Laptop undervolting and overclocking results... i7-10750H RTX 2060... MSI Raider GE75

[EEAM.Peru]

Thanks for the reply,

When set both PL's to 90W it draws 80W aprox and then drops to 50W because of thermals. So I set both to 69W and now Im getting 69 Max constant. The problem is the cpu doesn't go up to 4.3 Ghz it stays at 4.0Ghz at 98.8% cpu load and its not thermal throttling. I believe there lies the problem. Weird because I haven't changed any internal settings on Windows since it's a new laptop and my settings are identicalto yours.

Set both Iccmax to max btw

 

[unclewebb]

then drops to 50W because of thermals

It would be very unusual for a CPU to drop from 80W to 50W because of thermals. That sounds like power limit throttling.

Turn on the ThrottleStop Log File option or open Limit Reasons and show that in your picture when your CPU is loaded running Cinebench. Showing ThrottleStop after Cinebench is finished is not going to show what the problem is. Remember to exit HWiNFO when using Limit Reasons. A log file while running Cinebench is going to show the problem.

 

[rruff]

Your CPU will run at 4.3GHz in that test unless something is throttling it. It will be either a power limit or temperature limit. >80W sounds like no UV to me. If you set both PLs to 69W and get ~4.0GHz, that also sounds like no UV is being applied.

 

[unclewebb]

no UV is being applied

The FIVR monitoring table shows that the undervolt is being applied. The cache is at -60 mV. Not a huge undervolt but it is being applied.

CPU power consumption data from Intel CPUs is not measured power consumption. It is an approximation that can vary from one CPU to the next, even if they have the same model number. Some report high and some report low. Actual power consumption could be exactly the same. When it is not properly measured, no one knows.

The ThrottleStop screenshot posted shows PL1 is set to 55W. Hopefully that screenshot was taken before he increased the power limits.

 

[EEAM.Peru]

Thanks again both for your replies.

The following is TPL1-TPL2 55-75 and then 69-69
Cooler bost on (max fans button) + No other background program besides TS and R20 + Dragon Center doesn't make a difference on my tests (with and without)

69-69 seems to be the best IMO since it doesn't thermal throttle. all the options on TS are the same difference is only changing TPL1 and TPL2
Scores between both favors 69-69 by 50-60 on R20.

 

[unclewebb]

69-69

At 69W or even 75W, your CPU is power limit throttling. That is why it does not maintain the full 43.00 multiplier. If you found a way to improve cooling, you could raise your power limits higher and get a little more performance but not much. Maybe 4300 MHz instead of 4000 MHz. Most real world apps or games are not going to fully load all 12 threads of your CPU so they should be able to run at full speed without any power limit or thermal throttling. For Cinebench, you are right on the edge of both of those limits.

 

[EEAM.Peru]

At 69W or even 75W, your CPU is power limit throttling. That is why it does not maintain the full 43.00 multiplier. If you found a way to improve cooling, you could raise your power limits higher and get a little more performance but not much. Maybe 4300 MHz instead of 4000 MHz. Most real world apps or games are not going to fully load all 12 threads of your CPU so they should be able to run at full speed without any power limit or thermal throttling. For Cinebench, you are right on the edge of both of those limits.

Thanks again for the reply,

I suppose Ruff's model has better cooling being a 17 inch laptop (they usually have more room for cooling and fans) and thats why he can sustain those 4.3 Ghz. I forgot that he also has a MSI Raider which is more expensive and better on quality compared to this Leo GL65.

Tomorrow Im getting a cooling pad with fans. I will retest with that and post my final results here so it can help others.

 

[unclewebb]

cooling pad with fans

Most cooling pads only lower temps by a degree or two. Make sure you can return it to the store if it does not solve your problem.

 

[rruff]

I forgot that he also has a MSI Raider which is more expensive and better on quality compared to this Leo GL65.

In the 69W run you are power throttling at 4.0GHz, while I reach 4.3GHz with that reading. The reason for that may be what unclewebb stated regarding the "approximation" of power readings. Cooling wouldn't account for it. Your temperature is also reaching 95C which could be a cooling issue. I've been testing in ambient temperatures ~20C; maybe it's warmer where you are? The Raiders are a little more expensive but they share the same chassis as the Leopard... and I think the same cooling. The size might be a difference as you mentioned. Another contributor could be thermal paste. I regularly see 7-8C core temp variations in a test like this, and yours are even higher at 9C. Another thing you could try is simply raise the back of the laptop a little by resting it on something to improve airflow... just make sure you don't block the vents.

Tuning to run benchmarks is fun, but not so practical. I really like TS for being able to set different profiles. I set limits via frequency rather than power as I suspect this gives more stable performance.

