Why do odd multipliers work better than even?

[Rubycon]

When overclocking a socket 1366 it seems that stability is always greater on odd multipliers. Is there a reason for this?

 

[aigomorla]

I realized this also back in the 920 days...

but no one seems to listen to me. :T

 

[soccerballtux]

I realized this also back in the 920 days...

but no one seems to listen to me. :T

what?
(just kidding)

 

[ghost recon88]

It's just the way things are Just like how the 980xe and other Extreme Editions clock worse on air than their non-EE counterparts. Some thing with Intel just can't be explained by man. Well actually the latter can be explained due to leakage, but the odd multiplier still isn't explained.

 

[Rubycon]

I realized this also back in the 920 days...

but no one seems to listen to me. :T

Hmmm we're still in the 920 days!

Looks at 980X plugged into same socket fitting 920s.

Yep we're living in the past with thoughts thick as a brick while we bungle in the jungle!

Sorry cannot help it when I hear Ian's flute!

 

[aigomorla]

ummm my 920 days means the rest of you guys were still stuck on lga775.

:X

When you guys got 920's i was on a 975.
And when people got on a 975, i went gulftown.

Now people getting gulftown... so i guess i need to find a beckton from my sponsors back pocket with a board to go with it.

:awe:

 

[Rubycon]

ummm my 920 days means the rest of you guys were still stuck on lga775.

:X

When you guys got 920's i was on a 975.
Like how when people got on a 975, i went gulftown.

Yes but you are starving your chips! You know what I mean.

Lean is mean! Ask any racer that uses NOS!

 

[krose]

Hmmm we're still in the 920 days!

Looks at 980X plugged into same socket fitting 920s.

Yep we're living in the past with thoughts thick as a brick while we bungle in the jungle!

Sorry cannot help it when I hear Ian's flute!

Jethro Tull reference FTW. I can only use 19x or 21x , and with 8x memory I can only use 17x uncore. 16x doesn't boot.

 

[Anubis]

mine runs at 20x just fine /shrug

 

[RussianSensation]

I am pretty sure the answer is engineering based in nature -- in other words requires actual knowledge of how multipliers interact with the particular architecture in place. I actually thought that Core 2 Duo/Quad architecture overclocked better with odd multis as well - is this true too?

Any Intel engineers here?

 

[n7]

There has to be an actual reason, though i cannot understand why.

Also, i've not found this to make any difference for my i5 750 & i7 860.

Haven't tested this really w/ my i7 930 as i'm still trying to figure it out.

 

[iCyborg]

I've heard about this several times, but can't say it's been my experience. Sure I can jack up bclk a bit more on 19x (I have 920), but not the overall cpu clock. I didn't play with OC all that much though...

 

[ehume]

I'm running 182x22 on an i7 860 ATM, with Vcore set to 1.26875v. Never any problems with the even multi's.

 

[Rubycon]

I'm running 182x22 on an i7 860 ATM, with Vcore set to 1.26875v. Never any problems with the even multi's.

If you push things to the limit you will see that you reach a limit (at the same CPU speed) faster when on an even multiplier. You need an unlocked chip to see the correlation OR just be "on the fence" with a locked one.

 

[WhoBeDaPlaya]

I can't say that I've noticed any issues running either 210x20 or 200x21 on 3x i7 920 D0s (all same batch/step code/pack date on EX58-UD3Rs).

 

[Spikesoldier]

I can't say that I've noticed any issues running either 210x20 or 200x21 on 3x i7 920 D0s (all same batch/step code/pack date on EX58-UD3Rs).

what he meant was that it takes a higher amount of voltage to be stable at the same frequency on an even multiplier than an odd one.

ie: 4000mhz 20x200 1.30v vs 4000mhz 211x19 1.25v

anand mentions that on the new 32nm chips that this voltage disparity to be stable is less pronounced compared to the 45nm chips.

 

[n7]

This post is a ridiculously good source of information on X58 OCing, particularly regarding why higher memory speeds are difficult to obtain, & why QPI/uncore/RAM speeds need to be set up perfectly...
http://www.xtremesystems.org/forums/showpost.php?p=3495431&postcount=877
He also touches on the even vs. uneven multi thing.

I was having issues w/ my OCing on my new X58 till i discovered this...now i understand a lot better!

 

[WhoBeDaPlaya]

what he meant was that it takes a higher amount of voltage to be stable at the same frequency on an even multiplier than an odd one.

Yup, I know what he meant. I ran the exact same voltage for both configs.

 

[Rubycon]

I am pretty sure the answer is engineering based in nature -- in other words requires actual knowledge of how multipliers interact with the particular architecture in place. I actually thought that Core 2 Duo/Quad architecture overclocked better with odd multis as well - is this true too?

Any Intel engineers here?

Yes that's what I'm looking for. Specifically if equipment is attached to the motherboard what signals should one be looking at?

 

[WhoBeDaPlaya]

Yes that's what I'm looking for. Specifically if equipment is attached to the motherboard what signals should one be looking at?

I wonder whether it's got anything to do with the loop response from the relevant phase locked loops, since changing the multiplier does change the loop gain.

 

[Rubycon]

I've highlighted the signal values in the Classified 4X 762 BIOS that nobody seems to discuss.

Anyone - yes anyone - "play" with these?