Seeking Wisdom about "Per-Core" Over-clocking on the SB-K processor

[BonzaiDuck]

Every so often, I try to explore new approaches to my OC settings.

I'm mildly intrigued with "Per-core" multiplier settings, as opposed to "All cores."

So far, the information I've turned up on this topic -- derived from experience with a P67 motherboard -- isn't encouraging to my own standards.

Supposedly, one has to use a "Manual" overclocking approach, and any attempt to use "Offset" or other ("Extra Turbo") voltage adjustments will make the computer freeze or BSOD.

The other problem: from the same information source, I must use an LLC setting of "High" or higher. But I'd rather use "Medium" for a "4-core" limit of 4.5 Ghz, with the remaining multiplier settings at 4.6, 4.7 and 4.8.

Before I give up early on this pursuit, I was wondering if there were "better insights" or collegial experiences with "Per-core" that are more encouraging, or which would at least allow me to use "Offset" and "Extra Voltage for Turbo" settings instead of a fixed, "Manual" VCORE.

 

[Dufus]

I use to use additional turbo voltage for my old 2500k for 4.6,4.7,4.8,5.0 turbo, somewhere between 0.15 to 0.2V IIRC. Problem with using fixed for stepped bins is that you'll need the highest voltage for the highest bin which means you'll be running all 4 cores at a higher voltage than necessary. A bit of vdroop can help as well so medium LLC should not be a problem.

 

[BonzaiDuck]

I use to use additional turbo voltage for my old 2500k for 4.6,4.7,4.8,5.0 turbo, somewhere between 0.15 to 0.2V IIRC. Problem with using fixed for stepped bins is that you'll need the highest voltage for the highest bin which means you'll be running all 4 cores at a higher voltage than necessary. A bit of vdroop can help as well so medium LLC should not be a problem.

Thanks. I may give that a try.

I was a bit concerned by the lockup caused by my first attempt at "per core." It forced a CHKDSK of my boot drive, and it seemed to require more than pressing the reset button to re-initialize.

But you DO indicate the problem with it. While the "guide" I'd consulted suggested lower power and voltage overall, "lower" may be in reference to "highest."

My attempt at preparation for the "per core" adventure at least resulted in my finding stable settings for 4.5 ("all cores") at lower voltage, power and temperatures. The profiles for 4.6 and 4.7 are also acceptable from those perspectives, but it's nice to have more "options."

Incidentally. I've been using CPUID's HWMonitor to track voltages, temperatures and fan speeds. I notice occasionally that the Min and Max values for one fan can get 'out of whack." For instance, Min and Max might trap 0 and 19,200 RPM as opposed to what actually occurs. And I'm wondering if this might be the result of a PLL Voltage setting insufficient for the SMB. It is currently set between 1.65 and 1.68V, depending on the profile.

Maybe? Maybe not?

 

[BonzaiDuck]

Ho-ho!! That . . . is . . . . nice!!

So . . . I get the most rock-solid stable settings, with the lowest temperatures, acceptable VCORE -- for the highest frequency.

Then, change the OC profile to "per core" so there are a mix of speed settings: 47, 47, 46, 45. If one or two cores are loaded up, AIDA-64 shows "4700 Mhz." It bounces around between 4700, 4600 -- 4500 -- depending on whether there are one or two cores loaded, 3 cores loaded, etc.

Suddenly, the voltage under stressing drops, whether the test is affinitized or loading under all eight threads. The temperatures are now also lower.

Yes!! That IS nice!! The temps were "nice" for "all cores" @ 47, but this . . . is . . . nice!!