Question GPU caught fire, what happened? (pictures)

[teknow]

Hello,

as you can see from the pictures below, the GPU failed catastrophically and I was wondering what might have caused it. The build is obviously a few years old (Nvidia GTX760), but not ancient. It has worked without any issues so far and this happened completely out of the blue. The computer was not being stressed hard when it just shut down on its own. I looked around for a minute to see if there's anything weird that might explain it and noticing nothing I turned the computer back on to see what's up. As soon as I pressed the ON button flames burst out of the GPU, I saw the orange flickering lights and heard the cracking, fizzing noise of components burning. I cut the power asap and luckily everything stopped.

There was good cooling with plenty of airflow, nothing on this rig was ever overclocked or tuned above factory defaults. The components are all from what are supposed to be reliable brands (Asus mobo and gpu, XFX psu). Obviously there's some dust, but the GPU closeup photos are somewhat deceiving I'd say, it doesn't look that dusty in person. See the orange metal part on the GPU, or the mobo pictures. That's closer to how things look to the naked eye. Managing computers of friends and family this one is easily cleaner than most. I can't really see a good reason for such a violent failure and I've never seen this happen before.

Is there anything to be gathered from what part of the GPU caught fire? What about the mobo? As you can see it has clearly been blackened by the fire, is it likely damaged as well or could it just be dirtied? I can try it without a different GPU since the onboard video works, or at least did work before this mess. Everything except the CPU and RAM I already disconnected but I was still a bit weary about turning this on again. I'm guessing even if it works fine with onboard video, trying a different GPU on it would be risky as the PCIe slot might be damaged, right? Can we at least say if the problem originated with the GPU itself, or could some other issue have caused the GPU to burst into flames? What's more likely?

Thanks for reading. BIG pictures:

 

[Leeea]

Reminds me of the good old days when nvidia chips would spontaneous melt down back in the mid-2000s.

. . .

That looks like it happened right under a capacitor. Staring at it, said cap does not appear to have burst, but rather the board opened up underneath it. The cap filters the power before it is fed to the GPU, most of the amps pass right through that area of the pcb.

note: the above post was edited many times

 

[Stuka87]

Hard to tell for sure what the chip was. Its an 8 leg chip, so could be a rectifier. Looks like it either internally shorted, or had too much current drawn through it.

 

[Leeea]

Hard to tell for sure what the chip was. Its an 8 leg chip, so could be a rectifier. Looks like it either internally shorted, or had too much current drawn through it.

I do not think it was the surface mount chip.

I am going to argue the way the pcb opened up underneath said component, the heat came from from the other side. The main hole is north east* of that chip also, it looks to me that it was just unlucky to be in the area.

*https://forums.anandtech.com/attachments/gpu-closeup-2-png.35018/

Took at how the pcb layers burst toward the camera just north east of that square component.

-----------

I originally though those black screws were the gpu heatsink retention screws, but on zoom how that is not the case. The four gpu screws are to the left. My guess is they are holding a heat sink against the VRMs on the other side.

Thanks for reading. BIG pictures:

Could you give us a picture of the other side of the graphics card? Maybe remove the heatsink?

 

[teknow]

Could you give us a picture of the other side of the graphics card? Maybe remove the heatsink?

Can't do that now (not near the card), so I did the next best thing and found some nice pictures (both sides, with and without the heatsink) of this same card when it's not burned. Here they are. Hopefully it's of some help, if not I should be able to post the pictures of my card without the heatsink later.

 

[fleshconsumed]

Reminds me of the good old days when nvidia chips would spontaneous melt down back in the mid-2000s.

Huh.... I had EVGA GTX260c216 burn out on me after a few weeks/months of usage. EVGA replaced it under warranty. I had no idea there was a widespread issue with nvidia cards back then.

 

[Leeea]

Can't do that now (not near the card), so I did the next best thing and found some nice pictures (both sides, with and without the heatsink) of this same card when it's not burned. Here they are. Hopefully it's of some help, if not I should be able to post the pictures of my card without the heatsink later.

It looks like the hole in the bottom of your board is directly under the 7805* 2M01 voltage regulator. I only see one voltage regulator on the board, so that likely supplies all of the vrm circuits. It also has a heat sink placed directly against the pcb right there.

*https://en.wikipedia.org/wiki/78xx

 

[Stuka87]

I do not think it was the surface mount chip.

I am going to argue the way the pcb opened up underneath said component, the heat came from from the other side. The main hole is north east* of that chip also, it looks to me that it was just unlucky to be in the area.

Its possible one of the main VRMs failed, they are right on the other side of the board, and pull enough current to burn through the board if they internally shorted.

