This is the main problem of the top NVIDIA video cards of that time.
And no conspiracy theories.
They all dried out completely, overheated and burned.

No conspiracy theories here - Xenos is one of the prime examples of bumpgate related failures and it was an ATI chip, based on something from x1000 series IIRC. This is pretty much the only fact i am aware of - everything else would be speculation. What was defective and what was not is unknown, the same as it is unknown for nvidia - lists from well known lawsuit(s) are most definitely incomplete.

Dried thermal compound would have played a role once the stuff was old enough (several years at least), but also it usually is not as critical as people may think. I've seen a lot of different hardware with 20+ years old thermal compound and my opinion is - all the horror stories about it are no more than myths. Thermal transfer is hindered, but more often than not - inconsequentially so. You'd see something like a 5C difference, or with temperature controlled fans - slightly higher fan RPM and that's it.

Not to mention that what people often perceive as "dried thermal compound" can simply be very high viscosity thermal compound from the factory...

I am yet to see an example which would change my mind about it...

I would agree with you completely if we were talking about Tualatin.
No energy-saving technologies, the temperature is always 45 degrees regardless of the load and 30 watts of consumption.

But GT(x? I don't remember all these NVIDIA prefixes)560, which I personally had to delidded a year after the warranty expired, completely disagree with you.

All my 9800 GT cards have broken fan regulation. Not only they run fan at like 35-40% regardless of temperature, fan control is inverted. If you set it to 20% the fan runs faster. Re-pasting has lead to 4-5'C temperature reduction. I didn't delid any of them. Some had thermal compound that was completely solid, more like burned to the die, others had thermal compound similar to MX-2, very hard to remove. Keep them at about 60'C while gaming and they should last.

Back then people didn't use mesh cases and case airflow was very poor.

shevalier wrote on 2025-08-09, 07:26:

I would agree with you completely if we were talking about Tualatin.
No energy-saving technologies, the temperature is always 45 degrees regardless of the load and 30 watts of consumption.

But GT(x? I don't remember all these NVIDIA prefixes)560, which I personally had to delidded a year after the warranty expired, completely disagree with you.

Well, it is completely possible and likely that i simply have not seen cases severe enough. And then it can lead to more subtle issues like local overheating in areas which are no longer making good contact...

But i've seen so many people break their stuff trying to fix something that was not broken in the first place that i am kind of allergic to whole idea of thermal compound related "maintenance".

At least with modern hardware there is no need to guess anymore or replace thermal compound "just in case", ability to monitor temperatures including hotspot makes decision when to delid or just repaste very easy to make.

AlexZ wrote on 2025-08-09, 08:24:

All my 9800 GT cards have broken fan regulation. Not only they run fan at like 35-40% regardless of temperature, fan control is inverted. If you set it to 20% the fan runs faster. Re-pasting has lead to 4-5'C temperature reduction. I didn't delid any of them. Some had thermal compound that was completely solid, more like burned to the die, others had thermal compound similar to MX-2, very hard to remove. Keep them at about 60'C while gaming and they should last.

Back then people didn't use mesh cases and case airflow was very poor.

Yep, this has been pretty much my experience as well. Pull apart a card with 20 year old completely solidified thermal compound, replace it, get ~5C improvement.

Also somehow i hate modern "mesh cases" and cases built out of fans. Over9000 fans blowing in random directions through overplayed mesh+dust filter do not result in good airflow. This becomes very evident looking at how industrial systems, or even just gaming consoles are designed.

It is possible to get great airflow with cooling better than on open stand with just 1-2 correctly positioned fans in a case which is properly designed and is not made out of mesh. It is impossible with mesh though, so mesh is actively harmful in many situations.

All my 9800 GT cards have broken fan regulation. Re-pasting has lead to 4-5'C temperature reduction. I didn't delid any of them

That's nonsense. G92 cards had no lid.

Archer57 wrote on 2025-08-09, 08:52:

But i've seen so many people break their stuff trying to fix something that was not broken in the first place that i am kind of allergic to whole idea of thermal compound related "maintenance".

Likewise, considering that people don't care whether it's an industrial Honeywell or a GD900 from AliExpress.
Most often, they can't even get the temperature they had before disassembling.

But with the GT560, there were no options. The video cards were from the kids of colleagues who were hanging out in online games.
And they glitching and/or turned off.

P.S. I'll be honest about myself.
Of my retro PCs, I can hang out for a few hours in a few games on Turion + Vortex.
The rest - well, they turn on sometimes.
But not for hours-long battles in Stalker and FEAR.
Just to look at the interface, run 3Dmark once.
With such use, well, what kind of overheating is there.

The Serpent Rider wrote on 2025-08-09, 09:14:

All my 9800 GT cards have broken fan regulation. Re-pasting has lead to 4-5'C temperature reduction. I didn't delid any of them

That's nonsense. G92 cards had no lid.

Mine certainly don't. I haven't see all of them that were made. Picture above in the thread does show de-lidding, maybe for a different core.

The Serpent Rider wrote on 2025-08-09, 00:11:

Depends on who you trust more. According to findings from Felix, 55nm should be fine, because Nvidia has started fixing their shit somewhere during 65nm production. So any late GTX 260 that you can identify to have a die shrink 55nm chip.
Byt according to old articles from Semiaccurate all 55nm chips are still cursed and Radeon should be safer choice.

Is the Felix you're talking about the one who made this video? https://www.youtube.com/watch?v=I0UMG3iVYZI
I'm watching it now.

