TELVM wrote on 2026-06-12, 09:21:

Feallan wrote on 2026-06-11, 22:13:

... I would think the worst case scenario is a blown fuse in the PSU ...

Disclaimer - Not for the faint of heart ...

Worst Case Scenario

Why am I looking at video of a P3 setup with no HSF on the CPU, GPU and northbridge? Like, it is almost as if this was built to burn. A lot of the fires here are as a result of no thermal cooling, rather than voltage I would say. Even IF you supplied the right voltage to this setup, it would probably burn. No wonder the fire started at the AGP port.

TELVM wrote on 2026-06-12, 09:21:

Worst Case Scenario

According to the ATX standard, this is perfectly normal behaviour.
It has an incredibly wide operating range for the OVP.

appiah4 wrote on 2026-06-12, 09:43:

Why am I looking at video of a P3 setup with no HSF on the CPU, GPU and northbridge? Like, it is almost as if this was built to burn ...

Because that's not "real combat footage", it was staged.

To illustrate (in a "controlled lab test") what will happen to the system if the PSU's voltage regulation ever goes FUBAR.

Two compilations of genuine chinese fireworks:

https://www.youtube.com/watch?v=Ao0X7aouMB8

https://www.youtube.com/watch?v=Vb9zjxqklJA

A notorious scandal just a few years ago:

Exploding Power Supplies: Gigabyte & Newegg Dumping Unsellable Product

Gigabyte GP-P750GM Review - A Power Supply with an explosive attitude

Wish they wrote how they staged it in the video. It only says "wrong voltage supplied", but wrong voltage could be 12V into the 5 and 3.3V lines, or they could have rigged it to just pass through 230V AC... I don't think we should consider such scenarios. If we go with the other videos you posted, it seems the most common "explosive" part are the switching transformers - which you mentioned as what most likely killed my own Corsair 550W PSU. Never opened it so can't confirm, but I have indeed been lucky, that it died so gracefully. Although I'm not sure how exactly this kind of explosion plays out. When the transformer blows up, is there some sudden voltage spike before it goes to 0? Also "good" PSU design should ensure at least some overvoltage protection, so maybe these kick in then. Not very knowledgeable on the topic.

@shevalier - Wouldn't these 12V caps discharge still be protected by OVP circuits? I would think even 20+ year old PSUs with moderately good design should have some zener diodes at least.

@cyclone3d - I did see many bulged and leaking caps myself, especially in things made in the capacitor plague era. Even bought a cheap used 5V heavy Chieftec PSU, advertised as 100% working, I measured it without load and with some small load (2 HDDs and 2 CD roms I think), it was indeed fine. When I opened it up, it had 2 leaked caps, a lot of Fuhyuu ones + Teapo. Well I ended up buying another one, Delta this time. Also measured fine, with nothing suspicious inside so I ended up using it for my build. It does have CapXon capacitors... But I took the risk, it would take me a lot of time to find proper replacement caps

Feallan wrote on 2026-06-13, 15:48:

But I took the risk, it would take me a lot of time to find proper replacement caps

It's not like *finding* capacitors takes a lot of time, as you can just buy them from Digikey or LCSC. Although it does take some time to read the datasheets and so on to figure out which ones to buy.

shevalier wrote on 2026-06-11, 14:41:

Have you seen the prices for the Pentium 3S at a CUBX with 1 GB of RAM?
It’ll be a shame if some modern Chieftec goes burn and takes everything down with it.

Decent modern PSUs are also available, and something with 120-150W for 3.3V and 5V should be good enough for a P3 build, as apparently not all of them are dumb enough to turn fan off when 12V is lightly loaded. This is a much easier route for someone who is not comfortable with soldering, and I think he meant that.

Btw, why is that motherboard so expensive, what's so special about it when compared to other 440BX boards? 1.4GHz PIII-S CPUs are surprisingly expensive as well, although slower models aren't that bad.

Still, socket A CPUs are more powerful than Tualatins, and ISA boards can be had for a relatively low price as well. Especially if you are fine with non-A KT133.

Feallan wrote on 2026-06-13, 15:48:

@shevalier - Wouldn't these 12V caps discharge still be protected by OVP circuits? I would think even 20+ year old PSUs with moderately good design should have some zener diodes at least.

There are no Zener diodes capable of withstanding a load of more than 300 W per pulse.
It is entirely possible to design such a circuit, but I am not sure it would be economically viable.
This is by no means a typical load for modern power supplies.
If I were a power supply designer, I wouldn’t even consider such a load scenario. Because it’s a daft idea.

