LGA1156 Core i7 & i5 Overclocking Guide

[andy5174]

4C less with Prime95 as expected.

That's a good idea!

75C is just insane under 17C with Hyper 212 Plus.

Update: Yes, I have tried IBT as well and the core temps is just 4~5C higher. BTW, I just passed 20 runs IBT test 2 days ago.

 

[ehume]

I'm going at this from the opposite direction. I've had my Hotrod up to 4.5GHz (without the torture test), HT enabled. I've had my BCLK up to 230MHz. I've run stable LinX runs above 4.1GHz, but even with the MegaH it was touch and go because of the heat. All of this was at Auto voltage, but the MB had that up to 1.396v.

So now I am trying to find a way to run at 4.0GHz as cool as possible, with HT enabled. So far I've been able to do that with voltage down to 1.25v. But I wasn't paying too close attention to the various settings, changed some to "safer" settings and messed up the ability to run at low voltage.

The settings that seemed to work: BCLK at 182, multiplier at 22x, power-saving and shutoffs all enabled (EIST, C-states, etc.). In retrospect, seeting the multiplier to 22x automatically set the Turbo Boost to Auto. Setting it to Disabled lost me the ability to push the multiplier to 22x.

Disabling the LLC lost me the ability to OC at low voltages. In retrospect, who cares if enabling LLC causes spikes when the target voltage is so low? Next time, I'm leaving it at Auto, which is where the MB put it when I started OC'ing the P55A-UD3P.

As best I can tell, 182x22MHz gives core temps 1-2c less than 200x20MHz. The temps are OK in an 18c basement, but when I'm done testing, the machine comes upstairs where the ambient temps will be higher. For HT-enabled, 4GHz seems to be at the edge of what the Megahalems can handle.

 

[bob5568]

I'm impressed and surprised by your success! We have similar equipment. You've gotten a much higher stable overclock than I. Perhaps you got lucky and got a great cpu. Maybe the big difference is that our risk tolerance is much different.

I stay away from enabling LLC, as I love the idea that the mobo is protecting against transients. I also stay away from auto voltage settings as I like to see what I can do from within the specified voltage boundaries. I'm guessing you're at least busting the vtt max, but I'm not absolutely sure. What is your vtt running at? What temps do you show at 100% load at 4ghz?

When you say you've run stable LinX runs at 4ghz, can you be more specific? How much ram dedicated to each run, how many runs, or how long was your test? I found a large difference in what I call "stable", as I began to explore testing beyond 1 hour. I testing with 4gb ram. At 3.6ghz, my temps top out at 70C.

Cheers,
Bob

 

[ehume]

I wasn't paying any attention to vtt. And your idea of stable is beyond what I was doing: I was using the LinX default mem commitment and 20 reps. The whole point then was to get some preliminary notions of what the chip could do and what cooling I needed.

I no longer consider the LinX testing adequate. I cannot set a mem commitment that stays set. Also, the brief runs with rests do not represent a true torture test. So I now use OCCT. With that default I am now having trouble running even 4.0 GHz at acceptable temps. This is good, because it represents how hard I might push this machine. Also, when I bring it upstairs I will need a 12c headroom, since my basement ambient is around 18c and the upstairs room may get as hot as 30c.

So, when the new MB comes (the old one was making spurious sounds in the audio system) I will work on trying to make the machine rock stable at 1.25v. At that voltage LLC should not be a problem, since a spike would have to be 0.15v to touch the nominal limit.

For my part, I admire your success with P12's on the Mega. When I tested fans on a Megahalems, my Noctua results weren't great. But then, I did forget to test a P12 with an S12B.

 

[bob5568]

I will work on trying to make the machine rock stable at 1.25v. At that voltage LLC should not be a problem, since a spike would have to be 0.15v to touch the nominal limit.

For my part, I admire your success with P12's on the Mega. When I tested fans on a Megahalems, my Noctua results weren't great. But then, I did forget to test a P12 with an S12B.

