Disable Speed Shift EPP Control on Windows 10

[xink64]

Hi all,
New at this forum!

I've got a Razer Blade Pro 17 2020 with a RTX2080 TI Super and Intel 10850H.

I've unlocked the bios with a hardware programmer, which means unlocked thermals, unlocked tdp, undervolting etc. so it performs really well as the geekbench5 scores testify.

I can't seem to set the Cache Ratio higher than 47 in TS, which means it never runs 5.1 Ghz.
I think this is why Linux does a lot better single core, or Windows just been busy with stuff in the background, but otherwise I'm really impressed with the multi core score on both OS.

Clearly not a standard intel 10875H score as those numbers are between Ryzen 9 5800H and 5900H, crazy!

Wonder why Intel did not allow for those kind of performance numbers, but instead locked 10th gen down hard.
It makes the chips look terrible compared to the 5th gen Ryzen.

Anyway, my problem is the EPP control in Windows 10.

Setting power settings to high performance gives EPP value 0 or mid or low gives 84 and battery at a much more relaxed 128.

If I change it myself in throttlestop 9.2, FIVR shows first 84 and then jumps to the value I've set it for.

I know TS is not in control, since the CPU is not clocking down properly if I set it for 255 for example.

I want a lower EPP as the laptop runs pretty high frequencies under not much load.

So any game runs at maximum thermals basically being at 100 Celsius constantly.
The simple solution is that I force the max frequency to around 3.7 to 4.0 ghz which averages around 85 degrees instead, but the EPP way would be much more elegant, I think.

Does anyone know how to disable Windows 10 own EPP control and take the power back to TS as it used to be?

Thanks,

 

[unclewebb]

I know TS is not in control

Did you use ThrottleStop to switch to the Windows High Performance power plan? This might allow the Speed Shift EPP value you set in ThrottleStop to go to the CPU with less interference from Windows 10. This used to work but any Windows update might have changed this.

Watch the Speed Shift EPP value reported in the FIVR monitoring table. If this value changes back and forth or if the Speed Shift EPP value in this table is not the same as the Speed Shift EPP value that you have requested then Windows is in control and there is nothing you can do about it.

Some BIOS versions will allow you to disable Speed Shift Technology. This might let you use Speed Shift EPP with no interference from Windows. The problem is that the Disable Speed Shift setting in the BIOS might completely disable it. If your BIOS does this, the Speed Shift EPP box on the main screen will be gone. It will not be possible to enable Speed Shift while you are in Windows. If this box does not disappear, maybe this trick will let you fully control Speed Shift. Windows will not be Speed Shift aware.

Setting EPP to 255 does not guarantee that the CPU will always run at the lowest speed. If Speed Shift EPP is showing 255 in the FIVR monitoring table then the CPU is using this setting.

Your Geekbench comparison is using two different versions of Geekbench so it might not be a 100% fair comparison.

The 10850H has a locked maximum CPU multiplier. Be default, the maximum cache ratio is also locked. The cache is usually locked to 300 MHz less than the core. When the core slows down, the cache automatically slows down too.

What does ThrottleStop report for C0% when idle? Some users are unaware of how much stuff is running in the background. I like consistent benchmark numbers so I keep background tasks to a minimum. I use software like ShutUp10 to keep Windows 10 from being too bloated. When running single threaded tests, the maximum CPU multiplier and CPU speed is highly dependent on this.

 

[xink64]

Thank you for the elaborate answer, very helpful!

I've set power profile with ThrottleStop instead and now FVIR reports correct EPP value immediately.

I agree, Geekbench5 versions might show slight differences, but 10% higher SC is a lot.

I'm also very aware about my background services and I use Shutup10 as well, really useful for privacy!

The combined C0% value is around 3%, but this is with everything enabled. IntelliJ IDE, Docker, Brave Browser /w 9 tabs, Steam, Discord, Viber and a bunch of others.

On my old 8950HK and 4710HQ, I'm pretty sure 255 EPP meant minimum clock speed all the time or very close to it.

On 10875H with Win10 20H2 it means, whatever random clock speed under the sun. I see spikes to 3.0,4.0, 4.4 Ghz constantly while just browsing.

If I disconnect the power brick, 255 EPP means at least less than 2 Ghz, with occasional 2.2, 2.7 Ghz spikes.

