CPU running 100mhz lower than its supposed to?

[USRNME]

I have a i7-10750H and despite setting the 6 core turbo ratio to 43, when running Cinebench or any other CPU intensive task, the maximum it will reach is 4.2 ghz. This happens regardless of what I set the turbo ratio to, for example: If I set it to 42, it will only run at 4.1ghz etc. Speedshift is already at 0 (Windows battery slider). I have attached my settings below. Thank you in advance.

 

[DarkDreams]

What other CPU intensive tasks have you tried? C20 uses AVX instructions afaik, so it might be an AVX offset.

 

[TheLostSwede]

Supposed to? You are aware that the clock speed is somewhat up to the notebook manufacturer and the cooling solution they implement, right?
There's Intel's spec, but most mobile CPUs have variable TDPs these days that the notebook manufacturer decides over, which in turn means variable clock speeds.

 

[Flaky]

See if disabling Thermal Velocity Boost helps.

 

[USRNME]

What other CPU intensive tasks have you tried? C20 uses AVX instructions afaik, so it might be an AVX offset.

Its the same in C23, in games and in timespy, always 100mhz lower than what I set in the turbo ratios.

 

[unclewebb]

See if disabling Thermal Velocity Boost helps.

That is the problem.

Open the FIVR window and clear the Thermal Velocity Boost box. What Intel calls boost is really throttling in disguise.

Also update your version of ThrottleStop. The new version properly supports your 10th Gen CPU. Limit Reasons will show TVB throttling in red when you have this throttling problem.

 

[USRNME]

That is the problem.

Open the FIVR window and clear the Thermal Velocity Boost box. What Intel calls boost is really throttling in disguise.

Also update your version of ThrottleStop. The new version properly supports your 10th Gen CPU. Limit Reasons will show TVB throttling in red when you have this throttling problem.

Thank you very much!

 

[unclewebb]

Thank you very much!

Post a screenshot if you found your missing 100 MHz.

 

[USRNME]

Post a screenshot if you found your missing 100 MHz.

haha I did find the missing 100 mhz, but it brought a little hitchhiker called power consumption, so now I just use the 4.3 profile for high score runs.


However, I have another question regarding the actual undervolt itself. I have noticed that my laptop tends to crash right after a benchmark / closing a game, does that just mean it needs more voltage? Because I've read somewhere online that Intel CPUs tend to undervolt better at higher clock speeds / higher load, so using speedshift to maintain boost clocks gives more undervolting head room, is this true from your experience? I guess I just don't want to admit that I lost the silicon lottery big time lol (Can't even get close to the -75mv most 10750H's can )

 

[unclewebb]

but it brought a little hitchhiker called power consumption

As the old saying goes, there are no free lunches. That last 100 MHz takes lots of voltage and creates lots of extra heat. Rarely worth it except maybe when bench testing.

Offset voltage reduces voltage at both ends of the MHz curve. Reducing the core voltage -75 mV when running constantly at full speed might be fine but as soon as you remove the load and the CPU is allowed to idle down, an instant crash is not unusual. That is definitely a sign that your undervolt is not 100% stable. A computer should never crash at any speed.

Intel CPUs save most of their power when they are idle when the cores enter low power C states like C7. This means slowing the CPU down does not reduce power consumption as much as you would think. Idle cores in C7 are already sitting dormant at 0 MHz because they are disconnected from the internal clock. If you are plugged in, use the Windows High Performance power plan to keep your CPU at its maximum speed. This might allow you to get away with a -75 mV or higher undervolt because the CPU will not be dropping down to 800 MHz. It will run at full speed when it has a task to perform or it will be at 0 MHz in C7 when it has nothing to do. Right away people are thinking, you cannot do that with a laptop. It will be like an oven when running at full speed when idle.

Here is the testing I did which compares my 10850K when idle at 5000 MHz vs idle at 800 MHz. Not a huge difference. No difference in idle temperatures. If a fast CPU allows you to undervolt a little more, you might be further ahead going fast instead of slow.

 

[USRNME]

If you are plugged in, use the Windows High Performance power plan to keep your CPU at its maximum speed.

Thanks for the detailed answer, I'm already using the high performance plan (I assume this ultimately does the same thing as the battery slider method when it comes to setting speedshift to 0, right?). Could the crashing be due to the voltage offset being stable at all core turbo boost clocks e.g. @4.2 ghz, but not stable when each core is fluctuating between 4.3-4.6 ghz?

