i5-750 OC w/ TB and EIST

[El Jefe]

Late to the party but here it goes.

To start off, I've read this: http://www.anandtech.com/show/2850/6
Which happens to be my exact mobo (GA-P55M-UD2) and processor (i5-750)

I'm trying to achieve a BCLK of 160 with turbo-boost and EIST enabled (so a range of core speeds from 1.4GHz to 3.8GHz) and memory multiplier of 10x for 1600Mhz.

For the most part I've been able to configure it and it runs.

However, I was having stability issues at idle. So I enabled Load Line Calibration in the BIOS, which seemed to stabilize everything.

But then the above referenced article states: "We still recommend having LLC disabled in most situations, but if you have to reach a certain overclock for benchmarking, then you should probably enable it." I interpret that as for my situation of everyday use, I shouldn't need it enabled.

In addition to this: http://www.anandtech.com/show/2404/5

And the fact that my mobo manual give this stern warning: "Note: Enabling Load-Line Calibration may result in damage to your CPU or reduce the useful life of the CPU."

So my question is this: is it more appropriate to adjust vcore and other voltages to get what I want? Or leave my system as is?

I have no experience tinkering with voltages, so if that's what I need to do please guide me. All I know is that Vcore is the target applied voltage to the processor. I keep seeing others and I have no idea what they do.

 

[sm625]

I ran an i5-750 for two years. I too had stability issues at near idle loads. My only solution was to reduce my overclock. I started at 3.8 and at the end of two years I was down to 3.2 and started having problems there too. I had tinkered with practically every possible voltage. I finally just went ahead and "upgraded".

 

[Yuriman]

EDITED : Decided to retype my thoughts.

My experience with LLC on socket 1155 has been that it doesn't work very well. The idea behind LLC is that it reduces the delta between low and high load voltages. The way it works on my board is to increase high load voltage a little and low load voltage a lot. For instance:

LLC off:
Idle - 1.0v
Low load - 1.25v
High load - 1.20v

LLC enabled:
Idle - 1.0v
Low load - 1.32v
High load - 1.32v


Depending on which level of LLC you use, the high load voltage may be a bit under or even overshoot your low load voltage. This sounds like a good idea, but in practice I've needed to use higher voltages to attain stability with LLC on for some reason. Example:

LLC off - 4.6GHz lowest stable voltage
Low load - 1.32v
High load - 1.29v

LLC enabled - 4.6GHz lowest stable voltage
Low load - 1.32v
High load - 1.32v

^ So, rather allowing me to have lower partial load voltages, it increases my minimum high load voltages for stability.

 

[El Jefe]

This is my experience with LLC, as it relates to socket 1155:

EDIT (1 minute)

You mean 1156

 

[Yuriman]

I don't own an 1156 system, so unfortunately I can't speak for that socket specifically.

 

[Flapdrol1337]

Normally if your cpu is heavily loaded the vcore drops a bit. LLC compensates for this, so you maintain the vcore under load.

On my i5 750 system if I enabled LLC I had a higher vcore under load than in idle, which was a bit strange.

Did you set your vcore manually or did you keep it on auto?

 

[El Jefe]

Normally if your cpu is heavily loaded the vcore drops a bit. LLC compensates for this, so you maintain the vcore under load.

On my i5 750 system if I enabled LLC I had a higher vcore under load than in idle, which was a bit strange.

Did you set your vcore manually or did you keep it on auto?

Auto. I didn't have the courage to tinker with them.

 

[mindbomb]

LLC and eist don't play well together I hear. Also, bclock changes and eist don't play well together either, since you are overclocking the p states as well.

 

[El Jefe]

EDITED : Decided to retype my thoughts.

