dulu wrote on 2024-08-30, 17:20:

Based on these symptoms, the service -man claims that the reflow was 100% successful and the problem lies elsewhere. The only thing that comes to his mind is to replace the aluminum capacitors

Depends on who made the card... Some of them need their capacitors replaced, some come with good quality parts... Re-flowing can bloat the capacitors. There's no way to tell without looking at the card.

I suppose BGA reflow is ok where they used leaded solder...

I can't say if the GPU chip is damaged already or not... but it sure is degraded. The fact that you can keep the card from crashing with good cooling and a slight underclock is somewhat promising, though.

My personal opinion is that most stock coolers/heatsinks are shit anyways and should be replaced with the get-go with something better. They are usually specced to cool the GPU just enough so that it lasts for the "design life" of the card, which often is just a few years. Of course, many GPUs lasted much longer (especially some older ones) mainly because they were build on large litography node and all in all not pushed to the edge like modern silicon is.

So yes, just get a much bigger cooler for it and do a slight underclock on both the GPU chip and RAM. Try 10% reduction in the clocks first, then 15%. And if still unstable, go down 20%. Yes, you will lose some performance, but what's better? - A fast GPU that crashes after a few minute or a slightly slower one that works OK. Either way, even 20% reduction won't be that noticeable of a hit if you use the video card for much older titles accordingly. Once you find a clock that works stable enough on the card, maybe also consider flashing the card with these lower clocks so that they stay the permanent default clock going forward.

The voltage of the Ti500 is indeed pushed to the limit for the Geforce 3 at 1.65v. The Ti200 runs at 1.52v to 1.55v and the original GF3 runs at 1.5v on the dot. This is from my measurements of all the different GF3 cards I have and calculating it using the pair of resistors that make up the GPU's VRM resistor divider.
However it's not a crazy voltage, the Geforce 4 Ti4200 cards that I've seen run between 1.65v to 1.66v and I expect that the Ti4600 runs at a higher voltage than that.

What I have found is that these cards all had thermal pads for cooling for a *reason*. That being that the BGA package has a plastic outer circle and a metal inner circle, that metal circle is inset by maybe 0.2mm and that's a distance that thermal paste doesn't do well with. See the contact on the GF3 in this thread, the paste in the metal circle is thick and that's not ideal: Geforce 3 Ti200 Overclock
I have had the same heating issues with one of my GF3 Ti200 cards after repasting it and putting on a larger heatsink, it would crash after a short time trying to game on it because there was too little paste on the middle.
Your pictures look like a GPU core clocked too high type crash, to me.

What I do on all of these metal circle BGAs now is get some 0.1mm or 0.2mm copper and cut that in a circle to fit into the metal circle area. Then paste the GPU, put the copper shim into place and put more thermal paste on that, then put the cooler on.
That resolves the airgap on all these metal circle BGA chips I've come across and in use I don't think they really need more than the heatsink they come with.

Hope that helps, what you're describing says to me it's a GPU core thermal issue, but it might be worth checking the GPU vcore is correct and that the capacitors feeding it in the upper left of the card by the bracket are all functioning properly.
Sometimes the resistors that set up the voltage regulator output voltage can go bad so they should be checked out of circuit against their markings if the vcore is higher than 1.65v.

But the thermals are the first item to check 😀 See how the paste looks when you take off the heatsink, if you put a thin layer on the GPU, does that only spread on the corners and stay untouched in the middle?

edit: I forgot how overbuilt the voltage regulation is on the Ti500 with the poly tantalum capacitors, it's probably *not* capacitors if your card is a reference design. The poly tantalum capacitors are the little black rectangles and they're quite high quality and not susceptible to degradation like electrolytics are. The GPU vcore side doesn't have a purple cap so not one to fail there either. The purple caps on the end of the card are for the memory rails.

endor is creative. After reflow there is no visible signs of high temperature. Even the sticker on the bottom is still white and has not burned. GF3 is 2001, 100% lead solder. service-man confirmed this because he used a solder profile for leaded solder.
n my opinion, if the GPU needs a non-factory heatsink to function, then it is damaged and has no value from a collector's point of view.
As for the metal ring, I was also wondering about it. Interesting observation.
Update. I installed junk cooling out of curiosity. Copper core, fan about 50mm. The card has been working in 3D for an hour now - with stock clock.

dulu wrote on 2024-08-30, 17:20:

The question is whether anyone has experience with professional repair of cards from that period and whether these symptoms tell him anything. Is the core permanently damaged? The problem is clearly dependent on the GPU temperature. I had identical symptoms with the radeon 9800PRO. It hung on the stock cooler, but after installing water cooling it worked normally. I'm sure it would look identical here.

