Offset Voltage, raise min Vcore?

Hello all, I made a new thread for this just because it seems that each overclock is vastly different, and a lot of material I was reading was in regards to "good" overclocking i5 chips. Mine does not appear to be very high on the spectrum.

-----
I spent this weekend overclocking my computer, and things went decent until I got to tuning the offset Vcore.

Asus P8Z77-M (2105 Bios, newest)
i5 3750k
Corsair CX500M
GSkill Ripjaw X pc3 12800 (8-8-8-24 @ 1.5V)

So it is pretty clear that my i5 isn't as good of an overclocker as most. Using fixed voltages I require 1.12-1.25V to maintain 42 ratio. 1.30V for 43 ratio. I'd prefer to stay under 1.3V in all situations, so I will probably stick to 42.

I wanted to use offset Vcore, and began to experiment. To get any stability at all I need to use a +.015 Voltage. This puts my min Vcore at like .984 and max at 1.296. (Or something in that area.) In all situations any change from 75% LLC down to say 50% LLC causes problems, at least based on my current settings. If it didn't I would use this to raise the offset voltage but lower my max Vcore.

A few notable facts:
-I can't seem to change my PLL voltage. Only Option I have in my Bios is +.1V (which would be from 1.8V to 1.9V)
-Speed step enabled
-C1E, C3, C6, C state all enabled at the moment (I realize this may be the culprit)
- LLC at ultra High (75%), Fixed Freq 350, Phase control extreme, Current capability 120% (max in my bios)


From my testing at +.015V the system is very solid under prolonged Stress testing, benchmarking software, etc. At +.005V I would get some weird Chrome errors, so I upped it. At .015V I get some odd crashing of applications. (counterstrike crashes at map change only) These seem to be low load situations, which leads me to believe my min voltages are possibly to blame. Although it may also have something to do with my C states (which I admittedly do not understand)

I am trying to utilize Speed step and Offset to lower my clockspeeds/idle Voltages. This isn't so much to save power as it is just to unload the components. I will not claim to understand the benefit of C states, but as a general direction I prefer to have as many "features" running as possibly, especially since I have such a tame overclock.

Are my symptoms pointing to a common issue? What benefits do the C states give? Is there any way to control your min Vcore in offset mode with anything other than offset voltage and possibly LLC changes?

Thanks all!

Did some more digging and found these threads:

http://www.overclock.net/t/1079110/idle-bsods-and-c-states

Which specifically say to Disable C3 and C6 for idle BSODs. I'll play around with that and then come back to this. If anyone wants to chime in in the mean time I appreciate it, but for the moment I'll due some more diligence.

Those changes to C states definitely made this much stabler and brought up my min Vcore values with different offsets.

Crappy part now is I am having a hard time stability testing. I seem to get a bogus voltage value over extended stress testing. Whether its a random 127 C mobo temp or a bogus voltage detected.... HW Monitor has impossible values scattered through it. (2.000 Vcore for instance) I don't know if these bugs are what caused Prime95 to fail, but they are certainly annoying to see what values things peaked at since it masks the true values. I did have logging running on core temp and that is really the only thing that's worth looking at.

I was using in parrallel: Cpuid, CoreTemp, HW Monitor, and Asus AI Suite II. I think I will cut those down to 1, probably HW monitor... and run the fan's at the moment off the bios control instead of the software until I determine stability.

Happily though, with the most recent offset settings the Vcore only spikes to 1.248v and doesn't break 75C even under extended stress testing.

This has become an ongoing log and I am talking to myself, but that's okay! It will help someone someday!

Tried again last night, Prime ran for ~7 hours until it errored out. I upped the voltage a bit and set it in motion again. Can anyone chime in about LLC? I know it's intention is to prevent drooping of voltage as load conditions vary. I notice it lowers my max Vcore under full load (which is what I was trying to achieve) I have 50% LLC right now to keep the max vcore down. I think my range is like (-.010v offset (.984v - 1.256v)) With LLC set to 75% the max is up at like 1.275.

I think I'd prefer to not be so aggressive on my offset but keep LLC at 50% to lower the max vcore and thus max temperatures at load. Shall keep testing! I just don't want to assume that my stability issues are because of too low voltages, when in reality its because of fluctuations due to LLC.

With my p67 and z68 boards I have to keep most of those settings at default, or auto. Changing them only introduced new issues. Between those 2 boards I required less offset voltage for the z68 one with the same cpu and other hardware. Probably differences in power phases.

Running p95 on win7 produces slightly more heat than running it in win8.1. a solid base for more investigation from the offset, and try different values.

So all in all, offset can be tricky.

ps, while you're talking in yourself, we are still reading it.

ps2 in answer to your questions, altering llc and pll overvoltage etc create more voltage when required, thus you could lower offset values to compensate. But an x value of cpu speed still requires an x value of voltage, regardless of how you get it.

I've found so many threads here and across the internet of people talking to themselves that in the end find a good solution. There's a comic somewhere about a guy finding an old abandoned thread about a troubleshooting issue with the caption "Where did you go, what DID YOU SEE?" I try to conclude my random issues with final updates for this purpose. If it helps one person the next 5 years it will have been worth it

-.010 Offset voltage with 50% LLC. Vcore range .984-1.264. Stable for 12 hours of prime95. Apparently I was just right on the line with my first try. May try 25% LLC with the same offset.

Coretemp says my VID at that ratio (42) is 1.0558 idle and 1.3260 at load. Not sure if that means anything, but yea.