My sense would be to DISable LLC, and then set the vcore in bios to whatever you need to (within reason) to end up with 1.25v as measured by cpuid under load. Now you would have the 1.25v that you desire, AND a known maximum limit to any transient (defined as that value in the bios). Thanks for the link on your fan tests. Most impressive. My results are not necessarily super great, I suspected your findings also, that the p12 may not be the best choice. I was swayed by the Noctua marketing that claimed the high pressure was the way to go for a cpu hs fan. I may try others at some point, but temp is not the only consideration for limiting my overclock to 3.6. I think I'm at the sweet spot of balancing performance against power, and against system stress.

 

[soccerballtux]

In retrospect, who cares if enabling LLC causes spikes when the target voltage is so low? Next time, I'm leaving it at Auto, which is where the MB put it when I started OC'ing the P55A-UD3P.

Gillbot made that thread about LLC but I have some posts in there where I sufficiently debunk his claims.

On top of all that, there are multiple filtering caps on the side of the CPU socket for this very reason.

 

[ehume]

Gillbot made that thread about LLC but I have some posts in there where I sufficiently debunk his claims.

On top of all that, there are multiple filtering caps on the side of the CPU socket for this very reason.

Good to know.

My main trouble was that when I disabled LLC I needed more than 1.3v to get OCCT to run five reps without crying Error. I'd rather start with a low volt and spike than run with the higher voltage.

@bob5568 - the preliminary results I got suggest that a Megahalems is more responsive to rpm and cfm than static pressure. Looking at the 38mm vs 25mm fans, the 38mm fans have no advantage, where 25mm fans with higher rpm's do have an advantage. But I repeat - all preliminary. I'm working on a more definitive study for our cooler.

 

[bob5568]

When you disable LLC, the number you are entering into bios has a different meaning, is it no longer the actual voltage your cpu will see, but rather a boundary for transients, a true "max voltage".

So, by setting vcore in bios to 1.3, LLC disabled, your cpu would see, something much less, and under load, even less. You'd need to query your vcore (cpuid) once under load before you'll know what vcore is actually appllied in that condition.

This is better than setting 1.25v with LLC enabled, because in that condition you have no information about transients. You have no defined true "max". Certainly at voltages that low, I'd expect it to work ok, but why take the chance. There is no advantage.

 

[ehume]

When you disable LLC, the number you are entering into bios has a different meaning, is it no longer the actual voltage your cpu will see, but rather a boundary for transients, a true "max voltage".

So, by setting vcore in bios to 1.3, LLC disabled, your cpu would see, something much less, and under load, even less. You'd need to query your vcore (cpuid) once under load before you'll know what vcore is actually appllied in that condition.

This is better than setting 1.25v with LLC enabled, because in that condition you have no information about transients. You have no defined true "max". Certainly at voltages that low, I'd expect it to work ok, but why take the chance. There is no advantage.

Thanks for that. I think I will need to explore that, perhaps with monitoring software.

 

[Gillbot]

gillbot made that thread about llc but i have some posts in there where i sufficiently debunk his claims.

On top of all that, there are multiple filtering caps on the side of the cpu socket for this very reason.

lol!

You made ONE post in the thread and you hardly "debunked" anything. It was an ongoing discussion and by no means a dictionary of facts. Nice try though, but if that is what you need to believe to help you sleep at night so be it.

 

[Gillbot]

Ironically, the current issue of "The Overclocker" covers a lot of this with much more in depth info. I suggest everyone go check it out and draw your own conclusions.

 

[bob5568]

Gillbot, I just read the current issue of The Overclocker (as per google). didn't see anything on LLC. Happen to have a link?

 

[Gillbot]

It talks about overshoot and undershoot voltages.

http://cde.cerosmedia.com/the_overclocker_issue_9/1S4b87c336a9eda012.cde/page/12

 

[imported_RJR]

When you disable LLC, the number you are entering into bios has a different meaning, is it no longer the actual voltage your cpu will see, but rather a boundary for transients, a true "max voltage".

I don't think that is 100% accurate bob, it appears that the power management design of the motherboard would have the last say as to the extent of the transients vs any voltage set in the Bios. The article referenced by Gillbot would seem to confirm this.

It talks about overshoot and undershoot voltages.

NICE article, I just wish they could have taken it ONE step farther and included LLC data as well. They sort of did will the cheaper boards (but not really) that eliminated droop from the get go, but it still doesn't completely answer the question of a better quality board will better designed power management circuitry and LLC enabled. Oh well someday.