Discussions about Bumpgate are somewhat pointless in the context of retro hardware.
These products have long been on the right side of the graph, while soldering problems are considered "infant mortality failures."

At that time, there were many surprising problems associated with the transition to lead-free solders.
- tin pest
- tin whisker
- ball crack
- replacing HASL with organic solderability preservative (OSP)
- uncompatible flux, solder and PCB coatings
- A banal saving on electricity when melting the solder - the boards often did not heat up enough.

And none of that matters now.
Either you have a working video card that you can buy, or you don't.
There won't be any other video cards like that.

Bumpgate has absolutely nothing to do with lead-free solder though.

The reason why it still matters is that bumpgate-affected chips have significantly reduced lifespan, especially in certain conditions (so on certain cards). Even if you find completely new unused card it is still going to die within short period of time. Kind of the same thing as with capacitor plague - yes, any capacitors go bad over time, but stuff from that period is pretty much guaranteed to have dead capacitors while newer or even older stuff might not have that issue.

Of course if someone wants a card from that period they'll just have to buy one and hope for the best. But IMO it makes sense to consider the alternatives and especially - avoid paying a bunch of money for such cards. The simplest example is one discussed here - G92. IMO it makes sense to buy later G92 based cards, like GTS250, instead of early ones, because ultimately it is the same chip but later ones should be fixed.

I may be wrong here as everyone has their own goals, but IMO there is very little reason to mess with bumpgate-affected pci-e hardware. Go just a few years forward and it was resolved. It is not even going to be that unrealistic because replacing videocard without upgrading whole system is a common thing to do and system with videocard a year or two newer than the rest of the hardware are completely normal.

Sadly with late AGP there are no options - everything is going to be affected and there is basically no choice. Late AGP in general seems to be very cursed period for PC hardware - several large screw-ups/defects and said transition to lead-free solder on top...

And of course it only matters if the system in question is going to be practically used. If not it may be much less important...

Archer57 wrote on 2025-08-10, 05:28:

Bumpgate has absolutely nothing to do with lead-free solder though.

Is this due to the insufficient ductility of the lead-free solder, which cannot compensate for the expansion of the underfill material between the crystal and the chip substrate?
There are exactly the same problems there, because it is exactly the same technology.
And for the user it makes no difference whether the crystal has peeled off from the substrate or the chip from the board.

shevalier wrote on 2025-08-10, 06:25:

Is this due to the insufficient ductility of the lead-free solder, which cannot compensate for the expansion of the underfill material between the crystal and the chip substrate?
There are exactly the same problems there, because it is exactly the same technology.
And for the user it makes no difference whether the crystal has peeled off from the substrate or the chip from the board.

They did not use lead-free solder for bumps. In fact they were almost pure lead with resulting high melting point and everything.

The issue is that they used underfill with 70C Tg and then proceeded to set target temperature for cooling system above that. So that underfill softens/solidifies actively trying to tear bumps apart each cycle instead of supporting them.

Or at best if temperature is not going back and forth through 70C it just sits above that with underfill being soft and completely ineffective.

So the defective part here is underfill itself, not solder or soldering.

It actually is important what failed, because bumps are not fixable while if the issue is between the chip and the board the chip can be desoldered and soldered back ("reballing" which a bunch of repair shops offer, which basically is a scam because in 99% cases the issue is with bumps and it does nothing apart from making the chip temporarily work because of heat).

SapphireJet wrote on 2025-08-10, 01:21:

Is the Felix you're talking about the one who made this video? https://www.youtube.com/watch?v=I0UMG3iVYZI

Yes, there's a complimentary video that addresses common misconception about reballing the chips.

Archer57 wrote on 2025-08-10, 06:50:

They did not use lead-free solder for bumps. In fact they were almost pure lead with resulting high melting point and everything.

Classic lead-free eutectic, including for Flip-Chip Interconnects - 95.5Sn-3.8Ag-0.7Cu
Or the cheapest option - 99.3Sn-0.7Cu

They used almost pure TIN (Stannum - Sn), not lead (Pb) 😀

shevalier wrote on 2025-08-10, 12:17:

Classic lead-free eutectic, including for Flip-Chip Interconnects - 95.5Sn-3.8Ag-0.7Cu
Or the cheapest option - 99.3Sn-0.7Cu

They used almost pure TIN (Stannum - Sn), not lead (Pb) 😀

Nvidia used high lead bumps in defective chips. Which is 95% lead 5% tin IIRC.

I am not sure about ATI...

Though it does not really matter all that much in this case - underfill was defective which was the main issue.

Lead free bumps as a root cause is also a common misconception of the problem. It doesn't really matter, because even Jensen Huang admitted that it was an underfill problem.

I think my bump gate experiences include GeForce 7800 and 8500/8600/8800(G92). Possibly a 9500 too. I have not had problems with a G80-based card. Maybe the superior cooling with G80 cards is part of that.

With ATI the only thing that comes to mind is Xbox 360. Some of the 48x0 cards had such terrible cooling though that I would be hesitant to buy them. Some have no RAM or VRM heatsinks. The single slot 4850s often ran super hot, idling at around 80C and going above 100C in games. The single slot 8800GT had these extreme temperatures too.

G80 cares are definitely not safe with stock coolers, unless fan is ramped up to 100%.

I've never had a problem with any of the GPUs reported to be bumpgate effected, I thought it primarily ended up being chipsets effected?

The most unreliable line cards I've ever used has to be the GT200 series, like the GT240.