I had no luck with the 6600GT either... The GPU chip could be defective, or it could be the bridge chip. The 6600GT is native PCIe, so that's why it needs a bridge chip.

Do you have a magnifier that you could use to inspect the solder joints for the power connector pins? If one has a hairline crack it could be unreliable and could move as the temperature changes.

Or, if you are proficient with soldering you could reflow those joints just to be sure (you'll probably need to preheat the board a bit since it likely has lead-free solder).

mockingbird wrote on Yesterday, 19:00:

I had no luck with the 6600GT either... The GPU chip could be defective, or it could be the bridge chip. The 6600GT is native PCIe, so that's why it needs a bridge chip.

I encountered cards with defective GPU chip and AGP bridge chip. If the GPU chip is defective, I usually just get artifacts and immediate crashes. If the AGP Bridge is defective... I usually get no screen at all. But yeah, the seller confirmed that it did work with the driver installed and passed benchmarks. Then it remained into storage fora bit before I got it. It wouldn't be the firs time that I encounter problematic hardware that was working before entering storage.

Ozzuneoj wrote on Yesterday, 19:05:

Do you have a magnifier that you could use to inspect the solder joints for the power connector pins? If one has a hairline crack it could be unreliable and could move as the temperature changes.

Or, if you are proficient with soldering you could reflow those joints just to be sure (you'll probably need to preheat the board a bit since it likely has lead-free solder).

A magnifier is somethig that I should get, becase at a naked eye I don't see any issue but...
Unfortunatelly I'm not proficient (yet) in soldering.

Anyway, I managed to get another PSU out and, now, the card just straight up crashes after installing drivers. It won't even load into windows. Just crashes right after the load screen. And if I load into safe mode... The screen is full of artifacts.
The other 6600GT I own, the lucky card among many fallen soldiers, still works no issues even with this third PSU.

Guess that this is a tale as old as time itself: Functioning vintage hardware that stops working after being in storage for some time. But I'll keep trying a little bit more.

Small update that warrants another answer:

After going back to the second PSU, the one that I had more success with this card, the artifacts slowly disappeared one reboot and one swap of the connector after another. The card still shows issues, mind you, not 100% stable, with that "power issues" message popping up from time to time. As I'm writing this, I'm well into the "lobby" scene in 3DMark01, never reached it with this card before. It usually showed that message well before completing the first run of "Dragothic"

But I did notice something peculiar: The smaller artifacts would appear and disappears in unison with an acute wistles from the card. It did those wistles from the moment I started testing it and, at the beginning, I just thought it was the music of an old fan.
But now? With problems and artifacts going in unison with those squeeks and wistles? Maybe there really is something wrong with some capacitors or some solder joints on the card. Maybe even tinuy cracks or just oxidation from being into storage for some time.
I'll keep doing more tests because, when I finished writing this, I reached the end of 3DMark01 for the first time ever with this card, scoring a respectable 17541 points, in line with the other 6600GT I own. And no apparent artifacts showing up.

EDIT 1
I don't know how simply disconnecting and reconnecting to the GPU the molex connectors I have on this PSU, alternating bewteen the two of them, made the card go slowly from photo 1 to photo 2, with the error message reappearing briefly before itself going away too. And the squeek is now gone. On the way to complete a second run of 3DMark2001 before shutting everything down for the evening and coming back at it tomorrow.

EDIT 2
I've passed more 3DMark2001 runs with similar scores, even after rebooting the system and even after disconnecting and reconnecting the MOLEX plug again.
I'm starting to think I may be jinxed or something with these late AGP era cards...

Nice! So it's working now?

Maybe there was some kind of stubborn (and invisible) oxidation or gunk on one of the pins on the molex connector. Some PSUs may have connections that clamp tighter than others, so I guess that could explain the difference between power supplies... though what you've described isn't something I have personally experienced.

If you reconnected the cable multiple times it could have cleaned the connector enough to make it work without getting flaky.

Now, I wonder... does the card work with the other power supplies too?

Ozzuneoj wrote on Yesterday, 21:15:

Nice! So it's working now? […]
Show full quote

Nice! So it's working now?

Maybe there was some kind of stubborn (and invisible) oxidation or gunk on one of the pins on the molex connector. Some PSUs may have connections that clamp tighter than others, so I guess that could explain the difference between power supplies... though what you've described isn't something I have personally experienced.

If you reconnected the cable multiple times it could have cleaned the connector enough to make it work without getting flaky.

Now, I wonder... does the card work with the other power supplies too?

Well, yeah, it is working. Meaning the silicon on the PCB is good: The GPU, the memory and the AGP bridge are good, because even afte unplugging the MOLEX and reconnecting it again to the same connector, the card still works and passes 3DMark2001 on repeat.

The weird thing, though, isn't that things got gradually better while I kept switching between connectors and PSUs, but that they went gradually worse and then they improved to no problems at all. I was sure the card was a goner when it started artifacting heavily. But that squeeky sound that now disappeard... Yeah, maybe there was something electric going on between the connections.
Anyway, tomorrow morning I have to test the actual itemI went to pick up from this seller, the motherboard, but before testing that I'll first try switching again connectors on this PSU, then test again with on of the other two PSUs that showed problems before with this specific card. At this point I think that, it it wasn't oxidation or gunk, it's probably a defective connector on the card. Something I believe should be easy to fix for a professional technician.

