Undervolting laptop CPU

[Matthew05]

Hello,

I'm undervolting my CPU on my laptop using ThrottleStop. I have a Lenovo Legion Y530 with i5-8300H and GTX 1050.

 

I changed the CPU Cache offset voltage to -125mV and I also increased the turbo boost time limit to its maximum [3670016 seconds(?)]. I'm also leaving the fans on 100% for the whole duration of my testing.
I've been changing the core voltage offset by -10mV at a time and testing it for almost 7 minutes with Aida64 with CPU, FPU and Cache stress test and I'm logging everything I need through MSI Afterburner and then converting it to Excel table.
What I'm sceptical about is how much I've lowered the core voltage and how high it still boosts. Have I've done anything wrong?

 

I'm attaching my Excel table with a finalized sheet which compares the original and 2 undervolt offsets and also the logs I've aquired from MSI Afterburner.

There is also a table with a few Cinebench R20 scores (I didn't bother to do it after every change of the offset voltage).

Results.xlsx

[Sorenson]

That is a really high undervolt you either won the silicon lottery or there is instability that isn't showing up in just a 7 minute aida64 stress test. For your concern, My experience is that as long as the CPU is receiving enough power it will boost as high as its rated. When it does not receive enough voltage you generally experience an app crash, freezing, BSOD, or a shutdown. 

[Matthew05]

I've seen someone undervolting as far as -350mV on an i7-8750H. It showed no instability, but the clocks were limited to just 3000MHz. I will test for instability and if anything happens, I can always bump it up by 10mV a few times. I'm not doing anything critical that I can't afford to loose on the notebook anyway

[Matthew05]

I've looked at my core voltages, it seems that it just keeps them low instead of lowering it by 250mV all the time.

These readings are from HWMonitor:
0mV offset:
VID #0 max: 1.307V
VID #1 max: 1.295V
VID #2 max: 1.291V
VID #3 max: 1.292V
 

-250mV offset:
VID #0 max: 1.069V
VID #1 max: 1.075V
VID #2 max: 1.085V
VID #3 max: 1.085V

 

I've included the voltage reading from AIDA64 as well:

Original.bmp

UV -250mV.bmp

[unclewebb]

Are you adjusting the CPU Core and CPU Cache offsets equally?  When you are adjusting the CPU Core offset greater than the CPU Cache offset you will reach a point where you are not accomplishing anything.  The CPU will simply ignore your request.  The VID voltage is just a number.  It is not the same as the actual voltage going to your CPU.  Run the TS Bench test for a consistent load and then watch your CPU temps.  If a big change in CPU Offset voltage does not show a measured and repeatable change in the CPU temp then it is not doing anything.  You can make voltage changes back and forth for some A - B - A testing while the TS Bench is in progress.

Same thing with changing the Turbo time limit.  ThrottleStop lets you set this to a ridiculously high value but if it is outside what the CPU is expecting, the CPU will ignore the request.

[Matthew05]

Thanks! I'll look into it right now. I haven't touched the CPU Cache offset after changing it to -125mV.

[unclewebb]

Intel only allows the CPU Core and CPU Cache to be offset equally.  There is a lot of debate on forums about setting them differently and how much differently before there is no gain.  

For data logging I would use ThrottleStop over MSI Afterburner.  It does a better job of tracking the CPU multiplier when a CPU is throttling.

[Matthew05]

It seems that the VID 1-4 show the actual voltage, it dropped the moment I've lowered the cache voltage as well. I'll test it a bit (lot) more. Thanks!

[Matthew05]

Lol, the thermal paste on a Lenovo Legion Y530 is REALLY crappy. I've just moved from 82-88°C to a 65-75°C. This is really great, now I can make even more and a lot nicer graphs