MSI GL73 vs Lenovo Legion power throttling

[freezerain]

Hello forum. Long story short my MSI laptop was thermal throtling, 20-25 watts at max. Today I changed thermal compound to liquid metal. My friend has Lenovo laptop with same cpu and stock thermal paste, so we did small comparison.
My MSI GL73 with 8750h have core/cache offset of -50 and have AC/DC loadline set to 1 in bios by advice from the post on notebookcheck (I think author of the original post is hanging around here as well). As you can see on the attached screenshot I got 12 thread / 960m test completed in 86 sec, max temp was 89, max power was 57w, no thermal throttling.

What my buddy got from stock Lenovo Legion something-something is 113sec, 95 temp, 73w, definitely thermal throttling.

I was interested in is how he was able to achieve 72w on his machine, so I reruned that test again and I got 96sec, 80max temp, 65watts, slower, less temps while more power on pl2 boost. My default values I believe was pl1-45w / pl2-60w.
I am understand that I am using this test for completly different reason that it was designed for, what I am trying to understand where this inconsistency in time, power package and thermals is coming from. I am not interested in getting 70w burst but I am worried that same mechanism can sometimes lower my pl1 behind my back while gaming and I am not completly understand how test with bigger max power was worth in time and cooler in temps.
In my multiple tests I have also noticed that after around 30sec wattage is droped to 45w, my TPL settings attached.
Can somebody point me what is happend, do I really even need to worry about this because my cooling system is already on the edge? Did some of my setting are messed up? Should I check unlocked bios for MSI secret capping power setting? Whats up with higher boost - slower result - cooler temps? Does my TPL "overlock" completly ignored? "Disable and lock turbo" in FIVR is enabled.

 

[unclewebb]

higher boost - slower result

A faster CPU will take less time to complete the TS Bench calculation. Less time is not a slower result, it is a faster result. Kind of like cars at the drag strip. The less time the better.

Intel CPUs do not report measured power consumption. It is an approximation. The only purpose of this number is to control the turbo boost function. It is not accurate, especially when the load line settings are not exactly the same.

Use Cinebench R20 to compare to your friend. A good 8750H with no throttling issues can maintain the full 39.00 multiplier for the entire test.

 

[freezerain]

ty for the reply uncle, I will clarify.
1 Test - 86 second with 56w pl2, 89c temp
2 test - 96 second with 65w pl2, 80c temp.

I see 2 issues with this result.
1 issue: why in second test cpu decided to give almost 10 watt more power package and still got worse time while being much cooler. I am not understand where is a choke point
2 issue: even while my pl1/pl2 have 80w in TPL it is in fact clamped to 45/65 while testing. Thats why like you said in cinemabench after 30 second i got 45w and 34 multiplier while holding 85 degrees during enitre test. So with those thermals I should be able to maintain close to 39 FID but I have no ability to test it because I am 45 wattage locked by MSI?
I am not sure if I even diagnose the issues correctly.

And this is what I got with cinemabenh.
1 Test - I have no cpu info on that but the result is 6147
2 Test - 63w max, 94c max, 6700 score.
Like I said after first 30 second cpu switched to 45w and 85c. So I should be fine on 50w atleast but cpu ignoring TPL
And results again is very different, I am trying to not change anything in my system rn to be consistent and before test I am always making sure CPU is running like 2w idle so any invisible background windows updates shouldnt affect test
Am I stuck on 45w and is 15% difference in test is expectable behaviour?

 

[unclewebb]

10 watt more power package

Intel CPUs do not report measured power consumption. They have a formula that uses the VID voltage table built into the CPU. This approximates actual voltage going to the CPU and this approximated voltage data is then used to approximate actual power consumption. Two different 8750H can have significantly different voltage tables. I would not use the power consumption data when trying to make comparisons between different laptops even if they have the same CPU model.

it is in fact clamped to 45/65 while testing

This is a common problem. The 8750H has a 45W TDP rating.

Intel® Core™ i7-8750H Processor (9M Cache, up to 4.10 GHz) Product Specifications

Intel® Core™ i7-8750H Processor (9M Cache, up to 4.10 GHz) quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

Some manufacturers use an embedded controller (EC) to enforce the 45W long term TDP limit. The turbo power limits are duplicated within the CPU. ThrottleStop and Intel XTU have no access to the third set of turbo power limits controlled by the EC. If a manufacturer sets this third power limit to 45W then the lowest power limit wins and long term, the CPU will throttle based on the lowest set limit. On most laptops, there is no easy fix for this limitation.

In the advanced MSI BIOS you might be able to access the IMON Slope and IMON Offset variables. If you set the Slope to 50, the CPU will report half as much power consumption. When it is consuming 45W, it will only report that it is consuming 22.5W. That can allow you to run your CPU way beyond 45W. Now it will not start to power limit throttle until it hits 90W because the CPU thinks and will report that it is only consuming 45W.

This is another reason why power consumption data from Intel CPUs is a meaningless number. It can be easily manipulated. I have never done any modifications to the IMON Slope or IMON Offset variables so search Google for these terms and you should be able to find some more info.

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

I am more familiar with Cinebench R20 scores. With unlocked power limits and no thermal throttling, an 8750H should score over 3000 points in R20. If a 45W limit is being enforced, the typical score is more like 2500 or 2600. I have heard of some Razor laptops enforcing a 35W limit. If you have one of them, you will barely be over 2000 points in R20.

 

[freezerain]

Uncle this is exactly what I am was looking for. I dont understand why power throtling is even a thing. My bios for sure have IMON slope/offset however I dont know what they doing exactly and will investigate deeply, I think Falkentyne bios guide have something to begin with.

Ty for clarification with power consumption data, I was expecting it to not be 100% accurate but atleast somehow meaningful.
I will run CN R20 and will compare with yours data, however I am expecting it to be on lower end even with liquid metal aplied.

Ty for everything you doing and good luck

 

[unclewebb]

I think the default setting for IMON Slope is 100. If you set it to 75, the CPU will report 75% as much power consumption as default. Set it to 50 and power consumption will be reported half as much as normal. Set it higher than 100 and your CPU will report that it is consuming more power than what it actually is. Definitely do not want that.

Run a simple and consistent test like 4 or 8 or 12 threads of the TS Bench test. Have a look at power consumption. Reboot, change IMON Slope, run the same load and see if power consumption is reported differently. If your laptop blows up, it wasn't me that suggested doing this.