undervolting i7-9750H - still remaining heatpeaks

[Biohaze]

Hey there I'm the new one^^,

I'm trying to cool down my heat machine y540-15irh by undervolting it.

I've already done this steps (see pics, data) with the shown results.
Log-File: https://we.tl/t-jyYFGGIkDq

The RTX 2060 is remaining on a cool lvl 50-65°C

What can I do to get away these peaks and as result of it, the throttling of the CPU?

Thanks

 

[Kingtanar]

As far as I know, 8th and 9th gen Intel CPUs benefit from a 2:1 ratio when undervolting, which means, you could try undervolting the core twice the value you have on the cache. This way you would have -240mV on the core and the same -120mV on the cache you currently have.

If I remember correctly, after -225mV on the core there were no gains, but you should try that by yourself with Cinebench R20 (it's faster than R23). Do 3 runs with different voltage values on the core and see which setting gives the best average score from the 3 runs and, of course, CPU temperature.

With all that said, if you still have high temps, repasting could be a solution + lifting the back of the laptop to increase the space of air intake for the fans.

 

[unclewebb]

@Biohaze - Have a look in the Options window. Lenovo likes to low ball the thermal throttling temperature. This is controlled by the PROCHOT Offset setting. If you do not see a lock icon near this setting, you can lower the PROCHOT Offset value which will raise the temperature when thermal throttling begins. The Intel default for this offset is 0. Some laptops use a value of 2 or 3 so throttling starts just before 100°C. Do some testing. If you find a value you like, you can use the Lock PROCHOT Offset option.

MAXON Cinebench R20.0 Download

CINEBENCH is a real-world cross platform test suite that evaluates your computer's performance capabilities. CINEBENCH is based on MAXON's award-winn

www.techpowerup.com

Most 9750H owners get better Cinebench results with the core much higher than the cache. Run a baseline test with the core and cache set equally at -120 mV. After that, start bumping the core in steps of -25 mV.

-120 mV cache and -200 mV core usually works well for Cinebench.

You might need to reduce the core for some games. Maybe -120 mV cache and -175 mV core. This can help bring the temps down a degree or two. Not much else you can do. Even with undervolting, all laptops with 9750H CPUs tend to run at full temperature.

Try turning on the Log File option in ThrottleStop. These log files cover what is important and leave out what is not. Some log files are "too much information".

 

[Biohaze]

As far as I know, 8th and 9th gen Intel CPUs benefit from a 2:1 ratio when undervolting, which means, you could try undervolting the core twice the value you have on the cache. This way you would have -240mV on the core and the same -120mV on the cache you currently have.

If I remember correctly, after -225mV on the core there were no gains, but you should try that by yourself with Cinebench R20 (it's faster than R23). Do 3 runs with different voltage values on the core and see which setting gives the best average score from the 3 runs and, of course, CPU temperature.

With all that said, if you still have high temps, repasting could be a solution + lifting the back of the laptop to increase the space of air intake for the fans.

Hey Kingtanar,

thank you for your tip. I'll try the steps with Cinebench 20.0.

Repasting is, at the moment, no choice. I'll lost the warranty if i do so....unfortunately

@Biohaze - Have a look in the Options window. Lenovo likes to low ball the thermal throttling temperature. This is controlled by the PROCHOT Offset setting. If you do not see a lock icon near this setting, you can lower the PROCHOT Offset value which will raise the temperature when thermal throttling begins. The Intel default for this offset is 0. Some laptops use a value of 2 or 3 so throttling starts just before 100°C. Do some testing. If you find a value you like, you can use the Lock PROCHOT Offset option.

MAXON Cinebench R20.0 Download

CINEBENCH is a real-world cross platform test suite that evaluates your computer's performance capabilities. CINEBENCH is based on MAXON's award-winn

www.techpowerup.com

Most 9750H owners get better Cinebench results with the core much higher than the cache. Run a baseline test with the core and cache set equally at -120 mV. After that, start bumping the core in steps of -25 mV.

-120 mV cache and -200 mV core usually works well for Cinebench.

You might need to reduce the core for some games. Maybe -120 mV cache and -175 mV core. This can help bring the temps down a degree or two. Not much else you can do. Even with undervolting, all laptops with 9750H CPUs tend to run at full temperature.

Try turning on the Log File option in ThrottleStop. These log files cover what is important and leave out what is not. Some log files are "too much information".

Hey Unclewebb,


thx for your tip. I'll take -175mV core as target an will do some research about the 2:1 core:cache ratio mentioned by @Kingtanar .

At the moment my Lenovo Legion just hit's the 85°C mark with coolingpad and the undervolt specs posted at the beginning.
I'll test some variations.



Ah, I have read that the cache and core should have the same -x mV. This infos are outdated?

I dont find anything about a 2:1. Do there are some references?

 

[unclewebb]

The 2:1 ratio was an early theory but it is not true. Increasing the CPU core voltage beyond the CPU cache voltage is a trick that reduces the voltage the CPU uses when it is running AVX instructions. The maximum difference that works is somewhere around 100 mV.

If you set the cache to -100 mV then your benchmark results will probably continue to increase until the core reaches -200 mV. Yes, that is a 2:1 ratio but that is just a coincidence.

Ah, I have read that the cache and core should have the same -x mV.

This myth got started because that is all Intel XTU allows you to do. There are two separate voltage registers in the CPU. Only ThrottleStop gives you access to both registers individually. It was not until users started doing some proper testing that they discovered this trick.

A user sent me his Cinebench test results. Hard to argue with his numbers.

Cinebench Test.zip

drive.google.com