If heat fixes your problem I would guess it's a contact issue, either at the solder joints or inside the chip package. The internal wires are usually bonded with the actual chip pretty well so I'd start probing for increased resistance between IC leg and solder pad and if necessary reflow the solder joint.
Check the RAM chips as well.

Thanks, Doornkaat. I actually ended up checking the continuity of each individual pin for both TMUs and all RAM ICs... everything looks good.
Unfortunately, it's probably an issue inside one of the chips/RAM ICs, but I have no idea how I could ever identify which one it is...

Hey dude,
you could try heating up the TMUs and FBI individually with a hair dryer while shielding the other two chips with cardboard or something before starting the PC and running a 3D application to try and find out which chip(s) need to be warm to work correctly? If it's not one of the main chips, maybe start heating RAM in groups of four to narrow them down? Just an idea, again I can't gurantee anything.

Btw. do you know what the maximum resistance for continuity is on your multimeter? It may show continuity despite resistance of a solder joint being relatively high. Hence I suggested checking the joints for increased resistance, not just continuity.
I know it's a ton of work to redo this with that many legs and especially with the heatsinks in the way. It may very well turn out all joints are electrically 100% fine. So I'm sorry if there was a misunderstanding.
But if you can figure out the IC culprit maybe try again, this time measuring for resistance.
Maybe reflowing the chip(s) legs right away is a better idea than probing. If you're handy with a soldering iron and have good flux and a wide chisel tip you should be able to do this pretty quickly since there's no BGA involved.

Good luck, I'm hoping for a christmas miracle! 😉

That's actually great advice! Thanks a lot!
I will first try your suggestion of heating the TMUs individually. If I don't see any difference, I'll then try heating up the RAM ICs in groups of four and see if I can at least pinpoint where the faulty module is 😀
I don't think it's worth trying it on the FBI chip, since I tested the card with disabled textures and it works perfectly - so most likely it's either the TMUs or some RAM IC.

Again, thank you very much! I'll get back with the results in two weeks, when I return from my vacation 😀

bloodem wrote on 2020-12-23, 17:55:

That's actually great advice! Thanks a lot! I will first try your suggestion of heating the TMUs individually. If I don't see an […]
Show full quote

That's actually great advice! Thanks a lot!
I will first try your suggestion of heating the TMUs individually. If I don't see any difference, I'll then try heating up the RAM ICs in groups of four and see if I can at least pinpoint where the faulty module is 😀
I don't think it's worth trying it on the FBI chip, since I tested the card with disabled textures and it works perfectly - so most likely it's either the TMUs or some RAM IC.

Again, thank you very much! I'll get back with the results in two weeks, when I return from my vacation 😀

Good luck and have fun on your vacation!
Why do you use multiple accounts though? No shame in having to ask for help once in a while! 😀

I. will. kill him. 😁
We are actually two very, very, very different people 😁 and we're not using multiple accounts (err... or we shouldn't).
He is a colleague from work to whom I recommended this community, since we share the same passion.
Our workflow involves sometimes using the same VM (with the same local account) as a workstation, and I think my vogons account was active and he didn't realize it. 😀

Browsing Vogons at work? 🤔
Checks out!😅

Hehehe, did I say that? Oops! 😁

Call me crazy but if I were to bet money I would put it on those SMD electrolytics having leaked onto their pads and causing cold solder issues until the card warms up..

appiah4 wrote on 2020-12-24, 12:31:

Call me crazy but if I were to bet money I would put it on those SMD electrolytics having leaked onto their pads and causing cold solder issues until the card warms up..

No, the capacitors are fine, one of the first things I checked.

bloodem wrote on 2020-12-24, 06:43:

We are actually two very, very, very different people

😜 I got the joke 😜
Anyway, sorry for the screw-up, guys! That was my bad 😀

Have been very busy lately so the Voodoo card had to wait... but today I finally tried my luck with it again. I tried the test that Doornkaat recommended (preheated one TMU at a time), but this did not have any meaningful results.
However, I then thought of something else: what if it's not actually a TMU issue? What if the FBI memory is affected (or maybe the FBI chip itself?).
So my next test was to actually limit the FBI memory to 2 MB (something that for some reason I did not try before). And what do you know... the issue is completely solved. I left the PC turned off multiple times to cool down, restarted it, started GLQuake / Quake 2 (which were the most sensitive), no freezes, no corrupted textures, no problems at all.

So I now know for sure that either a portion of the FBI, or a memory IC that belongs to the FBI is affected. The simple (optimistic) assumption is that one of the memory ICs is to blame and the FBI is probably OK.
The question is where do I go from here? How could I narrow things down to at least a few ICs? The FBI has 4 memory ICs on the bottom front, and 4 memory ICs on the bottom back for a total of 4 MB... Now, the question is, when I disabled 2 MB, did I specifically disable one of these groups on the front or back, or they have been randomly disabled? How could I tell? I checked their temperature and they all seem to get equally hot...

All in all, I'm still glad that at least this card is now perfectly usable, although 800 x 600 will only work in just a few games now (probably those that only use double buffering?)

UPDATE:
I had some more time this weekend to play around and finally managed to fix the Voodoo 2 board. It was a very fun ride, so now I'm actually thinking of searching for more damaged Voodoo 1/2 cards and try to fix them 😀
Anyway, long story short, I narrowed the issue down to four FBI memory ICs, and using the very scientific "eeny, meeny, miny, moe" technique, I replaced two of those ICs and the issues were gone, I can now use the whole 4 MB of FBI memory without any glitches in GLQuake, Quake 2, or any other games.

I also replaced a few capacitors with new Panasonic low-ESR caps and in the following weeks will continue to replace them all.
The soldering looks decent enough (I think), even though this was literally the first time I ever replaced an IC. My intention was to first practice on a dead board but... I got carried away and just practiced on the Voodoo 😁

A few pics (the OKI memory ICs are the original ones, the EliteMT ICs were purchased on eBay):

Nice job 😎. How exactly did you desolder the defective chips and solder the new ones on?

Thank you!
Desoldering was very quick and easy with a pair of tweezers, kapton tape and a pretty cheap ($20) industrial hot air gun.
For soldering I also used a cheap Basetech ZD-99 soldering "station" with a 3 mm flat tip.