 

[dnm_TX]

@USRNME try also to lower your core UV to like....-125 mV and then gradually increase the cache until you reach something close to -75.2 mV(by testing of course). Same CPU here and i remember when i was doing my UV,setting up the core at -150 mV froze the entire system,so by my count,you're right there,at the danger zone . Also another thing to note is that when i was testing with CB23,anything past the -125 mV mark,not that it didn't give me better score but even lowered it with couple points after each run,so keep this in mind as well. Hope this helps.

 

[USRNME]

@USRNME try also to lower your core UV to like....-125 mV and then gradually increase the cache until you reach something close to -75.2 mV(by testing of course). Same CPU here and i remember when i was doing my UV,setting up the core at -150 mV froze the entire system,so by my count,you're right there,at the danger zone . Also another thing to note is that when i was testing with CB23,anything past the -125 mV mark,not that i didn't give me better score but even lowered it with couple points after each run,so keep this in mind as well. Hope this helps.

Interesting that your core UV was the bottleneck instead of the cache, I've always thought that the cache was the limiting factor. I'll try that tysm.

 

[dnm_TX]

Oh....one more advise. On the cache,i'd strongly recommend to test it with TS bechmark and set it to random(i believe it's called). If it goes through,consider it stable. Increase,'til you get the blue screen.

 

[USRNME]

Oh....one more advise. On the cache,i'd strongly recommend to test it with TS bechmark and set it to random(i believe it's called). If it goes through,consider it stable. Increase,'til you get the blue screen.

How long / which size?

 

[unclewebb]

setting Speed Shift to 0

When your computer is idle at the desktop, check the FIVR monitoring table to make sure that Speed Shift EPP really is set to 0.

Some 10th Gen mobile CPUs are simply not 100% stable at -75 mV. This is not a bad thing. It just means that Intel is doing a better job of matching the default voltage curve to the voltage that the CPU actually needs. You could run a -125 mV cache undervolt on the majority of 8th Gen 8750H but you will rarely if ever get away with that on a similar 10th Gen mobile CPU.

 

[dnm_TX]

How long / which size?

Don't matter(the shortest,i guess)...if the cache UV is unstable,you'll get the blue screen right away,trust me....i know

P.S. @unclewebb is there a new stable release planned in the near feature? Give us some update pls.

 

[USRNME]

When your computer is idle at the desktop, check the FIVR monitoring table to make sure that Speed Shift EPP really is set to 0.

View attachment 242209

Some 10th Gen mobile CPUs are simply not 100% stable at -75 mV. This is not a bad thing. It just means that Intel is doing a better job of matching the default voltage curve to the voltage that the CPU actually needs. You could run a -125 mV cache undervolt on the majority of 8th Gen 8750H but you will rarely if ever get away with that on a similar 10th Gen mobile CPU.

After taking @dnm_TX 's advice of lowering the core instead to push the cache , I did manage a higher cache undervolt (The more I tweak this CPU, the more it goes against my existing knowledge of undervolting LMAO). However, I ran into another problem; I can't seem to break 3200 on R20. I have attached a log file of my *hopefully current* highscore run of 3190. For the whole run the limit reasons remained blank, clock speed was completely locked at 4.3ghz, temps maxed out at 89 and power maxed out at 85.2W. What seems to be the problem?

EDIT: I know I've undervolted the cache more the core, but in my testing this way of undervolting has been quite stable, even in my most "sensitive" game / scenarios.

 

[unclewebb]

If you set the core to -120mV and the cache to -160mV, both will be set internally to -120mV. I do not think this accomplishes anything but I might be wrong. Setting the cache to -120 mV and the core to -160 mV might accomplish something when the CPU is processing lots of AVX instructions like Cinebench uses. This trick seemed to do something on many 8th and 9th Gen mobile CPUs.

Your CPU is running consistently at maximum speed. Not sure how to flog a better Cinebench score out of it.

You are not too far off the world's best scores.

Cinebench - R20 overclocking records @ HWBOT

Overclocking records

hwbot.org

What memory are you using and what speed is it running at?

 

[USRNME]

I do not think this accomplishes anything but I might be wrong.

Funnily enough, this is actually more stable than my previous setting of -143mv core and -43mv cache, despite it being a bigger cache undervolt, so at this point I don't even know lol.

I suppose at this point I should just accept my 3190 score, and in terms of my memory , AFAIK most 10750Hs can't run anything beyond 2933 - even with a BIOS mod since the IMC is locked at 2933 (And my sticks are actually 3200 lol) so the only memory improvements would probably just be from lower latency sticks. Anyways, TY for the replies