SpeedShift Value Reliable?

[a_bolog]

I'm trying to optimize the profiles for Full Performance, Internet, Gaming, etc.

I've noticed that the SpeedShift EPP value has an impact on the max CPU frequency in a rather "unpredictable" way. Maybe because I don't understand what it does.

Example:
Run a game, or static camera Heaven Benchmark.
EPP value is set to 100, max CPU is 3.6Ghz.
Change EPP value to 60, max CPU is 4Ghz.
Change EPP value to 0, max CPU is 5Ghz.

All good and as expected.
I'm using the High Performance Windows profile, so windows doesn't change or interfere with the value. Clock speeds are set to 5Ghz, so it's uncapped. Max PL1 and PL2, uncapped.


The question:
Is it a reliable way of configuring the profiles, purely based on the EPP values?

I'd like to say yes because I've noticed this other behavior:
EPP value 100, CPU is 3.6Ghz while running the game. BUT if I run something else in the background, open chrome, render a video, the CPU jumps to 4.5Ghz.
Additionally, if I run another game which is more CPU hungry, it clocks to 4.2Ghz, or something different.
So it's as if the EPP value is "choosing" the most efficient max CPU clock speed based on how much load there is and the EPP value (lower=higher clocks, higher=lower clocks).

Am I wrong? Is my observation correct?
For me it makes a lot more sense for the max clocks to be capped on a app by app basis, as one game might need more CPU than another, rendering a video needs less CPU than rendering in 3D, etc.
I would prefer to not "cap" the CPU to 4Ghz, and no matter the load, the CPU will be capped at 4Ghz.
It seems a lot more efficient to have different EPP values and let the CPU choose the max frequency it needs.

But maybe I'm just completely wrong.
Any opinions on this matter?

 

[unclewebb]

EPP value 100, CPU is 3.6Ghz while running the game.

What monitoring software are you using to report CPU speed when in game? The on screen in game MHz data is not that accurate when a CPU is partially loaded or when Speed Shift is enabled. Many monitoring apps choke when Speed Shift EPP is being used.

Turn on the ThrottleStop Log File option when gaming. It does a great job of tracking the CPU multiplier and MHz. Attach a log file from a gaming session.

EPP value is set to 100, max CPU is 3.6Ghz.
Change EPP value to 60, max CPU is 4Ghz.
Change EPP value to 0, max CPU is 5Ghz.

Run a TS Bench - 1 Thread test. ThrottleStop should report the same max MHz whether Speed Shift EPP is set to 0, 60 or 100. Post some ThrottleStop screenshots so I do not have to ask 101 questions.

If the low power C states are enabled, there is no need to micro manage the EPP settings or MHz. Intel CPUs do a great job managing themselves when they are setup correctly.

 

[a_bolog]

What monitoring software are you using to report CPU speed when in game? The on screen in game MHz data is not that accurate when a CPU is partially loaded or when Speed Shift is enabled. Many monitoring apps choke when Speed Shift EPP is being used.

Turn on the ThrottleStop Log File option when gaming. It does a great job of tracking the CPU multiplier and MHz. Attach a log file from a gaming session.


Run a TS Bench - 1 Thread test. ThrottleStop should report the same max MHz whether Speed Shift EPP is set to 0, 60 or 100. Post some ThrottleStop screenshots so I do not have to ask 101 questions.

If the low power C states are enabled, there is no need to micro manage the EPP settings or MHz. Intel CPUs do a great job managing themselves when they are setup correctly.

Monitoring with HWinfo and default TS Clock Mhz display.

Running single thread test, the max Mhz and the final score does change when changing the EPP value:
EPP 180 = 3Ghz
EPP 120 = 4.3 Ghz
EPP 70 = 5 Ghz

I've uploaded the log and screenshots here: https://mab.to/t/a1XQsN31Mi6/eu1

 

[unclewebb]

Monitoring with HWinfo

HWiNFO does not accurately track the CPU multiplier during some situations. There is no need to include HWiNFO in any of your screenshots.

I tried using your download site but I could not figure it out. Maybe it doesn't like my popup blocker or who knows why. I am not interested in registering to download files.

Attach your image files and your ThrottleStop log files directly to this forum like everyone else does. Not sure if this forum will allow you to upload your .mp4 video file.

 

[a_bolog]

I've done a screen capture and couldn't upload it here.
Here's with another website, it doesn't ask to register, I think the pop-up is to allow multiple-file download: https://we.tl/t-DYCnapAW0g

 

[unclewebb]

When taking screenshots of TS, can you show me the speed of the CPU while it is half way through the test and there is still a load on the CPU? The CPU speed after the test is over and is mostly idle is not as important compared to when the CPU is under load.