Mike_ wrote on 2026-06-13, 17:25:

Btw, why is that motherboard so expensive, what's so special about it when compared to other 440BX boards? 1.4GHz PIII-S CPUs are surprisingly expensive as well, although slower models aren't that bad.

I don’t know.
I’ve got two CUBX-L .
One has been modified to CUBX spec (with a voltage regulator for the north bridge and RAM), the other is factory-standard.
It’s a bit of a piece of rubbish, in my opinion. But retro enthusiasts seem to like it.
In theory, you can reach 1 GB of RAM on a PC133 system. In theory.
That means either 4 PC133 CL2 modules or you'll need to look for FB-DIMMs.

Mike_ wrote on 2026-06-13, 17:25:

Feallan wrote on 2026-06-13, 15:48:

But I took the risk, it would take me a lot of time to find proper replacement caps

It's not like *finding* capacitors takes a lot of time, as you can just buy them from Digikey or LCSC. Although it does take some time to read the datasheets and so on to figure out which ones to buy.

Yeah that's the problem. Desolder 20 caps to read the markings, as the interior is usually cramped, find all the datasheets, go through them, read datasheets for modern replacements, find the exact (or close enough) replacements. I'd do it for the chieftec PSU if I really had to, but fortunately the Delta one looks fine

Feallan wrote on 2026-06-13, 20:28:

Yeah that's the problem. Desolder 20 caps to read the markings, as the interior is usually cramped, find all the datasheets, go through them, read datasheets for modern replacements, find the exact (or close enough) replacements.

Manufacturers of ATX power supply units seem to have a strange penchant for using the wrong types of capacitors.

Back in the day, they used 0.1–0.22 μF*50 V electrolytic capacitors in the frequency correction circuits. And there might be as few as five of them scattered across the entire board.
In more recent times, with SMD assembly, they used X-type ceramic capacitors instead of C0G.
I understand that cost-saving falls on the development department, whilst RMA falls on the warranty department.
The summarise effect is zero, but it looks fundamentally corporate. 🙁

Feallan wrote on 2026-06-13, 15:48:

... it seems the most common "explosive" part are the switching transformers ...

The primary switching transistors:

^ KABOOM KABOOM in a Gigabyte GP-P750GM.

In the GN video they comment that one of these fireworks murdered a GTX 3080. Ouch!

Mike_ wrote on 2026-06-13, 17:25:

... Decent modern PSUs are also available, and something with 120-150W for 3.3V and 5V should be good enough for a P3 build, as apparently not all of them are dumb enough to turn fan off when 12V is lightly loaded ...

Paradoxically, though they never monitor +5V load or´+5V DC-DC FETs temp to spin up the fan, most budget modern DC-DC PSUs "idle" the fan at a relatively high speed (usually 800+ rpm).

They do so because, to keep price low, they're typically stuffed with bottom-of-the-barrel crapacitors (ChengX, Jun Fu, etc.), whose only chance to survive more or less "within specs" (at least until warranty expires) is overkill cooling.

From TechPowerUp - CX450M review:

^ Fan control only spins it up when +12V load increases. Fan control couldn't care less about high +5V load, high +3.3V load, or combined minor rail high loads, it just keeps the fan "idling" under such conditions.

But the "idling" speed is relatively high (~950 RPM in this PSU), and that's miles better for DC-DC FETs cooling than a stopped fan.

* BTW the CX450M ain't so bad in the cap department: Elites, Teapos, and one undersized Chemi-Con KMR bulk cap.

.

shevalier wrote on 2026-06-13, 18:34:

I don’t know. I’ve got two CUBX-L . One has been modified to CUBX spec (with a voltage regulator for the north bridge and RAM), […]
Show full quote

Mike_ wrote on 2026-06-13, 17:25:

Btw, why is that motherboard so expensive, what's so special about it when compared to other 440BX boards? 1.4GHz PIII-S CPUs are surprisingly expensive as well, although slower models aren't that bad.

I don’t know.
I’ve got two CUBX-L .
One has been modified to CUBX spec (with a voltage regulator for the north bridge and RAM), the other is factory-standard.
It’s a bit of a piece of rubbish, in my opinion. But retro enthusiasts seem to like it.
In theory, you can reach 1 GB of RAM on a PC133 system. In theory.
That means either 4 PC133 CL2 modules or you'll need to look for FB-DIMMs.

You can get 1.5GB of memory with most KT133/KT133A boards as well, most of them seem to support 512MB DIMMs...