My experience with LLC on socket 1155 has been that it doesn't work very well. The idea behind LLC is that it reduces the delta between low and high load voltages. The way it works on my board is to increase high load voltage a little and low load voltage a lot. For instance:

LLC off:
Idle - 1.0v
Low load - 1.25v
High load - 1.20v

LLC enabled:
Idle - 1.0v
Low load - 1.32v
High load - 1.32v


Depending on which level of LLC you use, the high load voltage may be a bit under or even overshoot your low load voltage. This sounds like a good idea, but in practice I've needed to use higher voltages to attain stability with LLC on for some reason. Example:

LLC off - 4.6GHz lowest stable voltage
Low load - 1.32v
High load - 1.29v

LLC enabled - 4.6GHz lowest stable voltage
Low load - 1.32v
High load - 1.32v

^ So, rather allowing me to have lower partial load voltages, it increases my minimum high load voltages for stability.

Thanks for this. And I hope you didn't think I was being snide with my socket comment.

In the first link in my OP, this is what they say "Vdroop was approximately -.005V with Load Line Calibration (LLC) enabled and -.01V with it disabled at idle. Under load Vdroop was -.015V with LLC enabled and -.03V with it disabled."

So yeah, at idle Vdroop is a smaller value with LLC enabled than with it disabled. Which would keep any negative overshoots above the minimum for the processor. Is that correct?

EDIT: And for my particular Mobo, there aren't any levels of LLC, just enabled or disabled.

 

[Yuriman]

If implemented correctly it should reduce the overshoot voltage at partial load. On my board it causes me to need more voltage rather than less, though.

 

[Flapdrol1337]

Auto. I didn't have the courage to tinker with them.

If you want a stable overclock at 3.8 I think you're going to have to.

anyway, these chips are pretty hard to kill, usually the board breaks before the cpu does, but since you're not doing a 4 core static overclock the extra load on the board probably isn't much of a problem.

what board are you using?

 

[El Jefe]

If you want a stable overclock at 3.8 I think you're going to have to.

anyway, these chips are pretty hard to kill, usually the board breaks before the cpu does, but since you're not doing a 4 core static overclock the extra load on the board probably isn't much of a problem.

what board are you using?

GA-P55M-UD2, just like what they used here: http://www.anandtech.com/show/2850/6

 

[Flapdrol1337]

whelp, abort!

the UD2 doesn't have a lot of power electronics around the socket, maybe see how far you can overclock with an undervolt instead.

edit: looked up some images, there's still a decent bit of stuff around the socket, but I remember gigabyte implemented a vcore limit in the newer bioses because people were killing them with overclocks, I'd be reluctant with long duration stress testing.

 

[El Jefe]

Thanks everyone for your feedback. I think I'm going to leave it as it seems stable and running. If something fails, I guess it'll be time to upgrade. This hardware is all beyond warranty at this point anyway.

 

[Ryanrenesis]

Thanks everyone for your feedback. I think I'm going to leave it as it seems stable and running. If something fails, I guess it'll be time to upgrade. This hardware is all beyond warranty at this point anyway.

I had the i5-760 and I upgraded to the i5-4690K and let me know you: OC'ing the 4690K is a breeze compared to the 760--took me literally 2 hours in total (figuring out BIOS + stress testing) compared to like 2 days to figure out my i5-760.

 

[BonzaiDuck]

EDITED : Decided to retype my thoughts.

My experience with LLC on socket 1155 has been that it doesn't work very well. The idea behind LLC is that it reduces the delta between low and high load voltages. The way it works on my board is to increase high load voltage a little and low load voltage a lot. For instance:

LLC off:
Idle - 1.0v
Low load - 1.25v
High load - 1.20v

LLC enabled:
Idle - 1.0v
Low load - 1.32v
High load - 1.32v


Depending on which level of LLC you use, the high load voltage may be a bit under or even overshoot your low load voltage. This sounds like a good idea, but in practice I've needed to use higher voltages to attain stability with LLC on for some reason. Example:

LLC off - 4.6GHz lowest stable voltage
Low load - 1.32v
High load - 1.29v

LLC enabled - 4.6GHz lowest stable voltage
Low load - 1.32v
High load - 1.32v

^ So, rather allowing me to have lower partial load voltages, it increases my minimum high load voltages for stability.

While the OP may have obtained his answers or made his decision, I'm looking at Yuriman's numbers, and thought I'd throw in my thoughts.