 

[LcT89]

Thanks rruff for the detailed guide. Helped me with a MSI GL65 Leopard same cpu but 2070 Super.

Just a heads up for anyone reading this post. I followed this guide exactly but couldn't reach the 68W draw PL1 at first. It always went down to 45W after the sustained PL2 of 75. Looked everywhere turns out the MSI Dragon Center was set to Balanced. Changed it to Extreme Performance and now I can get that 68W sustained power draw instead of defaulting to 45W.

Besides that I dont know how are you hitting a 3300 score on Cinebench R20. Could it be because I haven't undervolted my GPU? (although it doesn't make sense since Cine doesnt use the GPU). Im hitting 2800 on an average of 5 tests (was 2700 on the same average before Undervolting). Can you tell me what tests you used and for how long each you used for testing stability? Thanks in advance.

PS. Disabling Prochot and amping the limit from 95 to 99 didnt increase the score on R20 just upped my overall temp limits (like it should I know)

Screens from my TS settings and HWINFO64 when running R20 Multi.

Just a question: I have your exact same laptop model, but in you screenshots cache ratio min/max is 8/47. Mine can't go higher than 43 (i have alredy unlocked the overclock inside the bios). How is that possible?

 

[rruff]

Just a question: I have your exact same laptop model, but in you screenshots cache ratio min/max is 8/47. Mine can't go higher than 43 (i have alredy unlocked the overclock inside the bios). How is that possible?

Mine says 8-47 also, and I'm pretty sure I didn't touch that. Don't know what it means.

 

[unclewebb]

Don't know what it means.

It does not mean too much. When ThrottleStop starts up for the first time, if there is no previous ThrottleStop.INI configuration file, it will read this information from the CPU. Whatever the BIOS sets this to will be read by ThrottleStop. It might vary from one laptop to the next depending on what values the BIOS programs this to.

The important number is the Cache Ratio value in the FIVR monitoring table. That is the cache multiplier that the CPU is using. The cache multiplier on the locked non K CPUs are usually limited to 3 less than the core multiplier. Even if you set the cache Max to 100, you will still be limited by what the CPU is capable of.

 

[LcT89]

It does not mean too much. When ThrottleStop starts up for the first time, if there is no previous ThrottleStop.INI configuration file, it will read this information from the CPU. Whatever the BIOS sets this to will be read by ThrottleStop. It might vary from one laptop to the next depending on what values the BIOS programs this to.

The important number is the Cache Ratio value in the FIVR monitoring table. That is the cache multiplier that the CPU is using. The cache multiplier on the locked non K CPUs are usually limited to 3 less than the core multiplier. Even if you set the cache Max to 100, you will still be limited by what the CPU is capable of.

It's possible that something happened on my CPU and now can't go above 43? Watching the table and making stress tests, even resetting the ini file and bios, it varies from 8 to 43.

 

[unclewebb]

@LcT89 - What CPU model do you have? If you try to set Cache Max to a big number in ThrottleStop and then press Apply, it will automatically be reduced to what the CPU supports. What happens when you do this?

The cache ratio in the FIVR monitoring table can vary depending on what speed the CPU cores are running at.

 

[LcT89]

@LcT89 - What CPU model do you have? If you try to set Cache Max to a big number in ThrottleStop and then press Apply, it will automatically be reduced to what the CPU supports. What happens when you do this?

The cache ratio in the FIVR monitoring table can vary depending on what speed the CPU cores are running at.

i7-10750h, laptop MSI GL65 10sfsk Leopard.
If I set any number above 43, it resets the number to 43.
It seems strange to be because when I search screenshots of this with i7-10750h, I always see the 47 limit

 

[unclewebb]

It seems strange

You are right. It is strange. In the ThrottleStop FIVR window on the right hand side, what does it show for your Microcode version? I have seen 10750H screenshots that have a Max 47 cache multiplier. They show microcode 0xC8.

When fully loaded, the maximum CPU multiplier is 43. Try doing a quick 12 thread TS Bench test. What cache ratio is reported in the FIVR monitoring table when the CPU is loaded? Typically the cache runs at 3 less than the core so you should see a cache ratio of 40 during this test if the multiplier is able to maintain 43. If there is power limit or thermal throttling, the cache ratio will drop with the multiplier.

Edit - Are your turbo ratios set to their default values?

 

[LcT89]

You are right. It is strange. In the ThrottleStop FIVR window on the right hand side, what does it show for your Microcode version? I have seen 10750H screenshots that have a Max 47 cache multiplier. They show microcode 0xC8.

When fully loaded, the maximum CPU multiplier is 43. Try doing a quick 12 thread TS Bench test. What cache ratio is reported in the FIVR monitoring table when the CPU is loaded? Typically the cache runs at 3 less than the core so you should see a cache ratio of 40 during this test if the multiplier is able to maintain 43. If there is power limit or thermal throttling, the cache ratio will drop with the multiplier.