This is interesting - the two cards are actually very very different in the kinds of problems they face. The Geforce 3 is a classical BGA chip with I think gold bond wires connecting the silicon to the BGA lead segment. These are much more robust and I think will hold up better long term than the cards like the Radeon 9800 and later will.
The Radeon 9800 is a flip chip which connects the silicon to the BGA package with even tinier solder balls called bumps and is held in place with I think glue? Over time these bumps and glue can separate a bit leading to bad connections which is why you'll find most Radeon 9700 / 9800 cards are broken now (they were complex to start with).
It was a problem that we now call 'bumpgate' with video cards affecting Nvidia and ATI, breaking lots of Geforce 6000/7000/8000 series cards as well as ATI cards and the Xbox 360 / PS3 chips until the recipe for attaching silicon to BGA was improved.

I've got similar experience with a flip-chip card, my Geforce FX 5700 LE card would crash in games almost immediately, until I took the spring clips for the heatsink off and replaced them with plastic M3 nuts & bolts. Now the card works happily again: Re: What retro activity did you get up to today?
Probably only a temporary resolution but I'm happy it's working again for now.

Haven't had any problems like that with GF3 / GF4 cards.

The card is actually works.

I took the second, functional ti500 which I hadn't used for a long time. Surprise - the same symptoms, the same hang, after the same time. I connected the reballed GPU to the computer on which I used working ti500 in the past. Everything works fine

It remains to be explained why both cards hung on the GA-BX2000+. It looks like service-man and Thermalwrong were right. Hang indicated that the overclock was too high, which is equivalent to too low voltage. This is also partly explained by the connection with temperature. As the silicon temperature increases, its resistance increases, so the card needs more current to power it and, at the same time, a higher voltage.
Both gpus started working normally on bx2000+ after closing the JP20 and JP21 jumpers, described as "Enable for Voodoo3 card". Their operation is described in thread below.
Gigabyte GA-6BA and Voodoo 3 jumpers
When enabled, the gpu gets 3.3V directly from the PSU. This was enough for both GPUs to stop hanging.

dulu wrote on 2024-08-31, 12:24:

The card is actually works. […]
Show full quote

The card is actually works.

I took the second, functional ti500 which I hadn't used for a long time. Surprise - the same symptoms, the same hang, after the same time. I connected the reballed GPU to the computer on which I used working ti500 in the past. Everything works fine

It remains to be explained why both cards hung on the GA-BX2000+. It looks like service-man and Thermalwrong were right. Hang indicated that the overclock was too high, which is equivalent to too low voltage. This is also partly explained by the connection with temperature. As the silicon temperature increases, its resistance increases, so the card needs more current to power it and, at the same time, a higher voltage.
Both gpus started working normally on bx2000+ after closing the JP12 and JP13 jumpers, described as "Enable for Voodoo3 card". Their operation is described in thread below.
Gigabyte GA-6BA and Voodoo 3 jumpers
When enabled, the gpu gets 3.3V directly from the PSU. This was enough for both GPUs to stop hanging.

Ohh you've just reminded me of a similar fault I had testing Geforce 3 cards on an Abit VH6T motherboard that I'd partially recapped - turned out I hadn't recapped enough and the capacitors for the AGP slot power weren't holding up, it'd crash getting into Windows 98, especially with the Ti500 rather than the Ti200.
It may be worth looking at the caps inbetween the PCI / AGP slots to see if any appear to be bulging or leaking, but great that you've otherwise resolved it 😀