i7-920 can not get to 4GHz

Nameless_87 · 24 Apr 2011 at 18:19

Ok, something is happening but I do not know why...

Reduced VCore to 1.25V and it was prime95 stable for an hour.
Then I stopped it and started LinX, using all memory and run 10 times.
However during the first pass, after 3 mins, an error occured and it stopped, however everything still worked (no freezing or BSOD).

What is happening and what would the error be on the first pass?

Edit - Been reading about overclocking... is this true: if I getr a stable CPU overclock, lets say at 1.275V, then in order to increase the memory multiplier to 8.0 all I would need to tweak to get it stable would be the QPI? Since the CPU was already tested stable.

Thanks

Speed · 24 Apr 2011 at 18:30

It has likely detected an error in one of the calculations it has done and stopped, this is why I use both Prime95 and LinX, they both do things differently. Prime95 stresses the CPU at 100% for the whole test where as LinX seems to use more memory and picks up errors quicker.

Up the Vcore slightly and retest with LinX then Prime95.

Nameless_87 · 24 Apr 2011 at 19:01

ok I increased the Vcore to 1.275V and ran LinX again... It stopped on the 4th pass under same conditions as before...

What is the most likely error now? Is it Vcore again?

Speed · 24 Apr 2011 at 19:05

Could be either, but start with a few more notches of Vcore first as that's the lowest, test each time.

Nameless_87 · 24 Apr 2011 at 19:14

increased vcore from 1.275 to 1.2875V... will see what happens... If/when I get cpu stable I will TRY to increase memory multiplier to 8.0.

Also another question, everytime LinX fails, what errors could they be? Could it be for example that the QPI is a lot at 1.34V instead of low Vcore for example? As LinX does not give any indication what the error is :/

Edit - With that Vcore it stopped after 2 passes... now I am really lost!

Nameless_87 · 24 Apr 2011 at 19:38

I reduced Vcore to 1.28125V and reduced QPI to 1.3V

LinX 4th Pass - got the first BSOD since I started overclocking yesterday

lol

I am actually happy cause now I know what is happening... error code is alotofzeros124...

0x124 = increase/decrease vcore or QPI/VTT

since it was working good at 1.275V Vcore and 1.34V Qpi I am guessing it is the QPI... so going to put it back to 1.34 and re run test with the same vcore.

Edit - Failed after 2 passes as well... decreasing Vcore to 1.25675V...

Edit 2 - tried increasing/decreasing vcore, increasing/decreasing QPI but nothing seems to work... with LinX trying to run 20 times, everytime it is failing on either one or two passes

dunno what to do anymore. Thing is... prime 95 is stable with the things I tried

twist3d0n3 · 24 Apr 2011 at 23:08

try bumping up the ram voltage a notch or 2. don't worry about the 'exceeding 1.66v = kill cpu' stuff, it's only applicable if you have a 0.5v difference between ram voltage and qpi.

you may just have a crap cpu needing that many volts.

Speed · 24 Apr 2011 at 23:09

Good idea, I'd also try increasing Vcore and QPI above 1.35v. Lowering the voltage is rarely the solution.

Nameless_87 · 24 Apr 2011 at 23:27

Since prime 95 is stable I am going to leave it running overnight with 1.675V Vcore and 1.34V QPI and memory multiplier to 6.0

If I can get it stable like this I will leave it with this ram multiplier as I don't want to overvolt the CPU and damage the components

bifday2k · 25 Apr 2011 at 00:16

Whoa, that vcore is seriously way to high, i would not leave it over night at the Vcore.

Unless thats a typo which i hope it is.

Speed · 25 Apr 2011 at 00:43

Yeah I think he means 1.275 lol.

You won't damage the components using that sort of voltage, anything sub 1.4 is fine for 24/7.

bifday2k · 25 Apr 2011 at 00:49

Really hope so lol, even I'm not mad enough to go to 1.6v and im on water lol.

But yeh sub 1.4 is totally fine on both volts, can go over slightly if your cooling is upto it.

Nameless_87 · 25 Apr 2011 at 07:04

Oh yeah sorry my bad... yes it is a typo... im not mad lol

I left it overnight at 1.2675V or 1.2685V I don't remember the number exactly...

Been stable for 8.5 hours so far and still going on prime95 blend test with temps around 78 degrees.

After 12 hours I will stop the blend test and start large ftt test... If that passes as well I will reduce the Vcore further.

About the memory 8 multiplier what are safe values over 1.35V for QPI and Vcore? Also what about uncore, QPI link, QPI PLL, CPU PLL, IOH, ICH? They are all on auto... can I tweak them to 'spare' some volts from QPI and Vcore with memory multiplier 8?

Speed · 25 Apr 2011 at 14:08

Just because it is Prime95 stable doesn't mean the system is stable if it fails LinX so quickly. Also make sure to run LinX as admin.

1.2675v is very low but anything under 1.4v (even a bit over) is perfectly safe. The other settings can be tweaked, but it is best to get it stable with just Vcore and QPI voltage increases and go from there. However tweaking down PLL on gigabyte boards can aid in stability.

Memory issues are almost always QPI related, so you may need higher QPI voltage to get it to work at the 8.0 divider, I certainally did. It varies on how good the onboard memory controller is.

Nameless_87 · 25 Apr 2011 at 15:13

so I should increase QPI to something like 1.38V instead of 1.34 i am using and increase Vcore to 1.30V for example and go from there?

to what level is QPI safe?

Also I will try with 8.0 memory multiplier immediately.

Btw, it has been stable running prime95 till now so about 17 hours..

setter · 25 Apr 2011 at 15:16

17 hrs on blend is good going, id try with LinX on your current settings.

Nameless_87 · 25 Apr 2011 at 15:19

i did and failed in the second pass using all memory and 20 counts :/

That's what I don't understand!

Speed · 25 Apr 2011 at 16:46

Up the QPI yes, anything up to and slightly over 1.4 is safe. See if that solves your LinX problem.

Nameless_87 · 25 Apr 2011 at 16:57

Will try to up it now and report back...

prime95 was stable for 18.5 hours till I stopped it (now)

Temps reached a max of 82 which was 17 hours ago... but have been stable at 77-78 since then.

Will report back on how it goes. I will also up the Vcore to 1.3 Volts and memory multiplier to 8, uncore to 17x and QPI Link to 36x. Everything else will be left on Auto.

Speed · 25 Apr 2011 at 16:58

Frankly changing uncore and QPI link may cause more issues, leave it all on auto for now, you can use that for trouble shooting later. The logic bug that required uncore to be increased to x17 has long since been sorted. It should be twice the speed of the DDR3, which is what Auto does.