Overclock stable, but not when gaming...

Roverno · 4 Jun 2012 at 19:29

Hi all,

I have overclocking my 2500k in sig below and was running 5GHz on 1.425V, that was fine for 20 mins but not overnight so i clocked back to 4.9 and reduced voltage to around 1.408v. That was stable for an overnight run of prime 95 blend test but when i was playing diablo i got blue screen with error 0x124. As far as i know that means i need increased Vcore, why on earth would it be able to run prime for hours but not be capable in gaming??

What is going on here?

Cythx · 4 Jun 2012 at 19:31

No idea.

Decrease the clock and/or increase the voltage.

Do you really need 5 Ghz for Diablo...? You only have one GPU and I assume one screen, I'm almost sure 5 ghz doesn't benefit you any more than 4.8 ghz.

Roverno · 4 Jun 2012 at 19:34

The overclock was not for diablo, it just so happens that it crashed while i was playing it. I just wanted to clock my CPU as high as i could really now that i have watercooling, and for some reason 5Ghz just took my fancy.

EDIT: I understand 5Ghz is not necessarily achievable with every chip, but why would it be fine on prime but not on gaming?

Durzel · 4 Jun 2012 at 20:05

Extra load from graphics card causing voltage dip? I've heard some people talk about running Furmark and Prime in tandem, brutal!

Justintime · 4 Jun 2012 at 20:16

What version of prime95? Anything under around 27.4 can test falsely positive for SB and onwards as it dosen't use AVX extensions.

Leeroy123 · 4 Jun 2012 at 20:19

Overclock isnt stable then is it lol...... more voltage.

Justintime · 4 Jun 2012 at 20:22

Also forgot to add, i've noticed prime failing at 13-16hrs on a regular basis with the newer versions, usually 101 or 124 BSODs, so an overnight or 8hr run just dosen't cut it imo.. many will moan but overclocking takes patience. 24hrs seems to be a good indicator and haven't had a rig fail at anything once that is passed.

Leeroy123 · 4 Jun 2012 at 20:26

101 and 124 are both voltage errors.

Justintime · 4 Jun 2012 at 20:32

Leeroy123 said:
101 and 124 are both voltage errors.

Er.. thanks. To be specific, it can be related to Vcore, Vtt, Vdimm and PLL up or down, 90% vcore but not always and adding more vcore may not be necessary, you can tweak the LLC to compensate for vdroop.

Roverno · 4 Jun 2012 at 20:50

Durzel said:
Extra load from graphics card causing voltage dip? I've heard some people talk about running Furmark and Prime in tandem, brutal!

I did run both for a bit but it only used 16% Gpu so i thought it would be fairly pointless, perhaps not, will try again.

Justintime said:
What version of prime95? Anything under around 27.4 can test falsely positive for SB and onwards as it dosen't use AVX extensions.

Turns out im running v25.11. That should probably be my first port of call. I have heard OCCT is quite good, or AIDA64. Would you recommend p95 though on new version?

Justintime said:
Also forgot to add, i've noticed prime failing at 13-16hrs on a regular basis with the newer versions, usually 101 or 124 BSODs, so an overnight or 8hr run just dosen't cut it imo.. many will moan but overclocking takes patience. 24hrs seems to be a good indicator and haven't had a rig fail at anything once that is passed.

I personally like taking time over it, but mates been heckling me for a week to play diablo lol. Just tried some stuff to see what would happen.

Roverno · 4 Jun 2012 at 20:53

Justintime said:
Er.. thanks. To be specific, it can be related to Vcore, Vtt, Vdimm and PLL up or down, 90% vcore but not always and adding more vcore may not be necessary, you can tweak the LLC to compensate for vdroop.

I did know most of this but i have just thought about my RAM. I have hyper x grey which i have had running at 1600 on 1.5v for a while but in this new board maybe it wont work quite so well? I should perhaps push that voltage up a bit as that would explain the crash in game?

Justintime · 5 Jun 2012 at 12:03

Yes i would say a 24hr Prime run and about 50-100 passes IBT/Linx with AVX (make sure Windows 7 SP1 is installed) is good practice.

joeyjojo · 5 Jun 2012 at 13:02

Forget prime, I've been saying for years it's not a good stress test.

Get the newest IBT, set it to max memory without paging, and do 5 or 10 loops (more is unneccesary).

