Pretty significant difference in power readings (Package Power vs VR Power) ?

[MiniSpip]

Hello,

CPU Package Power max is 175W, which is normal since PL1/PL2 are limited to that value in the BIOS. VR VCC Power peaks at 214W and I was wondering whether that 40W difference is something to be worried about, or normal ? Peak VCore is 1.426V.



The CPU is not overclocked it's actually the opposite: It is slightly "underclocked" by tweaking LLC and AC_LL values. Could this be the cause of the difference ? Or is it simply because the VR is also supplying power to other components ?

Thanks in advance.

 

[Martin]

Actually the VR VCC value should be lower than CPU Package Power as it drives the cores rail only, so it should be closer to IA Cores Power.
If the board has dedicated VR sensors you might check those.

 

[MiniSpip]

Thank you for the quick reply.
I activated Asus EC support and some VRM sensors. The "VRM Vcore Power (POUT)" is also higher than the "CPU Package Power" and even a bit higher than the "VR VCC Power (SVID POUT)"...


The fluctuations are a bit weird too, as for current readings: VRM Vcore Power (POUT) can be much lower than CPU Package Power (5W vs 30W) and then peak above it. I checked with a very fast polling period and it's the same.

Is this indicative of a power issue, or something else ?

 

[Martin]

It's hard to compare in scenarios when the power is fluctuating much. Best would be to test at relatively stable load/power.

 

[kryponix]

Hello,
My first post. If I forget sth, pls, let me know. I have a related issue. Most likely because there's sth I forgot to enable or disale in BIOs or hwinfo, but after hours of searching and reading I'd given up.
Long story short, my vr vcc power values are always way higher than CPU Package Power. For what I've read everywhere it should be the other way around. No?
I did not play much with bios setup options from auto or default (a bit with CPU Lite Load mode, but changing this to different modes, does not seem to change the readings).
I attach a screenshot. Mobo is MSI MAG Z790 Tomahawk wifi (ms-7d91) (Nuvoton NCT6687D)
P.D. Sorry if I miss sth basic that explains this situation.

 

[Martin]

Hello,
My first post. If I forget sth, pls, let me know. I have a related issue. Most likely because there's sth I forgot to enable or disale in BIOs or hwinfo, but after hours of searching and reading I'd given up.
Long story short, my vr vcc power values are always way higher than CPU Package Power. For what I've read everywhere it should be the other way around. No?
I did not play much with bios setup options from auto or default (a bit with CPU Lite Load mode, but changing this to different modes, does not seem to change the readings).
I attach a screenshot. Mobo is MSI MAG Z790 Tomahawk wifi (ms-7d91) (Nuvoton NCT6687D)
P.D. Sorry if I miss sth basic that explains this situation.
View attachment 12050

Indeed, looks like either of those values is not correct. What's show at full CPU load? Can you attach the HWiNFO Debug File for analysis?

 

[kryponix]

Indeed, looks like either of those values is not correct. What's show at full CPU load? Can you attach the HWiNFO Debug File for analysis?

Thanks Martin,
Agree. The previous screenshot is iddle (more or less). When iddle, VR VCC is constantly much higher than CPU Package Power. I attach a debug file in this situation. I had to compress it, as there's a size limit in the server for upload (it was originally around 6 M). If readings were correct, it is like there were a multiplier in the power requested by CPU rails as long as it does not reach PL1. All of this is IF I understand what VR VCC POUT means (not sure at all). And in that case there is sth set up incorrectly (by default I guess) in my bios.
When at full load readings change and both measures are closer (perhaps as expected?). I attach a screenshot. I also have the debug file (11 mins info) but I do not know how to attach it since even compressed with the best method it is greater than the server limit (do I need to change default polling interval?). I can truncate it, or take another shot with less time, but not sure whether that will miss important debugging info. Do not hesitate to let me know if you need it and how to proceed.

P.D. Yep, I know, I'm very conservative with my PL1 conf

 

[Martin]

I think there are 2 factors playing a role here:
- The VRM is rounding up power values via SVID interface, so at lower load the results are not accurate.
- At lower load and transient situations, it picks up short spikes.
Important is that at full load the results match.

 

[kryponix]

Thanks Martin.

 

[falkentyne]

Asus EC has resolution down to 1mv on vcore and 1A reporting, and 1W on POUT.
(You can see these values on Asus OCTool "Raw VRM"). I am pretty sure Asus EC Field in HWinfo uses these values also (though I have Asus EC disabled due to latency polling issues, similar issues happen when Raw VRM window is open).

SVID I think is like 2A on IOUT and 15mv(!) on vcore.
Super I/O Vcore (e.g. CPU-Z vcore) is (at least on Asus boards), I think 9mv on vcore.