I'm no expert on EPP, but that sounds wrong to me.

I will take a look if I can either find or hack my way to disabling EPP in BIOS.

UPDATE:
I found a few switches, but none for EPP.

I did try running 5.4 Geekbench on Windows with and without background services.

Might be interesting, but problem is, everything turned off is not a meaningful way to find out how the machine performs in daily tasks. I don't just do one task at a time

Anyway, version bump aside, I'm still 73 points short on SC against a standard Manjaro installation without tweaks to it.
MC looks mighty good though!

 

[unclewebb]

On my old 8950HK and 4710HQ, I'm pretty sure 255 EPP meant minimum clock speed all the time or very close to it.

The 4th Gen 4710HQ did not use Speed Shift. You are correct that setting EPP to 255 used to be far more restrictive. With 10th Gen, Intel must have changed the algorithm. My 10th Gen desktop will slow down to 800 MHz when idle. At full load, it goes up to full speed whether EPP is 0 or 255. The EPP value controls how much load it takes to get up to full speed but it no longer prevents the CPU from reaching its maximum speed. ThrottleStop is still writing 255 to the EPP register just like it always has.

http://imgur.com/vV91Elm

I think the reason why this has changed is because holding a CPU to a low speed when it has something to do is inefficient. The "race to sleep" model is what is recommended. Allow the CPU to quickly get up to full speed so it can quickly process whatever needs to be done. If it has nothing to do, go back in to the low power core C7 state where the core sits at 0 MHz and 0 Volts. When people see an idle CPU at 5000 MHz, they automatically think that it is consuming a huge amount of power. If the C states are enabled, that is not true.

Power Optimization - A Reality Check

https://www2.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-140.pdf

You can still use Speed Shift Max to control the maximum CPU speed and maximum temperatures.

10% higher SC is a lot.

There is only a 0.3% difference in the multi core scores you originally posted. That is well within the margin of error and shows that there is not a huge difference between Windows and Linux in this benchmark. Not sure why you are seeing a big difference in single thread. If Linux was magically extracting way more performance out a CPU core, this would also show up in the multi thread tests but it doesn't. I would not lose too much sleep over this one test.

everything turned off is not a meaningful way

But it is the only way to make a meaningful comparison when benchmark testing. It sounds good if a CPU is only averaging 3.0% in the C0 state when idle but with a 6 core - 12 thread CPU, Windows 10 can be as low as 0.1%. Your idle has 30X as much stuff running compared to my idle. This can interfere with the CPU using the highest single core turbo multiplier.

Here is a good test. Run something simple like a single core Cinebench test. The 10850K can use a 52 multiplier when 1 or 2 cores are active. The 10850H has a 51 multiplier available when 1 core is active. What multiplier does ThrottleStop report during this single thread test? Is it consistent or does it vary significantly while testing? Lots of background tasks will force the multiplier to drop down. When additional cores go active, the maximum turbo boost multiplier automatically drops.

http://imgur.com/qXowc5y

 

[xink64]

Sorry for my late reply, been busy.
But you are right, the 4th gen Intel was a simpler time, I totally forgot!
That chip was easy to control!

About the Race to Sleep principle:
It makes sense towards energy and battery usage, but not heat build-up causing fans to spin loudly while doing lighter tasks.
You could argue that Razer's fan speed control is to blame, but fact is, the idea leads to less thermal headroom when a harder task coming along because the cooling system is already heavily loaded beforehand.

This is all in the realm of how to do CPU scheduling. I remember playing a great deal with Android's modding scene's many schedulers and kernels for the perfect balance between performance, battery and heat on a host of devices over the years.

Right now, I think the best solution is to force max clocks to be lower, but that's not very clever with a chip that can ramp 5.1 Ghz

Since EPP is no longer helpful, could a lower thermal max temperature be the way? F.ex. 85 or 90 degrees, or am I just reducing performance in a slightly different way than before?

I will try your idea for testing ability to reach 5.1 Ghz at some point, when time permits!

Thanks for all the insights and guidance, much appreciated!

 

[Kugumin]

I want a lower EPP as the laptop runs pretty high frequencies under not much load

You can reduce your EPP value from power plan settings, but as far as I know it will be hidden for you, so at first you have to unhide it. I can help you with that, just PM me