Try washing it.
Seriously. Soap, water, toothbrush, dry thoroughly.

Do you have Deoxit? If so, use it on both the AGP and Molex connections. I agree also reflowing the Molex solder joints might help.
If you have a multimeter, monitor the voltages from an unused molex and see if the 5V or 12V are sagging under load.

Does anyone know how these cards detect a power fault? Can that error be caused by weak power from the motherboard, or is it only detected from the molex side? Is it based on sensing 1 or more of the voltages somewhere on the card? Is it sensing the supply voltage(s) or output(s) from an onboard regulator?

So, typically, the card gets power from the slot, and from the plug.
Nvidia were trixy and they can output a standard VGA display in my experience with completely fd ram, and only slot power.
Which is enough to show a "needs more power" error, get into windows and code 43 IF a proper driver tries to load.

Presumably this means the core has a tiny vga only section that runs from the slot, and the 3d part runs from the plug VRMs.

In this case, I personally suspect something's wrong with the VRM circuit from the plug, but it could just as easily be random semiconducting dirt somewhere causing a partial short, and/or high resistance plugs.
It can be the detection circuit too, or a bad pad, or bad solder/infill (we are talking about the start of bumpgate here), or even a bad cap (smd or electrolytic).
It IS pretty old now.

Hence... give her a wash and see what happens.

It can't hurt to log the reported voltages in afterburner or something too. Without proper tools it's a shitshow.
Best an amateur can do is wash, and if that fails, gamble that the electrolytics/polymers are bad, replace them and hope for the best. At least they're fairly easy.

I fixed about a dozen gpus and half of those were from a wash. Most of the others were VRM related. Usually swap the mosfet and all good. One... I think was SMD caps on the PCIE lanes. I forget. Anyway I had a solid crack at learning.
I failed a bunch. Some were just doomed. Like the 1080 with the hole through the board. Doubt the core survived that.

myne wrote on Today, 07:31:

Try washing it.
Seriously. Soap, water, toothbrush, dry thoroughly.

shamino wrote on Today, 10:06:

Do you have Deoxit? If so, use it on both the AGP and Molex connections. I agree also reflowing the Molex solder joints might help.
If you have a multimeter, monitor the voltages from an unused molex and see if the 5V or 12V are sagging under load.

Does anyone know how these cards detect a power fault? Can that error be caused by weak power from the motherboard, or is it only detected from the molex side? Is it based on sensing 1 or more of the voltages somewhere on the card? Is it sensing the supply voltage(s) or output(s) from an onboard regulator?

myne wrote on Today, 10:23:

So, typically, the card gets power from the slot, and from the plug. Nvidia were trixy and they can output a standard VGA displ […]
Show full quote

So, typically, the card gets power from the slot, and from the plug.
Nvidia were trixy and they can output a standard VGA display in my experience with completely fd ram, and only slot power.
Which is enough to show a "needs more power" error, get into windows and code 43 IF a proper driver tries to load.

Presumably this means the core has a tiny vga only section that runs from the slot, and the 3d part runs from the plug VRMs.

In this case, I personally suspect something's wrong with the VRM circuit from the plug, but it could just as easily be random semiconducting dirt somewhere causing a partial short, and/or high resistance plugs.
It can be the detection circuit too, or a bad pad, or bad solder/infill (we are talking about the start of bumpgate here), or even a bad cap (smd or electrolytic).
It IS pretty old now.

Hence... give her a wash and see what happens.

It can't hurt to log the reported voltages in afterburner or something too. Without proper tools it's a shitshow.
Best an amateur can do is wash, and if that fails, gamble that the electrolytics/polymers are bad, replace them and hope for the best. At least they're fairly easy.

I fixed about a dozen gpus and half of those were from a wash. Most of the others were VRM related. Usually swap the mosfet and all good. One... I think was SMD caps on the PCIE lanes. I forget. Anyway I had a solid crack at learning.
I failed a bunch. Some were just doomed. Like the 1080 with the hole through the board. Doubt the core survived that.

I'll treasure all these infos, and I bet they'll be useful in the future for anyone else experiencing this problem and looking for solutions on search engines.
But even now, after keeping the system off for well over 12 hours, the card just works. I seriously don't know what could've been wrong, especially since cleaning the PCB and the connector with 90% isopropyl was one of the first things I tried and didn't solve the issue, and it kept getting worse the more I tried the card... Until it didn't got worse and magically fixed itself.

Now, I'll probably do the definitive test: I need to take this system off the test bench to test the MOBO I went to originally pick up from this seller (An ABIT KV7 that it was tested the day before picking it up). After confirming that the mobo POST and it's stable, I'll reassemble the current test system (an ASUS P5P800SE with a Cedar Mill 3.4ghz P4) and see if that 6600GT still works. If it does, it's a mistery. If it doesn't, then there's something wrong with the connector on it and it probably need replacing.

I wonder how easy are they to replace and if they can even be modded to use a 6pin PCIe connector.

The solder looks fine on the board. It practically looks new.