Weird Temperature Spikes after OC 6800k

Dobri · February 23, 2017

OK, so this is a weird and kinda lengthy one so please stay with me. Situation: Bought baller PC a few months ago, 2x1080, 6800k, all the good stuff (not interested in opinions). Overclocked 1080 to 2050 stable - success. Tried Asus software overclock for the 6800k, bricked my windows, cursed at them, figured it's crap and decided I'll do it manually later. Fast forward several months of heavy use with no issues. I decide one day "Oh hey, let's go back to that overclock". Overclock it to ~4.3GHz with a AIO water cooler. Success. Temps ~80 at full load on all 6 cores, messed around with individual core clocks to tighten the temp range, all is well. Stress test ~30 minutes, perfectly stable. Alright, let's get to gaming. Game for a few hours, computer freezes. Force reboot, Asus says "Overclocking failure". Well that's weird, let's dial it back a bit. Doesn't matter, won't turn on, "overclocking failure" gone, post code 61 (NVRAM Failure). Freaking out, shut down computer.

Turn it back on in 5 minutes, boots up just fine to Windows. OK, that was weird, let's stress test again. This time, I see these huge (15-20C) temperature spikes in the middle of stress testing (100% load throughout, no clock change but temps spike). Poking around at Hardware Monitor, I see matching spikes in power. Voltage - stable. Power - spikes from 90-ish reading to 120-ish reading for no reason (probably not real watts but that's what HW monitor was saying). Oh, maybe I have a crappy PSU. I was going to change it anyway so I bought a new one, well capable of supporting the 2x1080 and the 6800k + whatever other residual load. Same weird power + temp spikes. Play around with digi+ vrm settings to turn off all auto-power management and basically set everything to OC settings (disabled eco stuff, enabled power stuff). Hardware Monitor no longer reads Power but temps continue to spike. All throughout this experimentation rig is stable during 100% load in stress test but crashes in gaming load (???) with that same "overclocking failure" + post 61 + wait 5 minutes to use your PC. I've dialed it all the way back to stock clock and it still crashes while gaming, despite stable results during stress test (outside temp spikes). Oh and water temp never gets above the mid-40s (reported by Corsair Link) even after ~30 minutes of stress test.

Any ideas? I've attached the best graph I have of this in action, clock vs tmp of both the 6800k and 2x1080 while running prime95 and furmark. I saw them being stable at the start so I went ahead and did something else for a while or I would've stopped the test at the 100C (!!!!) peak for the CPU.

Dragonix · February 23, 2017

I do not have a ready solution for you, however 30 min. of stress testing is way too soon to call the machine "stable". I would suggest testing it at least 12 or 24 hours under stress.

When stress testing with Prime95 keep in mind the load will not be constant. The data is changing. I remember my i7 was reaching peak temperature around 15 minutes into the test. It was a sudden 10% spike in core temperature.

Once I owned a GTX 670. It was stable for 5+ hours of UNIGINE Heaven Benchmark on maximum settings but would crash in Crysis 2 within 15 minutes.

done12many2 · February 23, 2017

1 hour ago, Dobri said:

(not interested in opinions)

Sure you are or you wouldn't have said the following.

1 hour ago, Dobri said:

Bought baller PC a few months ago

If you are stress testing with Prime95 in blend mode, you are going to see temps ramp up and down as the test cycles between FFT sizes.

Stay away from the auto overclock crap and manually dial that badass rig in yourself.

Dobri · February 23, 2017

59 minutes ago, Dragonix said:

I do not have a ready solution for you, however 30 min. of stress testing is way too soon to call the machine "stable". I would suggest testing it at least 12 or 24 hours under stress.

When stress testing with Prime95 keep in mind the load will not be constant. The data is changing. I remember my i7 was reaching peak temperature around 15 minutes into the test. It was a sudden 10% spike in core temperature.

Once I owned a GTX 670. It was stable for 5+ hours of UNIGINE Heaven Benchmark on maximum settings but would crash in Crysis 2 within 15 minutes.

Hm, yeah, that's a good point. I guess I was mistaking processor use with the nature of the operations, I can see some prime95 cycles being more straneous on the CPU, even though the net "used cycles" does not change. That doesn't explain why I haven't seen it before but I'll just assume I didn't have a benchmark running long enough, considering I haven't tried OC the 6800 before.

That still leaves the open question - CPU was stable at stock clock for months. Why is it bringing it down to stock clock now makes the PC crash while gaming? And in general, what can I do to make the CPU stable, at any clock speed but preferrably ~4.1 which seems to peak in the high 80 during one of those spikes but runs at low 70 at full load otherwise? I know 100C is high but I don't think it's high enough to cause permanent damage to the CPU, especially for the 2 minutes it thermal throttled for and considering specs say 105 to be the T max, which is supposedly the maximum temperature before lifetime is significantly affected.

2 minutes ago, done12many2 said:

Stay away from the auto overclock crap and manually dial that badass rig in yourself.

That's exactly what I did that I perceive caused all of this. Auto overclock bricked Windows (couldn't even boot safe mode or run any repairs, literally had to reinstall). I promised myself I would not use it again until everyone starts badgering me how much better it is or manual OC is turned down completely, whichever comes first.

And I meant I wasn't looking for the usual "You don't need two 1080s and you definitely don't need a 6800 for gaming" opinion some people have a hard time keeping to themselves. Compliments on its ballerness are completely welcome, I'd do full specs but you already know the good stuff.