Does this happen with the card at stock clocks?

Found this… https://forums.evga.com/FindPost/3035532

No problem.

It could be tripping because the voltage needs a slight bump, but it seems like the power rail for the mem has an issue since the mem light goes out. If this issue occurs at stock, oc’d & oc’d and ov’d then I’d say the mem power rail has a problem.

Another update... Tested lower clocks until stock clocks. It still breaks after a couple days. No matter how I set clocks up, it only hangs the following day I booted up, in the least.
 
Remember "stock" clocks for KPE are 2012MHz core, 5500MHz (1375MHz) mem, that's what XOC reports when I apply its "default" settings. I didn't go underclock to see how low it goes.
 
If I don't game, it does not break at all, no matter how many days I keep the machine on. So there's probably an underclock setting I could work with to still play on that card.
 
Notwithstanding, I finally found my Zezzio thermal pads, ripped the whole GPU apart, cleaned everything. Painfully, yet patiently removing up old grease and padlets stuck here and there, to near-mint-condition.
 
Assembled everything and so far temps reads right across the board. If this doesn't fix it, I think I'll give up. I applied some thermal pads in additional areas with transistors that could contact heat spreaders.
 
I really hope the issue was a hot spot -- and that I didn't create new ones in the process.

And the next update is... still the same! :(
 
My conclusion so far is, the issue is due to components wear. This memory should run fine at 6,000 (1,500) when new. So it's probably just "spent out".
 
Well, at least re-applying the thermal pads and grease didn't worsen anything... :) I believe.
 
Now moving to my next resort: underclocking. Running now at around 1900MHz core and 5400MHz (1350MHz) memory and will run for a week or so of gameplay. If it gets stable I might try to improve one clock or another to see which one (if not both) is making the GPU go bonks. It should still perform kinda well as the core runs at about 500MHz more than reference card. But it's a pity I was counting on the "excess" this card provided for longer life. Anyways, if it still can perform as well as reference card, it's (much) better than returning to a GTX 670 or having to spend the bucks for a newer card.
 
For what I could see, without underclocking, the card goes on par with RTX 3070 (ray tracing aside). Which, interestingly enough, around my parts, goes around the price I paid for this GPU when I bought it -- a few weeks before 20 series was released.

A quick update, the underclocking did the trick. The card run smoothly for days.
 
I begun with:
- 80% power (priority)
- 70C temp
- -100MHz GPU clock offset (2012 - 100 = ~1912MHz, still above stock)
- -100MHz memory clock offset (1250MHz -- 5000MHz)

When I raised back to 100% power target, the problem happened again.
Then I begun fiddling with its limits (hope I'm not breaking it even more, but well), then worked stable at:
- 90% power (prio)
- 70C temp (I stick to this to avoid making air gap in the GPU like it happened when it was stock assembled)
- +37MHz GPU clock
- -100MHz mem clock
 
And now I'm going to change that to 95% power and run to see if it crashes.
If not, I will then return the memory to 0 offset and try...


Something interesting is, when I was at the 80% power setup above I had my system hard-shutdown, as if the PSU was broken. Now I suspect I may have an issue with either my Gigabyte motherboard, or the PSU itself.
 
The PSU is a supernova 850G5. I think it is pretty "high-end-ish" and has a long long warranty.
 
If in the end the PSU is borked, it could mean the issue is not with the GPU but PSU. If the culprit was the mobo, then the GPU issue persists. I have an old 500W (OCz StealthStream II) PSU that I can try plugging dedicated to the KPE 1080Ti. I will do the test when I get a chance. And if that solves the issue, I guess I will need to RMA my PSU.
 
I don't pronto assume the PSU is the culprit because the mobo is also showing signs of aging, with one fan header busted (I guess it was in the last time I cleaned up the rig guts). So, yeah, it really feels like time to upgrade the whole ~2016 stuff here... $$$. :P

I guess this should be the last update before the card dies. :P I don't think there's anybody else in the world with the same card and issue to exchange experiences about anyway. Hopefully one day this thread proves helpful to somebody venturing with these old (and legendary) cards. I will miss so much the power connectors location!!! No other brand seems to do what Tim and Vince envisioned. :S

At 90% power and stable. 95% and it dies just the same. I still didn't test with 2nd PSU (have to dig one out from boxes...). I moved the GPU and mem dials to +37MHz and +0MHz resp, and it didn't change anything.

Interesting that in the usual lit white LEDs next to card's power connector (12V, 3V3, PLL, MEM, GPU) the MEM LED is lit really bright (figures?) while 12V is a tidbit brighter than the rest, and then 3V3, PLL and GPU LEDs are least bright (but not really "dim"). D3909 LED only lits up when the the power "trips" and shuts its MEM rail down (thus the MEM LED becomes turned off).
 
I never paid attention to the LEDs before, but now what I see when I power up the machine is that MEM LED starts off and "smoothly" lits up, not instantly like the other, weaker, ones.
 
I believe it is a matter of time before 90% becomes unusable, but I hope it doesn't just die on me. I guess the advanced thingies in the card both pulled down its life span -and- allowed me to monitor and trim down pots (via software) I wouldn't be able to in other cards.
 
Besides, even downpowering it, I still read clocks above reference for this card when it throttles, so it probably should still be running faster than average. In-game FPS, by the way, is now noticeably low, it still kicks a great punch, and the throttle downs are really smooth. That is, compared to other FPS hitting aspects of the game, like more complex areas like forest and city, where the game I'm currently playing, Kingdom Come: Deliverance, becomes actually CPU-bound -- so I get lower FPS -and- GPU demand in those areas.
 
p.s.: this lighter MEM light could have been because I switched the DIP for the vDroop/memory in the board, I didn't test pulling it back in, as it didn't really change anything stability wise. I explained what I found about these vDroop switches in my previous posts.