I'd replace them with polymer electrolytic caps or MLCC.
Reason is the low resistance, as tantalums were back then considered as the caps with lowest ESR.

It is always a good idea to replace all tantalums on the board if one went kaboom. They are not really well suitable for continuous high ripple operation, and the others might already be borderline, too. I'd also consider measuring the PSU for ripple, as this might be the deeper-lying cause for the explosion.

MLCC is pretty expensive compared to cheap electrolytics.. Is low ESR the reason we are going for that?

Umm, I don't know what should be considered "expensive"...
Recently bought some Samsung MLCC.
10uF/16V cost me <10cent/pc, 47uF/6.3V <50cent/pc.
And yes, it's for speed. The lower the ESR, the better the caps can decouple spikes of all sorts.

retardware wrote on 2021-09-06, 15:28:

Umm, I don't know what should be considered "expensive"... Recently bought some Samsung MLCC. 10uF/16V cost me <10cent/pc, 47uF/ […]
Show full quote

Umm, I don't know what should be considered "expensive"...
Recently bought some Samsung MLCC.
10uF/16V cost me <10cent/pc, 47uF/6.3V <50cent/pc.
And yes, it's for speed. The lower the ESR, the better the caps can decouple spikes of all sorts.

OK then, I will replace them with electrolytics for piece of mind for the moment, I will replace them again with MLCCs later down the line then.

I think electrolytics will be just fine, in fact I have seen many boards of that era (286/386/486) where they used tantalums and electrolytics interchangeably, i.e. on multiple instances of the same board model you see some with tantalums and others with electrolytics. I think in most cases they just used whatever was available during production.

Regarding which ones to replace, the caps on the -12V line are usually the first ones to pop, followed by the 12V ones. I don't think I've ever seen a tantalum on 5V explode, maybe because that rail is usually better regulated and the spikes won't reach as easily the 16V the tantalums on that line are usually rated for.

TheMobRules wrote on 2021-09-06, 17:00:

Regarding which ones to replace, the caps on the -12V line are usually the first ones to pop, followed by the 12V ones. I don't think I've ever seen a tantalum on 5V explode, maybe because that rail is usually better regulated and the spikes won't reach as easily the 16V the tantalums on that line are usually rated for.

Now that is an interesting observation.
Personally, I was lucky, had no tantalum explosions, just one quietly shorted one that prevented the PSU from starting (Took some time to find out which one it was...).
Maybe it's a good idea to put the PSUs on the test bench before deciding whether to use or throw them away.
I threw away lots of PSUs after my big test session for my article on the German classic-computing.de forum.
The worst one ("working", but with leaky caps) had 14+V overlaid with 2V ripple... a true killer!

In short, there are many (and quite a few ATX)AT PSus that have very bad crossloading behaviour.
The worst ones that I tested almost reached 15V on the 12V lines when only (and heavily) loaded on +5V.
A 3.5" HDD typically draws less than 1A, which does not help much (the 5.25" drives sucked 2+A).
This is one of the reasons why I like Seasonic... they all stay inside the ATX voltage tolerances specs.

Maybe consider a modern DC-DC-DC PSU for your valuable hardware.
Or, use 25V caps instead of 16V ones...

MLCC are not that expensive now. Polymer capacitors are excellent subs too.

Cheers,

retardware wrote on 2021-09-06, 18:26:

Now that is an interesting observation. Personally, I was lucky, had no tantalum explosions, just one quietly shorted one that p […]
Show full quote

Now that is an interesting observation.
Personally, I was lucky, had no tantalum explosions, just one quietly shorted one that prevented the PSU from starting (Took some time to find out which one it was...).
Maybe it's a good idea to put the PSUs on the test bench before deciding whether to use or throw them away.
I threw away lots of PSUs after my big test session for my article on the German classic-computing.de forum.
The worst one ("working", but with leaky caps) had 14+V overlaid with 2V ripple... a true killer!

Sometimes there are well designed PSUs that are crippled by abysmal caps, Antec units from the early to mid '00s are the leading examples of that. Whenever I get a used power supply, I try to identify candidates worthy of being refurbished. In those cases a recap + a new fan is all it takes to have a solid power supply, but it is definitely a good practice to perform some reliability tests before using it on valuable hardware. Of course, any unit with burn marks or multiple component failures is a no go in my book, at some point the cost of replacing components is just not worth it, especially without knowing the previous history of the PSU and what kind of stress/heat it was subjected to.

retardware wrote on 2021-09-06, 18:26:

In short, there are many (and quite a few ATX)AT PSus that have very bad crossloading behaviour. The worst ones that I tested al […]
Show full quote

In short, there are many (and quite a few ATX)AT PSus that have very bad crossloading behaviour.
The worst ones that I tested almost reached 15V on the 12V lines when only (and heavily) loaded on +5V.
A 3.5" HDD typically draws less than 1A, which does not help much (the 5.25" drives sucked 2+A).
This is one of the reasons why I like Seasonic... they all stay inside the ATX voltage tolerances specs.

Yes, some AT units, especially earlier ones required a certain balance between the +5V and +12V loads... after all that's why IBM supplied a chunky 50W resistor on the 5170 to act as a dummy load on +12V when you didn't have an HDD installed. But that PSU was well designed, and just refused to turn on under inappropriate load conditions. The real danger here are the units that will happily turn on despite the voltages being totally out of whack (I wouldn't be suprised if the PG line was tied directly to +5V on those specimens).

retardware wrote on 2021-09-06, 18:26:

Or, use 25V caps instead of 16V ones...

Yeah, if I had to use tantalums I would definitely go with 25V caps, especially for 12V and -12V. If there's something tantalums doesn't like is when you go over the rated voltage, so that would at least add a layer of safety.

I replaced all the tantalums with electrolytics, and the first time I powered the thing on, I realized that the cap on the -12V rail was HOT. I mean, if I held my finger to it for 3 seconds it would start hurting like a bitch hot. So I turned it off, inspected everything, found nothing wrong, restarted and it was cool this time around? Tried it five more times, and it's just fine. I am using a cheapass noname ATX PSU with an ATX to AT converter, so I am thinking my -12V is probably shitty, which is why the cap blew in the first place. Regardless, I resurrected this motherboard and I'm happy with that.