Look very closely for any bent pins on the 3DFX chips that might be in contact with a neighboring pin or have broken soldering. The RAM chips may also have broken soldering.

Regarding thermal issues, adding heatsinks to the TMU and FBI chips on Voodoo cards helps.

I did run the Tomb Raider 3D Demo on Diamond Monster CD. Looked fine.

I also applied pressure to all bigger chips while the card was running hot and 'triangulating', but nothing noticably changed.

Today, I had another look at the card and I noted something peculiar, compared to the Voodoo². My Voodoo 1 gets HOT when it does nothing. The medium sized chip on top gets too hot to touch while the card is minding it's own business. The Voodoo² chips get warm in idle. But nowhere hot. If there are V1 owners around, could you please check if your cards get really hot in idle?
I have looked over the whole card with a +3 pair of reading glasses, but it looks like new.
Sound to me as if either the voltage is too high or if some signal is alway active.

It's normal for the TMU to get quite hot while in idle. Voodoo2 lowers the FBI/TMU clock significantly while in idle (usually 30MHz) , V1 doesn't do that.
The only power rail on the card is the 5V one. There is no regulation on the card, this comes directly from the PCI slot. Maybe check with a multimeter the 5V rail in your PSU.
Even if your card looks perfectly fine, try poking all the FBI/TMU pins with a needle. I had cards that looked fine even under the microscope, but some pins weren't actually connected.

sdz wrote on 2024-06-20, 21:28:

It's normal for the TMU to get quite hot while in idle. Voodoo2 lowers the FBI/TMU clock significantly while in idle (usually 30MHz) , V1 doesn't do that.
The only power rail on the card is the 5V one. There is no regulation on the card, this comes directly from the PCI slot. Maybe check with a multimeter the 5V rail in your PSU.
Even if your card looks perfectly fine, try poking all the FBI/TMU pins with a needle. I had cards that looked fine even under the microscope, but some pins weren't actually connected.

OK, so the chip being hot is not a symptom of the problem, but perhaps the trigger. Maybe due to thermal expansion?
Does it make sense to lower the card frequency to the min of 16 MHz while not using it? So that it doesn't literally glow in the dark all the time and heat up the neighbors?

As I have this issue with the card on various machines, I can rule out the 5V line. I even measured PSU voltages to check that end.

Poking leaves a lot of room for interpretation. So I take a steel needle and then do what? Try moving pins left and right by pushing a needle into the gap between pins? Try lift them? How much force is enough to test but not to cause the issue?

As the upper chip is the hottest, I would start there.

sdz wrote on 2024-06-20, 21:28:

try poking all the FBI/TMU pins with a needle.

I did. Pushed a needle tip between the pins and slowly moved it downward until it got stuck. I expected a loose pin to move sideways at the bottom, but none did.
However, the card has stopped triangulating for now. It seems to be working OK again. Nice. Thanks for the idea.

ux-3 wrote on 2024-06-21, 19:23:

I did. Pushed a needle tip between the pins and slowly moved it downward until it got stuck. I expected a loose pin to move sideways at the bottom, but none did.
However, the card has stopped triangulating for now. It seems to be working OK again. Nice. Thanks for the idea.

That would suggest you didnt notice a loose pin, but moving it made the connection temporarily stronger. Proper solution is resoldering all pins. Another option is loose connection on one of ram chips.

rasz_pl wrote on 2024-07-03, 20:05:

That would suggest you didnt notice a loose pin, but moving it made the connection temporarily stronger. Proper solution is resoldering all pins.

Too many, too risky. Should the problem return, I will repeat the procedure, but with only half a side of one chip. Then test again. That way, I can hopefully nail it down to a few pins and resolder those.

ux-3 wrote on 2024-07-03, 20:52:

rasz_pl wrote on 2024-07-03, 20:05:

That would suggest you didnt notice a loose pin, but moving it made the connection temporarily stronger. Proper solution is resoldering all pins.

Too many, too risky. Should the problem return, I will repeat the procedure, but with only half a side of one chip. Then test again. That way, I can hopefully nail it down to a few pins and resolder those.

Its the other way around, leaving dangling pins and hoping just wiggling them will somehow fix it will eventually lead to latch up (https://en.wikipedia.org/wiki/Latch-up) and fry your voodoo. Resoldering both chips and ram for good measure is the only proper course of action.