-Bigadv sensitive to high clocks?

[Eno]

Greetings folks.

My name is Eno, and I am a fold@holic.

I've been working hard towards getting the most out of my 980X processor and have achieved a stable 4.4 according to hours of folding, Prime, Linx, intel burn test and memtest. However, as soon as I start to breach the 4.1... mark suddenly things with the A3 core start to get VERY fragile. I folded for about 15 hours yesterday only to get to 60 some percent and have the A3 core crash on me. I tried numerous times to sort it out today, only to have it continue crashing. Linx shows stable, intel burn test memtest... everything is showing that the comp is stable.

The 980X is of course unlocked multiplier- so I am currently only using multiplier (33x133) for the clock. I have sufficient voltages, everything is balanced out very well.

My request is for information pertaining to where the instability is being caused for F@H. Client /core communication (C00000005) seems to be the prevailing error code. Is this in fact instability within the core itself? Is there a particular setting or feature of my BIOS that I'm overlooking?

I do thank you for your attention. I am excited to push my rig to the very limits of its ability to complete as much as possible, as so many others are as well.

Happy new year everyone... and here's to a great folding season ahead!


eno

[uncle fuzzy]

Welcome to the forum, Eno.

I see you didn't mention StressCPU2 in your list. It's a Gromacs-based cpu tester that comes close to the strain of folding.
viewtopic.php?f=14&t=52 (near the bottom of the list)

I can't fold on my 980X above 3.9 (133x29/30), although others have managed to be stable over 4.

[Eno]

I was doing quite well with 4.1... like I said- but ANYTHING over that seems to crash. I'm going to try stresscpu2 as well and see what happens! Thanks for the lead.

[uncle fuzzy]

It's very close to folding. If it fails StressCPU2, it will probably fail folding.

[Eno]

I'm running it and apparently the version I downloaded just goes indefinitely or until it fails. And will it give me an idea as to why it failed so I can correct the problem, or is it just going to crash out like a normal WU? What's a solid run at it? An hour? 12 hours? Infinity?

The thing with the 980 is it's really taken the challenge out of overclocking. You could technically just change a multiplier and hit a plus button two or three times at vcore voltage and walk away... So though I was perfectly happy with the ppd at 4.1... now it's stuck on me that I MUST figure out why it won't be stable at 4.4. I'm HOPING someone is just going to chirp up and say "the cores weren't tested that high because at the time there was nobody who could beta test them at that speed." I would GLADLY pull the throttles back and "ease" back into 4.1.

I would normally look to clock skew settings on the BIOS- if I was using any bclock... to help information match up a little more reliably- I'm still considering it as I review what the skew actually does.

Any thoughts?

[uncle fuzzy]

Yes, it runs 'til it crashes- BSOD.
No, it doesn't tell you what happened, just that you're not stable for folding.

[Eno]

Sorry, I'm new to this kind of stress test...

I have set my comp to not restart and wipe out my BSOD... at least then I can interpret an error code-

But the crashes I've been getting have just been windows based "A3 Core has stopped functioning" kind of errors... I'm not BSOD'ing (anymore.)

Any thoughts on what could be causing the instability?

[PantherX]

CoreStatus = C0000005 (-1073741819) can usually be caused by RAM (Details). Since you have overclocked your system, maybe you need to tweak the RAM settings to get it stable? I would suggest that you read section 1 - Problematic Hardware in this thread -> viewtopic.php?f=19&t=16526

[Eno]

I WILL do it! Thanks PantherX!

[Grandpa_01]

I am running 2 980X CPU's and yes they will fold at 4.4Ghz stable and yes there is a magic button both of mine will fold at 4.5Ghz but it takes allot of Voltage to do it with mine it takes 1.5+V to do it and that creates allot of heat even with liquid cooling it is higher than I am comfortable running (90+C). I can run at 4.4Ghz with 1.39V to 1.4V and (80+C) but I have found that my best settings are 144 X 30 for 4.318Ghz @ 1.325V and running my ram at 1728Mhz 7-7-7-21-1T-72. I run allot cooler there 70C and my frame times are only about 40 sec. slower about a 3000 PPD loss. So your magic button is your voltage setting. I do not know how far the 980X's will go but I have booted and ran Intel burn test at 4.6Ghz it just takes good cooling and allot of voltage. Do not forget that Intel recommends 1.375V max for the 980.

[Eno]

Eh Gramps...

Yes, they recommend a lot of things, don't they?

On the recommendation of PantherX I've also taken a close look at my RAM / QPI settings. Interestingly enough, I realized that with my TOTAL absence of touching anything other than multiplier I should have probably foregone my adjustments in those respective voltage areas. Who am I to tell the BIOS what it needs... when in fact it doesn't even know I've overclocked anything other than the CPU directly via multiplier.

So my course of action was to go back to auto on certain settings- they might be slightly over-volted... but they won't be under-volted! Don't get me wrong- I'm still sitting at 1.418 or so on vcore so things are still somewhat on the toasty side but still seem to be sitting sub 60 with my twin ultra Kaze 2k flanking an H70. Say what you want about the H70- with decent fans the thing kicks butt.

Prior to my switch a 2684 crashed on me in about 15 minutes- actually, I should say the A3 errored out on me and caused a windows error.

I'm running quite a bit more voltage than your INSPIRATIONALLY low vcore settings- My RAM is 1600 stuff and I haven't touched it in terms of timings nor frequency. (I obviously had enough problems just with cpu!)

Thanks for giving me some feedback though... Tough to just automatically equate errors with vcore though now I know I may have to bump it up even higher!


Happy new year!

[7im]

Obviously some applications are less stable at higher clock speeds. Windows probably doesn't care if you flip a few bits the wrong way while browsing the internet. Hence Prime testing and Windows will run "stable" at higher speeds.

However, data intensive applications, with very good error checking, will find those flipped or errant bits at slightly lower clock speeds.

A system is not stable until it runs all the applications that you want to run, and it doesn't affect how those applications perform. FAH may be the lowest common denominator. Doesn't mean that it is sensitive to overclocks, just better at demonstrating lower level instabilities, or instabilities in parts of the system that those other test programs don't test well.

[Eno]

Thanks Tim... all good points. I've done lots of reading on the page and come to the conclusion that the extra 200mhz just isn't worth it. I've crashed on so many WUs and had them abandon and restart so many times I'm POSITIVE it has- as it says it would in the forums- cost me more than I'd EVER hope to gain from it. I'm parked at 4.123 again and just going to stick to it.

Thanks again to all of you for your help and your commitment to the folding community.

And of course, happy new year!

Tim, see you over in mpc forums!

[k1wi]

Yeah, folding-wise rock solid is absolutely more important than cutting a couple of seconds off the tpf.

[Eno]

I agree overall- but if you buy a race horse you don't exactly put it on a treadmill either... so pushing the chip to its very limits was / is a challenge. What is daunting is the prospect of actually being penalized because the WUs are returned unfinished... It's tough to test the OC when they only thing that makes it fail is actual WUs.