads the 3.3V rail was designed for. For older 3.3 & 5V-heavy PCs, the 3.3V rail was designed for greater loads and thus it may not do to well without any load (often the case with the older half-bridge -based FSP PSUs, which is why they always put those big load resistors on the output voltage rails.)
tehsiggi wrote on 2026-05-18, 18:06:
However, the spec sticker is a whole lie. The output rails use 2x s30d40c - one for 5V and one for 3.3V - those are rated at 30 amps each. Not 30A per diode inside them, but 30A for both diodes together. The 5V rail with it's rated 42A is a blatant lie. Same for the 12V - it's spec'ed at 24A - uses an F16C20C - 16A for both diodes together. That's far from the rated 24A.
Yeah, that's standard for cheapo PSUs (Codegen included), as shevalier noted.
tehsiggi wrote on 2026-05-18, 18:06:
For my purpose it does not really matter, I'll re-cap the PSU and be happy. Even if I can get 25A out of the 5V rail I'm already happy. It's not a 12V centric PSU, that's all that really mattered for me. And it's not like the russian reviews I found for these units didn't tell already that this is basically a fraud 😁
It's indeed not that bad of a PSU, the lying label aside.
You've got 30 Amp rectifiers for the 3.3V and 5V rail, and they are in the larger TO-247 case.
Likewise, the primary BJTs are in TO-3P case, so they will definitely be better than the standard E13007 affair that most other cheapo PSUs used at the time - basically guaranteed up to 300 Watts continuous load (though I wouldn't try pushing them more.) The main transformer also appears to be sized approprietly, certainly enough to sustain those 300 Watts (and likely 350W peak), which should be plenty for any old retro 5V-heavy PC.
And hey, you've even got a PPFC inductor! The manufacturer had to sacrifice "wasting" copper there for you 🤣.
The only thing that possibly looks a little worrisome is the output side - particularly the side of the output toroid and the filtering caps after it.
To me, that toroid appears to be "106" size (about 2.5 cm dia. core) vs the larger "130" core (about 3.3 cm dia.) And it's Micrometals -26 type (yellow core with one side painted white), which is more lossy than -52 type core (light green with one side painted blue). As such, I think the maximum continuous load on the 5V rail should generally be kept under 20 Amps... which again, should be plenty for an old system. The PSU may be capable of outputting more, but for PSUs with T106-26 output toroid cores, I find that output toroid will often start running very very hot. For good longevity of the core, it's better to not run it too hot.
Also, you may or may not be able to do 20 Amps on the 5V rail if the output caps are not "up to snuff". Generally, the 5V rail should have at least 2x 2200 uF caps and a PI coil / bobbin inductor between them if you want to pull anything close to 20 Amps without the ripple becoming a problem. Increase capacity one notch up for 25-30 Amps (so 2x 2700-3300 uF). This is where many cheapo PSUs cut corners and either don't install big enough output capacitors or use only one or don't include a "PI" coil/bobbin inductor. Any of these (but especially the missing PI coil) can result in much higher output ripple.
The reason I mention this is because when you do the recap, you should see what the PSU has (in terms of capacity for the capacitors), but not go below the minimum capacity mentioned above. I've often seen some cheapo PSUs put only 2x 1000 uF caps on the 5V rail, and that is absolutely no good for a 5V-heavy PC.
Same goes for the 3.3V rail.
For the 12V rail, at least one 1500 uF cap will generally get the job done for a not-12V-heavy PC.
There's generally no penalty for going with higher capacitance (even if doubling it), but sometimes there is for going too low with the ESR (certain older PSUs like this may oscillate very badly with caps with too low ESR.) Therefore, DO NOT use polymers. For old half-bridge topology like this, entry-level low ESR is more than fine. The switching frequency is fairly low, so it's really the capacitance that starts to matter more here.
shevalier wrote on 2026-05-18, 18:30:
And a transformer with the same markings as their 350-watt model.
It is clear that the transistors are not of the D209L (or their 12 A equivalents, which are used in 450–500 W models), and the input capacitors do not have a capacitance of 1000 μF / 2 = 500 W.
In all likelihood, their capacitance does not even reach 820 μF.
Well, the transformer is -proper- "35" size, at least (35 mm across the top and at least 40 mm tall). I've seen worst in L&C PSUs, where the transformer says ERL35, but it really is a 33-sized one... and about 1 cm shorter.
For oldschool half bridge PSUs like this (i.e. relatively low switching frequency), ERL35 is usually good for about 300-350 Watts max. Beyond that, BJTs tend to blow up (even if they are much bigger), because the core can saturate.
With that said, no need for very big input caps anyways. If the PSU is only realistically capable of 300-350 Watts max (which I think is the case here with tehsiggi's PSU), 2x 560-680 uF input caps should be perfectly adequate.