Edit - Are your turbo ratios set to their default values?

my microcode is 0xE2 AND i suspect that could be the answer.
Dell G7 i7-10750H undervolting | TechPowerUp Forums
also this guy has the E2 and i remember a recent intel optional update delivered by microsoft (laptop is brand new but a 2020 model, so i made all updates in the past weeks)


With all defaults (deleted ini, reflashed bios from official MSI laptop's repository):
on single core bench, cache stays at 43
on multi core (12) starts from 40, then drops at 38 when all CPU drops. Same behaviour with cinebenchr23 (except the fact that it drops earlier)
I have to redo the undervolt, but with a -125 on cache, -100 on core, -60 on gpu intel and -40 on igpu unsilce, also mantaining a lower prochot of 89 instead of 95, two days ago i scored 8100 on a single iteration of cinebenchr23.

Edit: 8500 with all fans at max speed
Edit2: 8200 with a little less undervolt for stability and auto fan speed

Ps. it's been a long time since I wrote in english, so i'm sorry for that

 

[unclewebb]

my microcode is 0xE2

That is probably what happened. ThrottleStop reads the maximum cache ratio value from the CPU. Intel must have changed this in the most recent microcode so now the maximum value is 43 instead of 47. This might reduce single thread performance a tiny amount. When fully loaded or playing a game, the cache ratio will probably be the same so this change will probably not make any real world difference.

 

[rruff]

Edit2: 8200 with a little less undervolt for stability and auto fan speed

I get >8800 in C23. Max temp 91-92C. All cores 4.3GHz. Ring/LLC at 4.0GHz... is this what you are looking at? If so then ours are running the same despite mine have a higher ratio. On single core Ring/LLC hit a max of 4.6, but it was ~4.35 on average. Score was 1265.

If higher fans help your score, then you must be temp throttling. 95C should be fine on your CPU; no good reason to drop it since it already has a big safety margin.

I've been using a core UV that is -.100V relative to cache. Cache is the critical one IME. Mine is only -.061V, with -.161V on core. If you can set you cache -.100 or less that is excellent.

Your english is great!

 

[unclewebb]

On single core Ring/LLC hit a max of 4.6

Do not update your BIOS or your microcode or your cache will be limited to 4.3 GHz.

 

[rruff]

Do not update your BIOS or your microcode or your cache will be limited to 4.3 GHz.

I think I have updates turned off...

I'm curious how much difference it makes in performance... or why Intel changed it? I tried setting the ratio to 43 and TS locked up my computer (twice).

 

[unclewebb]

TS locked up my computer

ThrottleStop did not lock up your computer. You did it!!

Did you lower the Cache Ratio Max value?

You should be able to adjust the CPU core or cache ratios at any time to any value without the computer locking up. If it does lock up, this is usually a sign that your undervolt is on the edge of stability. Try increasing your voltages.

Maybe Intel changed this because some of these CPUs are not 100% stable at the higher cache ratios. That is usually why something changes. Reduce the warranty claims.

 

[rruff]

ThrottleStop did not lock up your computer. You did it!!

Did you lower the Cache Ratio Max value?

Yes, that's the number I changed. I could expect that raising it might cause an issue, but lowering? Sure enough, I set all my voltages back to default, and no issues changing it to 43 then

I ran TS Bench single core a few times with 43 and 47 and there appears to be ~0% difference in speed. So I don't think having it maxed at 43 is something to worry about.

 

[unclewebb]

I set all my voltages back to default, and no issues changing it to 43

I think you just came up with a new stability test. No need to run Prime95 for hours on end torturing your CPU. To test your undervolt, just make sure you can adjust the core and cache multipliers without your computer locking up. That was the logic behind the Random option in the TS Bench test.

 

[LcT89]

I get >8800 in C23. Max temp 91-92C. All cores 4.3GHz. Ring/LLC at 4.0GHz... is this what you are looking at? If so then ours are running the same despite mine have a higher ratio. On single core Ring/LLC hit a max of 4.6, but it was ~4.35 on average. Score was 1265.

If higher fans help your score, then you must be temp throttling. 95C should be fine on your CPU; no good reason to drop it since it already has a big safety margin.

I've been using a core UV that is -.100V relative to cache. Cache is the critical one IME. Mine is only -.061V, with -.161V on core. If you can set you cache -.100 or less that is excellent.

Your english is great!

Wow, big score! Have you tried the tvb option?

Do not update your BIOS or your microcode or your cache will be limited to 4.3 GHz.

I don't think that is a bios update, but one of the many optional KBs from Microsoft...