iccMax, Speed Shift and possibly other functions not working with Intel 11th gen (11800H)

[tlac]

I am not sure if there are two separate issues or only one issue.

For some users, running TS 9.3 can instantly fix the less than 100% C0% problem. TS 9.4 and newer makes fewer automatic changes to fix throttling problems. I have no idea why TS 9.3 has this magic ability while newer versions do not, even when both versions are setup exactly the same. Without hardware with this unusual throttling issue, there is no way for me to find out what is causing it.

I have the same issue with Antimalware Service Executable on Windows 10. ThrottleStop 9.3 is the last version that properly throttles the MS Defender as well.
If you need some logs, I can provide them.

9.3 vs 9.3.1

 

[unclewebb]

Good timing. I finally found a way to recreate this problem on my desktop 10850K.

Windows Defender is enabling some monitoring timers within the CPU but only when Real-time protection is enabled. If you disable Real-time protection, these timers within the CPU are left alone and there is no more throttling.

HWiNFO might use one of these timers to report the Effective Clock speed. Defender sets the timer one way while HWiNFO changes it to try and get accurate monitoring data. This is like two people trying to use one stopwatch at the same time to monitor two different events. The results will not be accurate and will be unpredictable. Shared monitoring timers is a real problem for Intel CPUs. It seems to be impossible for software to get exclusive access to these timers.

Sometime between TS 9.3 and TS 9.4, I made a change so ThrottleStop would use a different set of timers. I was forced to do this to avoid any conflicts with another popular monitoring program. Not sure which one at the moment.

Now that I know exactly what the problem is, hopefully I can add some code to ThrottleStop to reset the timer that Windows Defender is using / misusing. My 10850K sees a boost in Cinebench R23 of about 900 points when the timers are not being abused by Windows Defender. It could also be a bug at the hardware level and Defender trying to use these timers just happens to get the blame for it.

Hopefully I have time later this week so I can release a new beta version of TS with this fix. I think you can sign up on the download page to be notified when the next version is ready for download.

 

[tlac]

Thank you for your explanation but there are still some parts that I don't understand.
I get the same bad result without starting the HwInfo at all (after a fresh restart). So it looks the Defender can cause some problems in itself for me. Or is there some other hidden monitoring programs that conflicts with the timers like task manager or amd control center?

What steps you did to reproduce this behavior?

 

[unclewebb]

I get the same bad result without starting the HWiNFO at all

This problem is not a HWiNFO issue. HWiNFO is more like a victim.

Windows Defender is taking full control of 7 out of 7 monitoring timers leaving nothing left over for any other monitoring programs. This is why HWiNFO reports a reduced Effective Clock speed. Windows Defender is constantly reprogramming and trying to wrestle control of one of the monitoring timers that HWiNFO is trying to use to report Effective Clock.

What steps you did to reproduce this behavior?

I used RW Everything to monitor the status of each of the 7 timers. When Windows Defender Real-time Protection is disabled, none of the 7 timers are used. As soon as Real-time protection is enabled, Windows Defender takes control of all 7 timers. If RW Everything tries to use one of these timers, suddenly all of the timers are reset and Windows Defender decides that it no longer needs to use any of them anymore. Usually every 10 minutes, Windows Defender will try to take over the timers again. If this is allowed to happen, Cinebench performance will drop. Stop Windows Defender from using these timers and performance increases. This pattern is all very repeatable.

If ThrottleStop decides that it wants to use one of these timers then Windows Defender decides that it really does not need to use any of them after all. ThrottleStop 9.3 used to use one of these timers so that is why this version was always able to fix this problem. After TS 9.3, I stopped using one of these timers. This gave Windows Defender the opportunity to become a bully. It decided that it needs all 7 timers when no user software should be that greedy with shared system resources.

 

[tlac]

Thanks again for your detailed answer.
I'm still thinking on one thing.
How the Intel 12th gen is not affected? Can you check the timers on those platform? Are they used by the Defender?

 

[unclewebb]

The monitoring timers are common to all Intel CPUs for the last 15+ years. I do not have access to any 12th Gen hardware so I cannot do any testing.

Hopefully when this new TS version is released, users can do some testing and share their results. I am not sure what percentage of Intel CPUs have this problem. On my 10850K, there are some timer issues after resuming from sleep. More testing needs to be done.

 

[tlac]

I have some 4th Gen i3, i5 but no problem on those old platforms.

 

[unclewebb]

I have only heard of this issue on 8th Gen and newer so far. I will check my 4th Gen laptop someday to see how Defender is using the timers.

 

[samot]

Yup, have an older laptop with a i7 4790K and there´s no issue at all.

 

[tlac]

This could be relevant why the Defender plays with MSR:

Finding and Exploiting CPU Features using MSR Templating (pdf)
msrevelio (github)

 

[TheWolfLoki]

Amazing you were able to diagnose this @unclewebb! I really wish there was a fix for defender so Realtime protection could remain on without breaking clockspeed, I wonder how many HxC OC'ers would have benefitted from this over the past 5 years haha

 

[unclewebb]

Real-time protection still works. The gpedit mod only disables real-time notifications. This issue has been around for a long time.

I wrote a separate utility to show this specific problem. It and TS 9.5 should be available some day soon.