Muel · 5 Jun 2012 at 13:07

If it's unstable, it's unstable lol. Overclocking takes a long time of trying different stuff and getting to know your own CPU. I know I have a very bad 965BE, it takes 1.55v to do 3.8ghz, but isn't 100% stable so I run 3.7ghz at 1.475v, which is stable.

Takes time man.

As a proper test though, Furmark and blend at the same time. Gets the ambient temperate inside the case as high as possible, drawing maximum power from the PSU, etc.

Justintime · 5 Jun 2012 at 13:53

joeyjojo said:
Forget prime, I've been saying for years it's not a good stress test.

Get the newest IBT, set it to max memory without paging, and do 5 or 10 loops (more is unneccesary).

This is utter nonsense. Sorry but theres opinions but yours is completely wrong.

10 loops of IBT is NOT enough, i've done extensive testing with it and have noted that it takes about 100 loops to suss out stability in a system, i've had numerous fails in the 50s, 70s etc.. but none thus far at 100-150 (i do 150).. maybe that number allows enough time to take environmental changes into consideration such as voltage quality etc..but it just works, prime blend is a good test of your IMC/Northbridge/whatever your system uses and RAM, for Sandybridge and up - the latest versions from about 27.4, he was not using the proper version to test its as simple as that.

I'm not belittleing you but i've got way more experience with virtually any system and am now at my 183rd Sandy bridge and 14th Ivy Bridge overclock and have always used:
Prime blend
IBT/LINX
Furmark
OCCT
on every rig and find all are useful, however i can without doubt tell you that a system that has passed 24hr blend with the right instructions and 150 LINX with AVX loops is cpu stable 99.99% of the time.
You may do overclocking as a hobby but i've got a vested financial interest to make sure my rigs are rock solid, and over the years i've rarely had an issue that wasn't a hardware failure or driver issue, not had any overclock related ones.

joeyjojo · 5 Jun 2012 at 14:27

Thanks Justintime, your experience and contributions to this forum are significant. However, you say yourself that you "have always used" that same test selection, which may give a good indicator of strability, but it gives us no information about which of the tests are most rigorous. I would guess that some of them are more intense than others, and that you could dispense with the less good ones, bu neither of us are qualified to say which.

To dispute my claim that "prime is useless and IBT is teh roxor!11" you'd need to find a system which is stable in 10 loops (say) of maximum memory (without paging) IBT, and unstable at 24 hours of prime. I doubt it exists. Also you probably can't be bothered

Edit: just reread - you've seen it fail after over 50 loops? I'm surprised at this. Got a screeny? If using sufficient memory, the program is designed to find instability far earlier than that.

Justintime · 5 Jun 2012 at 14:45

I've seen it fail as high as the 80s, however when it fails that high its usually just a halt program error as the outputs didn't match, as opposed to an early fail (10-20 etc..) that 9/10 times i find results in an all out BSOD. Sorry if i came off too harsh as i tend to get excited and overzealous about these things lol.

joeyjojo · 5 Jun 2012 at 15:02

That's fine mate thanks for the info

It seems to me (and I could defs be wrong) that the program which gets the chip hottest should be the one and only one you need to use. Why? A couple of reasons. First, if the chip is hot it's using more of it's transistors, which means if one is going to make a mistake it'll happen sooner than if the chip is not as fully loaded. Second, if it's hot it's drawing more current, meaning the regulators in the mobo and in the PSU are working harder, which means they're more likely to slip up sooner.

Justintime · 5 Jun 2012 at 15:15

Loading and heat is not the only reason for instability. What ibt and prime calculate and use (imc, memory etc.) can vary a lot from each other. You'll probably be surprised to know that prime95 with AVX is only a degree or 2 off LINX/IBT loads.
While IBT is good for finding the max heat output (which is artifical and will never really happen) and a good tester of stability, its rather brutal, prime is more sophisticated in that it tests the cpu/memory imc etc.. with different patterns whereas IBT AFAIK only repeats the same test expecting the same result over and over, hence prime may pick up on things IBT won't.

joeyjojo · 5 Jun 2012 at 15:38

I am surprised to know that. Presumably for small FFTs? Is blend cooler?

If you wanted to just test the CPU though (memory is at stock say) you'd want to stress only the CPU and it's power supply (PSU and mobo), as to test the memory would be pointless. So you'd choose the hottest prog, which is stressing the bits you're interested in.

Anyway, pointless debate, best let the OP get back to his stability testing