First. I may have been one of only a few who continued to point at the December 2007 Anandtech paper which used the QX9650 to explore the voltage concepts of VID, VCORE, Offset, vDroop and LLC. The pitfalls mentioned about LLC seemed to address the harmonics of voltage changes in idle-to-load and load-to-idle transitions. This raised the spectre of stability under load and instability going to idle, as well as the possibility of a momentary overshoot in the load-to-idle transition that was both hard to measure and a possible source of damage if it went too high, too often.

Yuriman's numbers seem odd to me, because they didn't reflect my own socket-1155 experience. Without LLC, vDroop is largest -- that is, it has its largest absolute value. I could implement vCore settings (Offset and Extra Turbo Voltage) without LLC and get instability. If I raised LLC, there wouldn't be a change in the other settings, but I'd obtain stability. And I have to apply a caveat: reading the 2007 article, I determined to assure there was ALWAYS a level of vDroop in load performance -- in the direction it was intended.

The monitoring software measure "current", "minimum" and "Maximum" VCORE. "High" load voltage would show up as minimum; "Low" load voltage would show up as maximum -- just at the time stress-testing ends. I never saw any change in the latter when I implemented modest LLC -- which I define as maintaining a certain level of vDroop.

So I would start with a vDroop of about 60 to 70mV with LLC disabled. If I set LLC to "High," there are two more extreme settings available on the scale and my vDroop absolute-value would shrink to 20mV.

You can monitor many things while stress-testing, so you can see VCORE and VID simultaneously as they change. I think when you see VCORE actually EXCEED VID, your LLC setting is excessive, and you would have eliminated vDroop (in the direction it was intended) entirely.

As to the harmonics and the overshoot. The overshoot will be higher whatever LLC you set above "none." The most severe spikes (which you will never really see) would occur under severe stress-testing as it transitions to idle or "high" load-voltage. You might attempt to estimate the delta in this spike, even if you don't know the actual values of the spike itself (which elude monitoring).

On the EIST idle side of the equation, you will actually see the EIST idle voltage increase: mine increased from about 0.995 to 1.013V. This effect was noted together with a discussion of choosing + or - offset and increased LLC.

There may be remedies to the mysterious idle-level instabilities. I thought maybe changing PHase and Duty Control settings to those which favor full-phase mode and current-balance versus thermal balance. My own idle instabilities would only occur in intervals of 10 days to a month, while those reporting similar problems for SB and IB chips were complaining of occurrences with much greater frequency. And I don't have those episodes anymore.

I've never noticed any degradation to my processor now over 42 months of 24/7 operation since my basic overclock settings were established.

That's all I can say in this matter. But I also believe things have changed with the processor generations after Sandy Bridge -- regardless what you do with LLC. I only say that, given the second-hand information I have from threads posted in the forums beginning with Ivy Bridge and ending with Devils Canyon and Haswell-E. But those changes seem to show up in high temperatures and limits to how far you can take your overclock.

 

[El Jefe]

Update:

So, I with the settings I had I was still getting BSOD's when starting cold (meaning my computer was off for the night and would start it up in the morning). It seems that after it had warmed up it was stable. (?????)

Not happy, I decided to try the base 133 BCLK on a cold start to ensure it works as designed. Which I'm pleased to report that everything worked fine.

So, I opted to try hitting 4Ghz with the following settings (taken from here: http://www.anandtech.com/show/2850/6)
VCore: 1.375V
VTT: 1.37V
PLL: 1.80V
VDDr: 1.66V
Additionally I disabled TB and the advanced core states, but left EIST enabled.
I set my BCLK to 200 and changed my memory multiplier to 8 and started the core multiplier at 20. LLC was disabled.

It ran, was stable, and was getting speeds between 1800 and 4000 Mhz (just like they said it would!). But it was pretty toasty. I think I was hitting 78C while gaming. (I've only got an air cooler.)

Going back, I changed my clock back to 160 and enabled TB and the advanced core states, and left LLC off and the voltages for 4Ghz. Seems pretty stable at the moment.

Though, I may try to get a more effective cooler and try pushing 4Ghz again. That was pretty cool.