VCore Spikes, GA-EX38-DS5, LLC

[thrnbot]

Hi there,

I am having trouble with VCore spikes for a while now and cannot figure out how to solve this issue. Would much appreciate your help! I already read many interesting posts in this forum and so I thought this might be the right place to ask.

My Hardware: GA-EX38-DS5 (BIOS version: F5B) + Q6600 (rev G0)

The Problem: VCore spikes (variations) make the system unstable. Lessened potential for overclocking.

What I observe: Say I set my CPU Voltage in the BIOS manually to 1.2250v and CPU clock to 320MHz. The actual idle VCore may then be 1.17v. However, once I put load on the CPU (e.g. small FFTs), it rockets to 1.30v (with high variations). See attachement "erratic". I assume that sudden load changes make the system unstable (because of voltage peeks).

What I want to know: How to permanently disable load-dependent VCore adjustments?

What I tried: I already read a lot about Loadline Calibration, EIST and C1E. As far as I understood, disabling all of them should answer my question. In fact, I did, and for some time, it did seem to work just fine (see attachment "stable"). However, for some reason unknown to me, after some time (havnt found a pattern yet), the system always seems to revert somehow to enable Loadline Calibration without me changing anything in the BIOS. I have not been able to disable the spikes since.

Best,
Matthias

 

[redduc900]

LLC should be enabled, not disabled to control Vdroop and voltage fluctuations. C1E and EIST are just power-saving functions built into the processor that allow the multiplier and Vcc to drop when at idle (from x9 to x6; 266Mhz x 6 = 1.6Ghz). If you haven't already, try flashing the BIOS to the latest beta F5B revision to see if that helps any.

 

[thrnbot]

Ok. Resetted BIOS. Loaded Failsafe-Defaults (LLC, EIST, C1E on). Voilá! Stable VCore again.

FSB 333 + VCore set in BIOS to 1.225V leads to observed VCore under load (sFFT) of ~1.18V (idle: 1.21V).

However, after suspending the system overnight and waking up the other day, same problem again. Now, voltage under load jumps up to ~1.36V (without any BIOS changes or restarts). How can this be explained? How can a suspend-wakeup lead to such drastic changes in VCore behaviour?

Attached the dmesg log of the suspend-wakeup part (I am running Fedora).

 

[redduc900]

I assume you've already tried flashing the latest F5B BIOS?

 

[thrnbot]

Jep. F5B is the latest version:

http://www.gigabyte.com/products/product-page.aspx?pid=2761#bios

 

[thrnbot]

Ok, update on the issue, insights explaining much of what was unknown to me.

The increases in VCore after standby are likely related to the core temperatures. As the temperatures after long standby/shutdown are very low, VCore goes up. This accounts for up to +.2V differences between idle and load VCore. With increasing core temperature, voltage drops, and needed power (Watts) increases.

EIST and C1E are unrelated. LLC, as far as I observed until now, does nothing more then a static +.03V voltage increase. So, instead of 1.28V idle VCore, I see 1.31V, given the same core temperatures. I have it disabled.

 

[thrnbot]

Hi there again,

new insights. I set up a more elaborate logging/testing framework (script attached).

Here is what I do: log sensors readings every 200 secs for idle and load (mprime sFFT).

So, clock is set to 300MHz, Bios Vcore to 1.175V. Normal readings are idle Vcore: 1.15V-1.17V, Vcore sFFT: 1.14V. Abnormal readings are idle Vcore: 1.12V-1.14V, Vcore sFFT: 1.18V-1.25V.

Now the interesting thing is that the switch between the two does not only occur afters suspends but also abruptly during system operation (see log).

Another very interesting finding is that the difference between the Core0 (c0) and the CPU temperature readings differ drastically between the described modes: in the normal mode the delta betwwen load and idle is about 10°C. Now, in the abnormal mode, delta under load is about 20°C! The CPU readings seems to be off by -10°C.

How can this be explained? I hope someone of you can shed some light into this. I have been pondering about this for weeks now.

Matthias