Adaptive Voltage with Haswell (And what is the benefit?)

[flexy]

(Asus Z87-Pro, 4770K)

I established that I can run my 4.5ish overclock at MANUAL 1.210 Vcore which I do for a while already.

I cannot see a benefit of switching to Adaptive (as many guides claim there is) since even on "manual" I see processor c-states active and HWInfo64 or HWmonitor tells me indeed that very little Vcore is on the CPU while idle.

(For example now, set to 1.210 V manual I have 0.144V on two cores and 0.000 V on two others, sometimes I actually see all FOUR cores at 0.000 V (HWInfo64). Because of that I don't see a benefit turning on Adaptive because there really is not much more power to save than what I do with c-states already??

* I understand the principle of Adaptive voltage but the Z87 Pro bios is not clear. It shows (for adaptive) one "additional voltage for core voltage" and another entry for "additional voltage for turbo mode" and then a total. Then it shows the current Vcore in BIOS, say, 1.080.

What I did I entered 0.065 in the first field and another 0.065 for "additional voltage for turbo mode" which added up (1.080 +0.065 + 0.065) to the 1.210 where I KNOW that this is my target voltage at load. (NOT talking AVX now, that is another story). However, with that setting OCCT freezes instantly.

(It is not entirely clear to me how to calculate the 'right' values for the two fields in BIOS...what to use as additional voltage for Non-Turbo and how much to add for "turbo" and of course the entire benefit as mentioned earlier). I am also confused since the bios has TWO voltages here and not just one of course...

 

[BonzaiDuck]

(Asus Z87-Pro, 4770K)

I established that I can run my 4.5ish overclock at MANUAL 1.210 Vcore which I do for a while already.

I cannot see a benefit of switching to Adaptive (as many guides claim there is) since even on "manual" I see processor c-states active and HWInfo64 or HWmonitor tells me indeed that very little Vcore is on the CPU while idle.

(For example now, set to 1.210 V manual I have 0.144V on two cores and 0.000 V on two others, sometimes I actually see all FOUR cores at 0.000 V (HWInfo64). Because of that I don't see a benefit turning on Adaptive because there really is not much more power to save than what I do with c-states already??

* I understand the principle of Adaptive voltage but the Z87 Pro bios is not clear. It shows (for adaptive) one "additional voltage for core voltage" and another entry for "additional voltage for turbo mode" and then a total. Then it shows the current Vcore in BIOS, say, 1.080.

What I did I entered 0.065 in the first field and another 0.065 for "additional voltage for turbo mode" which added up (1.080 +0.065 + 0.065) to the 1.210 where I KNOW that this is my target voltage at load. (NOT talking AVX now, that is another story). However, with that setting OCCT freezes instantly.

(It is not entirely clear to me how to calculate the 'right' values for the two fields in BIOS...what to use as additional voltage for Non-Turbo and how much to add for "turbo" and of course the entire benefit as mentioned earlier). I am also confused since the bios has TWO voltages here and not just one of course...

I'm not rocking a Haswell chip, so I'll approach your question with a bit more caution while inviting those with Haswell experience to comment.

Somehow, I think you "been aroun' for awhile" from your posting statistics, but I could bet (or just guess) that you're jumping from LGA-775 into "Gen 4."

You seem to indicate you are setting VCORE by "manual mode." There should be an "Offset mode" option -- alternative to "Manual". I also cannot tell if you have enabled EIST, C1E and the additional "C" BIOS power-saving options; I would recommend doing it with those items enabled, as much as someone else would say that it's somehow "easier" with them enabled initially to play with a fixed VCORE.

In Offset mode, the +/- voltage increment simply raises the effective voltage above (or below) what is given by the processor VID at some multiplier.

IF there is (and there should be) an "Extra voltage for Turbo" item, I argue that it is best to use that one to adjust the maximum voltage applied at the full multiplier-overclock. You could alternatively simply raise the Offset, but this can cause problems with idle-state instability if you choose some level of LLC. It is equally possible to reduce the Offset into the "minus" range, but that could potentially lead to the same difficulty. If there's an "Extra . . Turbo" option, it is best probably to simply set the Offset to "+" and the minimal increment -- probably 0.005V.

You're best to get some baseline readings from the BIOS monitor at stock settings, and use those baselines to adjust the voltages as you increase the multiplier -- beyond the point where the system is stable with "auto" settings at some higher-than-stock multiplier. By the time you've started to exceed that point, you should have a fairly good idea where to set the "turbo" voltage (and/or offset) simply by checking the BIOS monitor (to see if it's increased or decreased with an initial setting).

You should be able to trade Offset adjustments for "Extra . . Turbo" adjustments, but they are likely in different increments. On my motherboards, Offset is specified in increments of 0.005V, while "Extra" is specified in increments of 0.004V. So you could decrease "Extra" by two notches and increase Offset by two, to get a net increase of 0.002V under my motherboard's regime.

 

[Burpo]

6 1/2 minutes into this..
https://www.youtube.com/watch?v=Z7zPu9255ZI

 

[BonzaiDuck]

6 1/2 minutes into this..
https://www.youtube.com/watch?v=Z7zPu9255ZI

That nails it. And prevents him from detours due to my being short about these newest processors and boards.

 

[flexy]

Ok I see clearer now, see:

Only the first is an OFFSET, the second is not. I didn't catch that previously.

And it is correct as I read in some other forums there is indeed no "advantage" in using adaptive, NOT with c-states active. (Which they are for my system). There won't be more power savings in idle. (Which was the basic idea of "Adaptive" since it would add more voltages at higher clocks, depending on clock rate but leave voltage low at idle etc. But it's really moot with c-states).

 

[BonzaiDuck]

Ok I see clearer now, see:

Only the first is an OFFSET, the second is not. I didn't catch that previously.

And it is correct as I read in some other forums there is indeed no "advantage" in using adaptive, NOT with c-states active. (Which they are for my system). There won't be more power savings in idle. (Which was the basic idea of "Adaptive" since it would add more voltages at higher clocks, depending on clock rate but leave voltage low at idle etc. But it's really moot with c-states).

Did you get any comments or cautions about idle-level instability with negative offsets greater than minimal? The problem seems to walk hand in hand with use of LLC, but I think these newer boards don't require LLC. I think someone here mentioned that, or it was in the video link.

 

[flexy]

LLC only applies to VCORE and not to VID (confusing here, I know), not the actual VID, eg. the voltage you set in bios to overclock. VCORE here meaning the total voltage supplied to the CPU which USUALLY is about 1.800V. It has minimal impact on modern boards/CPUs. This is my understanding.

I have not experimented with negative offsets or looked further into adaptive or offset, simply because I see my CPU uses almost no power anyway due to c-states/EIST. I really don't see the point. I am ASSUMING that people who do use negative offsets might be unclear in how this CPU really works or maybe they have Speedstep etc. DISABLED for some bizarre reasons. Speedstep and c-states does ALL that already, eg. turning down multi and giving the CPU less VID, there is just no sense to me to mess with negative offsets too.