2600k limit search -CPU voltage reporting offsets

[daddylongs]

I have started using AI Suite to more rapidly determine the stable cpu voltage for a given speed (GHz). I am seeing a mismatch in what I tell Turbo Evo to use as compared to what CPU-Z is reporting.

CPU-Z reports only seem to vary in .08 voltage increments, so there is a bit of masked noise in this data.

Anyway, can you help me explain the offset I am seeing here?

IBT = Intel Burn Test 2.52

Each of those data points was found to be stable for a few minutes with IBT. The last one I tried to then push for a longer test and after 20 minutes it presented me with a BSOD. I can't be mad


It seems linear for now but I need more data points.

 

[Idontcare]

EVO = Targeted Voltage for CPU, this is the voltage value that the BIOS is attempting to drive the Vcc to.

CPUz = Measured/reported value of Vcc as far as the mobo's sensors are able to determine (CPUz value will be the same as that of the BIOS sensors)

Multi-meter = This is the "real" Vcc that the CPU is seeing, within the error of the multi-meter itself of course.

CPUz values are quantized in a more coarse-grained granularity than the BIOS settings for Vcc in EVO. The values in CPUz step in 0.008V.

You can use CPUz, just recognize that it is not giving you granularity out to the third decimal point as the reported value would have you believe.

^ measured values were taken with a multi-meter using the ProbeIT feature on the MIVE-Z with system under load from IBT

 

[daddylongs]

Thanks for the confirmation and the correction (.08 != .008)

My mobo software has a tool called Probe II but it seems to have the same .008 granularity.

I do have a nice Fluke multimeter, if that is helpful at all? :\

All 3 of your data columns' #s are close to each other. I was surprised by the constant 5% (.07v) offset on mine.

Nothing to worry about? Just keep watching what happens to CPU-z #'s? I'll try to keep tracking these #'s.

 

[daddylongs]

Data point on AVX results:

Only difference between these two runs was AVX or not. Results:

No AVX: 51.2 Gflops, Tmax = 62
Yes AVX: 100.6 Glops, Tmax = 67

Quite a difference there, thanks for the nudge.

 

[Idontcare]

All 3 of your data columns' #s are close to each other. I was surprised by the constant 5% (.07v) offset on mine.

As scary as this will sound when it sinks in, I have to tell you its BIOS rev dependent.

My MIVE-Z shipped with Bios rev 203 (or 202?). The delta between the BIOS/CPUz Vcc and the multi-meter was significant. On the order of 0.05-0.06V, just as you are seeing.

Then I updated my Bios to rev 403. Now you see the results, they are so close.

So yours is what it is, and worse yet is that every time you get a new BIOS you can't assume the old offset values are valid anymore.

Data point on AVX results:

Only difference between these two runs was ANX or not. Results:

No_AVX: 51.2 Gflops, Tmax = 62
Yes AVX: 100.6 Glops, Tmax = 67

Quite a difference there, thanks for the nudge.

Ah, excellent confirmation :thumbsup: Logic wins the day yet again

Do me, and the community, a favor and repost your results/post in the sticky thread so we have it documented there that using AVX really is necessary to get the most out of your IBT stress testing.

Good stuff, thanks so much for sharing it with us :thumbsup:

 

[daddylongs]

Done, and thanks for the BIOS tips in relation to T offsets.

I toyed with my 4.5-4.9 range last night: below points are min Vcc to be stable for ~7 minutes of LinX_AVX (~100GFlops):

And the same data, but max Temps:

And here is the round-up of my Voltage offset data with default settings.

======Round2:
After that I started playing with the other software OC options in the "DIGI + VRM" panel and now behavior is different. The offset seemed different (larger, more fluctuation - got as high as 1.42 from a Evo setting of 1.245), the results were different, like it does not reduce the cpu speed when idle, etc. But anyway it ran for hours @100 Gflops stable and then hours of P95 stable. Both of the below images are from the same run. Am I doing it right? Should I stay away from these DIGI+VRM settings - I am concerned that the Vcc will spike up too high unexpectedly?

 

[Idontcare]

daddylongs, you certainly have far more working experience and knowledge of the Digi+VRM panel than I do. But if you think it would help, I can capture the settings I have been using for all my tests and report them here.

It's not going to be very interesting though, I believe I am using the "Auto" setting for LLC in the BIOS and the "Optimized".

 

[daddylongs]

daddylongs, you certainly have far more working experience and knowledge of the Digi+VRM panel than I do. But if you think it would help, I can capture the settings I have been using for all my tests and report them here.

It's not going to be very interesting though, I believe I am using the "Auto" setting for LLC in the BIOS and the "Optimized".

The offer is appreciated, but don't worry about it. I'll keep doing some experiments.