In the TPL window is Speed Shift checked and are Speed Shift Min and Max set to the recommended values? I would set Max to 50 and Min to 8.

I think you need to check the FIVR Overclock box if you want to run the 50 multiplier when all cores are active.

Are you using any manufacturer's CPU control software on your computer? Perhaps some software that is designed to control the Windows power plans or the fans.

When I run the TS Bench - 1 Thread test on my desktop 10850K, I get the same CPU speed whether EPP is set to 0, 70, 120 or 180. The CPU slows down when it is idle if I am using any EPP value greater than 0. The 1 Thread load is more than enough to get the CPU up to full speed when using EPP. I am not sure why your computer is acting different.

 

[a_bolog]

I'm not using any software, I've specifically re-installed windows for that, so I can control the CPU with ThrottleStop.
On another note, even if I don't use TS, and change the EPP value through the default Windows Power Profile, it behaves exactly the same. (attached screenshot)

I think the video I recorded shows exactly how the CPU speed changes based on the EPP value.
The screenshots were to illustrate that even the score changes based on the EPP value.
And the log shows the exact multiplier/frequency it runs at.

I'm even more confused now on how it should behave.
Let me know if you want me to check other tests.

 

[unclewebb]

how it should behave

I am not sure how it should behave. I just know how it behaves on my desktop computer. I am using an older version of Windows 10, 20H2. It is possible that Windows changed what Speed Shift values are being sent to the CPU. If ThrottleStop is only changing these values once per second and Windows is busy changing these values 100 times per second, Windows will win the tug of war over the CPU.

Did you check the Speed Shift box in the TPL window for the above test?

Do you get the same 30 multiplier when plugged in?

If the Speed Shift EPP box is not checked and you change to the Windows Balanced power plan, what EPP value does ThrottleStop report in the FIVR monitoring table? Switch back and forth a couple of times and watch for any EPP changes. Does the Balanced plan give you the same 30 multiplier during a 1 Thread TS Bench test?

Without access to any recent mobile hardware, there is no way for me to investigate what Windows is doing.

 

[a_bolog]

Did you check the Speed Shift box in the TPL window for the above test?

Yes, I also disabled it and I don't notice any effect.

Do you get the same 30 multiplier when plugged in?

I tried both plugged and unplugged, it's exactly the same clock multiplier. The EPP value in FIVR window stays the same

If the Speed Shift EPP box is not checked and you change to the Windows Balanced power plan, what EPP value does ThrottleStop report in the FIVR monitoring table?

The EPP changes to 84, the Profile shows 33%, and clocks to 5Ghz

And If I change the percentage to 80 for example, the EPP value is 204, and clocks 2.4Ghz, so it's exactly the same behavior as if I would change the EPP through ThrottleStop.
Also, if I enable the SpeedShift box in TS while on Balanced, they conflict with one another, and I see the EPP value changing up and down.

And here is the High Performance Profile, it has the energy policy set to 0%, and EPP to 0, and clocks 5Ghz as well. The only big difference is when running the benchmark with more threads, the clocks and scores are higher in High Performance(0%, EPP 0) when compared with Balanced (33%, EPP 84). But on single thread, they clock the same 5Ghz.

 

[unclewebb]

EPP value is 204, and clocks 2.4Ghz

Your mobile CPU works a little differently compared to my desktop CPU. I still get the full 5.0 GHz during a single thread test when EPP is set to 204. It makes sense that a mobile CPU is going to be tuned more for efficiency and a desktop CPU is tuned more for maximum performance.

Did you run any scripts to add, change or uncover the hidden settings within your Windows power plans? There might be some hidden Windows setting that is slightly different in your Windows install compared to the same setting on my desktop computer.

I never recommend setting EPP higher than 84 because doing this can interfere with maximum performance. I do not bother checking the Speed Shift EPP box in ThrottleStop. Windows can manage EPP good enough for my purposes. The Windows High Performance power plan sets EPP to 0, Balanced to 84 and Power Saver sets EPP to 153 on my desktop computer.

On some CPUs, anytime you try to use a turbo ratio higher than the default values listed in ThrottleStop, you have to check the Overclock box in the FIVR window.

To see how your CPU works, run a TS Bench - 8 Thread test. Do you see the 48 multiplier during this test or is the CPU limited to the 42 multiplier? You might need to check the Overclock box to go higher than the default 42 multiplier when all 8 cores are active.

Now that you understand how EPP works on your CPU is everything OK? Any more questions? It looks like EPP is working as Intel intended it to work.

 

[a_bolog]

Yea my question was if it's a good way of configuring the profiles with the EPP value only?
So basically having all profiles on max clocks, max power limits, and only the EPP value different, and letting SpeedShift choose the clock speed for every app.

Thanks again for all the help.