Basically, under load for a couple hours and the GPU itself will shutdown in some kind of lock-up protection, although I'm keeping it behind 65 Celsius.
 
This happened to me some time ago (almost an year ago) when I was starting to overclock it and pushing it to its limits, until I settled in stable "long term ish" overclock numbers for it. So I just pushed down clocks and the shutdown gone back then.
 
A month ago, after months playing a lot of Fallout4, it then shut down again. I reduced memory clock to avoid it to happen again. Took a while (weeks maybe?) and again. Reduced even more mem clock. All temps checked all the time, and I set limits to XOC to throttle card once it gets to 65C.
 
Power temp is keeping the lowest reading, never reaching 60C, while GPU keeps around it, and memory seldom touches it.
 
I am blowing cool air straight from outside the room via a pipe to the case, and it helps a lot keep temps down.
 
But now it is getting more and more frequent. Yesterday it even "stuck broken". POST was ok, but when windows was about to show welcome screen, a line of frozen dots was displayed in the main display; disassembled the backplate to see if anything was really wrong because it stopped booting.
 
In all these cases I could simply click the power button in the case and it gracefully shutdown. So I am 100% sure PSU is not being triggered. Voltages in XOC report right.
 
When this happens during gameplay, I can even hear the "device disconnect" sound; next boot some windows are all tiny to the top left corner of the screen, indicating it became like 640x480 VGA. I don't have onboard video, so I really get blind when that happens.
 
I just wanted to know where to look to troubleshoot this specific issue. Renewing thermal pads? Power components must be worn out and beyond help? There are some green lights in the GPU that I don't know. Kingpin has several LEDs and the exclusive probe-it jack (I would need to grab one or make the actual connector and know the pinouts to put that in the multimeter).

I have recently suffered from artifacting and reapplied thermal grease to the GPU, and it's been working perfect (artifacting-wise) since then. As the GPU-shutdown issue is happening each time more often, I suspect some component is wearing out and triggering a lock earlier. It may be getting point temperature spikes (that the GPU thermal LEDs or software don't report) and I might be able to address that just by reapplying the right thermal pad (which I didn't swap when I reapplied die grease).
 
But I'm afraid it's beyond saving and the pads' swap might be a waste of time. So wanted to hear from somebody experienced if possible which tree should I be barking at for a bit of relief.
 
- The card is out of warranty
- I never used the card in the OC or LN2 bioses, it's always in the conservative one (when I am overclocking via XOC, I set fans to maximum and monitor LEDs are set to become red as it hits 60C)
- My usual stable overclock settings are: power target at 120%; voltage 100%; temp target (priority 65C), GPU clock offset +37 (2037MHz), Memory offset +497 (6003MHz).

With more "aggressive" memory and GPU overclocks what I get is either game crash to desktop, or I may get color artifacts on screen (after it goes beyond 70C), or game simply crashes. I couldn't rule out what triggers the GPU shutdown yet (as it does require painful power cycle to test the next "tick").
 
I'm going to make do with the voltage slider for now and see if anything improves... but I don't think hardware intervention could be avoided at this point as the issue is getting worse and worse over time.

HeavyHemiThat's a pretty hefty memory clock of almost 1000mhz. Personally I've seen artifacts and crashing on my 1080 Ti with as low as +300 on the memory and I've completed benches at +800.

 
The kingpin has some serious active cooling sinks, although the memory is nothing out of the ordinary. So that's why I guess it gets stabe at +497. Yesterday I reduced core voltage (I didn't really need to keep it high as it didn't bring any o/c stability at all). Its debut uncorking guide gives some serious insight on what's on what:
 

HeavyHemiAlso XOC has been known to cause issues. Have you run the GPU without XOC running and let it run unmodified?

 
Can I set the o/c profile and close XOC? Or are you suggesting I should use another tool to do the O/C? Yes, I have extra unstability long term with Fallout (more than people complain overall), so this might help. I'm worried cause Precision X1 doesn't have the same readings (it doesn't support extra sensors the kingpin has, that are supported in XOC). Well, I can still monitor temps via its hardware LEDs, maybe setting green to 63C and red to 67C.

I am thinking to try a whole weekend test without overclocking (at, maybe 2012 it defaults to?). See what happens. The increasing frequency of GPU shutdowns though is still a serious concern. You have a good point about mem and maybe it is taking its toll for running so high for so long -- although I am not getting the same behavior I get when I push it even higher (beyond stability, only done once, while looking for stable values). Maybe I just ease on mem clocks and give it a weekend long try.
 

HeavyHemiyou neglected to mention age model of your PSU... one of the most common culprits of GPU issues.

 
Yes, this was on purpose, I don't want to touch the PSU until I rule out other stuff. It would be the one and only I would suspect if whole system shut down. So it is a secret. And being a secret means we'll going to have a lot of posts stating I can't get help because I don't tell which PSU I got... lol
 
Okay you win, it's a (invoice says 220-G5-0850-X instead of -X1 or -X2) purchased mid-January this year. Back when I was pushing limits and also had this same GPU-only shutdown was on another PSU, the (). Again, I am ruling out the PSU because it is not whole system reboot. No blue screen. GPU simply plugs out, pulling less than 450W from the outlet in worst case (I have an APC Back-UPS RS 1300 which has a Watts usage display so I can tell what the whole system is pulling -- that is, GPU + CPU + peripherals).

an 850 W should be fine.
Sounds like it is definitely getting worse, so the obvious 'fix' is to stop overclocking it .. it just can't handle it anymore after years of it.
Definitely get rid of POX , just use Afterburner.   You won't get readings, but it doesn't' cause software/ hardware conflicts either 
Good luck.