I wasn't sure if it was one of the two chips there that failed, or something on the other side. Since we don't have a photo of the top side of the board (photos show the bottom side). The VRMs are up top, and now that I notice the hole in the PCB to the top right of those two chips, chances are it was one of VRMs.

 

[Leeea]

Huh.... I had EVGA GTX260c216 burn out on me after a few weeks/months of usage. EVGA replaced it under warranty. I had no idea there was a widespread issue with nvidia cards back then.

It was a massive multiway lawsuit fest with Apple, Dell, HP, and Compaq class action lawsuits. Most of the damage was in OEM systems. Then people realized it was nvidia's fault, and nvidia fought them before years, for finally admitting wrong* doing in 2010 and settled.

*It was discovered in court that nvidia knew they had a problem and denied it

 

[teknow]

Well I removed the heatsink now, but I didn't even bother taking a picture because the top side looks like new, much to my surprise given the noticeable damage of the PCB on the bottom side. There's really no trace of any damage whatsoever there. If you look at this picture, the fire was not directly opposite the bottom of the 3 big capacitors, but roughly in the middle between that one and the 7805 element below it. On my vertical closeup picture you can see the capacitor's protruding contacts north of the damage. Those two contacts are considerably off center, meaning most of the capacitor is further north still.

Any idea why the system simply shut down on its own at first and only when I turned it back on a minute later the GPU instantly caught fire? Obviously better that it happened this way as opposed to bursting in flames while running, if for no other reason than ensuring I was right there when it happened, but there's no reasonable way I could have predicted this would happen and avoided turning it back on, was there? And most importantly, how confident would you be about the other components, mainly the mobo (blackened in a spot as pictured) and the PSU? I wouldn't want to fry something else trying them out, although, admittedly lacking some relevant experience/knowledge, it feels like they should have a good chance of working fine.

 

[Leeea]

If you look at this picture, the fire was not directly opposite the bottom of the 3 big capacitors, but roughly in the middle between that one and the 7805 element below it.

That sounds like where the pin pad for the 5v 7805 voltage regulator sits/goes into on the pcb.

Any idea why the system simply shut down on its own at first and only when I turned it back on a minute later the GPU instantly caught fire? O

Someone will likely post something in a moment that makes this speculation seem foolish:

The solder joint on the 7805 regulator aged out prematurely. As it failed the resistance went up, creating more heat for the same wattage / amps. Eventually the slow build up of heat reached the melting point of the solder causing it to reflow up the regulator pin reducing contact area. This greatly increased resistance and heat, but as this failure took a while to occur, the pcb under the 7805 was already very hot. This increase of heat triggered the 7805's thermal safety cutout shutting down your computer.

The 7805 thermal protection is targeting the metal heatsink on the 7805, not the pins going to the pcb. It does not care about the temperature of the pins or pcb, just the temperature of its own heatsink. Even though the resistance is climbing, the 7805 will just continue to put more power down the line to counteract this, as it is designed to do.

On reboot the packaging and metal heatsink of the 7805 had cooled down, but the now failed solder joint transferred a great deal of heat directly to the pcb. The pcb could not transfer the heat away quickly enough to trigger the 7805's cut out, and the end result is visible in your pictures.

but there's no reasonable way I could have predicted this would happen and avoided turning it back on, was there?

999/1000 would not have caught it.

And most importantly, how confident would you be about the other components, mainly the mobo (blackened in a spot as pictured) and the PSU?

You have good quality components. I would be very surprised if any of them had any issues.

You have an Intel iGPU* right? Where you can plug a monitor right into the mainboard? Do that and start it up, you have nothing to lose.

*low chance you will have to do a CMOS reset if you intentionally disabled the iGPU in BIOS as many do
*what CPU are you running? Most Intel CPUs have an iGPU.

 

[teknow]

Someone will likely post something in a moment that makes this speculation seem foolish:

That actually sounds like a pretty sensible explanation to me, then again I'm no expert.

You have an Intel iGPU* right? Where you can plug a monitor right into the mainboard? Do that and start it up, you have nothing to lose.

*low chance you will have to do a CMOS reset if you intentionally disabled the iGPU in BIOS as many do
*what CPU are you running? Most Intel CPUs have an iGPU.

I do have an iGPU yeah, tried and enabled as I was occasionally using it to drive an additional screen. Testing what's left using the iGPU was my plan already, I just wanted to check if hopes of everything else still working aren't clouding my judgment pushing me towards an easily avoidable issue. I already disconnected all the peripherals the day of, so that nothing non essential would be exposed should anything else fail.

 

[Mopetar]

I don't have anything to add, but I hope the next episode of CSI: AnandTech is just as gripping as this one was. The analysis of the problem really got me fired up.



YEEEEAAAAAHHHHHHH!