My GTX 1080 FTW worked fine until it BSOD'ed one day (out of the blue, no software or settings change) and ever since it hangs or causes bluescreens as soon as the Nvidia driver is loaded, either right away or when some application (like GPU-Z) tries to access the API. It always POSTs and also works in Windows with the default MS VGA driver, but API calls through the Nvidia driver cause it to crash.
 
The model number is: 08G-P4-6286-KR
The original BIOS (according to sticker): 86.04.60.40.82
 
It also sometimes fails to show up in the device manager or has an exclamation mark on some PCs, which made the card also inaccessible by the nvflash tool, but in one particular PC it seems to be reliable on the PCIe device level (except for the driver issue of course) and the BIOS is accessible for flashing. I thought it might be a corrupted BIOS (though both the master and slave BIOS show the same symptoms) and tried to give it a shot. It turned out the current BIOS was 86.04.60.41.82 (different/newer than what's on the sticker). I have the card second-hand, so maybe it was updated before, but in any case the card used to work fine. I downloaded and re-flashed both BIOS versions, which didn't change the behavior.
 
Then I tried out BIOS files for other revisions, in particular versions 86.04.17.X.X and 86.04.3B.X.X. These BIOS versions solved the freezing/BSOD problem, i.e. the Nvidia driver can load without issues, accessing the API (GPU-Z, Afterburner etc.) doesn't freeze and the card "works" in applications. However, that's because these BIOS versions are apparently made for a different memory vendor. GPU-Z says the memory type is "unknown" and the memory clock is stuck at 405MHz for any workload, making the card basically unusable. This 405MHz issue is a well-known symptom for using a BIOS that's made for the wrong memory type.
 
So apparently only the 86.04.60.X.X BIOS version will work with my card's memory type, but each (sub-)version I tried is causing it to crash with the Nvidia drivers. I also tried to remove the Nvidia drivers with DDU while having the "correct" BIOS version flashed - this allows GPU-Z to load without freezing and it shows the memory type as "GDDR5X (Micron)" (while with the "wrong" BIOS it's "unknown"), but it doesn't show bus width and bandwidth (which is probably normal without Nvidia drivers?). But again, as soon as the Nvidia drivers are installed, any application that uses the API or 3D function crashes it immediately.
 
The only cause I can think of is a hardware defect at the memory modules or controller, but it's strange why it still works more or less "normally" when it's locked to 405MHz when using the wrong BIOS. I'd like to test a forced very low memory speed with the correct BIOS, but I can't access tools like Afterburner or Precision XOC without freezing. Is there a way to do it directly via the driver right at the system startup? Can I do anything else to revive the card or is it a lost cause?

I might get desperate enough to try it. But can we be fairly certain at this moment that it's indeed a hardware problem and can't be solved with some memory tunings?

Sajin
Might want to try the oven trick on it.... https://www.overclockers.com/the-oven-trick-repairing-your-broken-video-card-with-an-oven/
 
Note: Bake the card with the gpu die facing upwards instead of down.

Link writes: "So, basically, once you have completed this operation and (possibly) fixed your video card, BUY A NEW OVEN."

Is this really true? Can't you possibly give it a thorough clean and keep using it?

RealNovgorod

JohnnyBftw
Link writes: "So, basically, once you have completed this operation and (possibly) fixed your video card, BUY A NEW OVEN."

Is this really true? Can't you possibly give it a thorough clean and keep using it?

I wouldn't use the oven for food afterwards regardless. You can't clean out all the toxic deposits from everywhere and you'll get scolded really bad by your wife or mom.
 
In my case, since it's a memory issue, I will try using a heat gun on the memory chips first with a thermocouple to monitor the temperature. There's not much hope for the baking method in any case, since it's almost never a problem of the solder joints but rather the bonding wires inside the chip.

This is silly. You're not heating the GPU enough to cause any outgassing of hazardous materials. You're in more danger eating a well done steak.

Sajin

JohnnyBftw

Sajin
Might want to try the oven trick on it.... https://www.overclockers.com/the-oven-trick-repairing-your-broken-video-card-with-an-oven/
 
Note: Bake the card with the gpu die facing upwards instead of down.

Link writes: "So, basically, once you have completed this operation and (possibly) fixed your video card, BUY A NEW OVEN."

Is this really true? Can't you possibly give it a thorough clean and keep using it?

That is what they are claiming. Buying a toaster oven would be recommended.

Using Easy Off in your oven is more hazardous. We used to bake trays of devices in giant ovens. There's nothing there at the temp you use in your oven to cause issues. That link claiming baking is hazardous is to a site that does solder work, using solder irons. RoHS compliance makes this a non issue using your home oven barring using it during the cleaning cycle.

RealNovgorod

HeavyHemi
This is silly. You're not heating the GPU enough to cause any outgassing of hazardous materials. You're in more danger eating a well done steak.

You're heating it to about 200°C, which is just below the melting temperature of solder, but more than enough to cause outgassing of all kind of nasty stuff from the PCB, such as flux, conformal coating resin (acrylic, PUR or whatever), adhesives and plastics. RoHS doesn't mean that breathing soldering (i.e. flux) fumes all day everyday won't kill you. Of course doing it once in your oven won't poison you (just like breathing soldering fumes once in a while), but do you want your pizza to taste like a PCB factory for the next 2 weeks? The real health hazard in that case would be your wife ...

Again, nope. RoHS has everything to do with the hazardous materials allowed in production of electronics. I did this for a living. I sure hope you're washing your hands every time you take apart and reassemble a GPU. Thousands of times more potential exposure to hazardous chemicals via absorbing through the skin. You're not putting the entire assembled GPU in the oven. You've never done this have you?

Lol, I just got it to work again - in the dumbest way possible!
I finally got around to give it the heat gun treatment because I didn't want to screw up my oven and also didn't want to risk reflowing the GPU for no reason when the problem was obviously just the RAM. I used a normal general-purpose heat gun, not a hot-air station (because I don't have one) and just eyeballed it. There were some tinned but unpopulated pads at the rear part of the board behind the MOSFETs, so I used them to estimate how long to blow air on the card until the solder melts - around 45 seconds at the medium setting. Then I went over all the RAM chips from about 2-3cm distance (one group at a time) for about 45-60s. The air flow was quite low, so there wasn't any danger to blow components off the board. No components were moving and it was hard to tell whether the solder really melted, so I had to rely on my timing estimate from the unpopulated solder pads.
 
Anyway, after cooling down and reassembly it worked right away with the correct BIOS! It shows up correctly in GPU-Z and Afterburner and the memory clock boosts to the correct 4.5GHz (non-OC) under load. It performs as usual in all applications, just like before it broke. So somehow I lucked my way into fixing it with the dumb eyeballed heat gun method - so there's a little bit of hope for the future Google visitor, but your mileage may vary. I also don't know how long the fix will last - I try to remember to update this post if it breaks again...