The CPU is boosting incorrectly.

[Islam Ghunym]

The title of this topic is the best way to describe the problem.

 

I am running a QTJ1 (coffe lake I9 ES) 8 P-cores no E-cores on a Gigabyte H310 Motherboard which allows for Overclocking.

Stable OC on this motherboard was:

5.1 (1 core) 5.0 (2-3 cores) 4.9 (4-5 cores) 4.8 (6 cores) 4.7 (7 cores) 4.6 (8 cores), 4.6 ring/llc, vcore offset +0.1v, PL1 45W, PL2 56.25W and the CPU can reach max single core performance on any workload at these settings, DDR4-3300 (limited by the ES IMC), vddr 1.4v, tCL 16, tRP/RCD 22, tRAS 36, CR 1, tRFC 487, tFAW 16 tCWL 11.... etc

 

I was on Windows 11 22H2 (speed shift was disabled), and CPU cores was clocking fine. Each core had a different clock speed, and during single core benchmarks 1 core could go 4.9-5.1 while others are clocking lower. at other multi core benchmarks the all core ratio goes about 3.2 GHz AVX2 limited by the PL1. My system was stable for about a year.

 

Recently I reinstalled Windows 11 (23H2 version this time), and here I noticed my system being not fully stable. I checked  what is going on, and noticed that the CPU is doing all core boost 5.1 GHz which is wild, and all cores clock speeds are tied together no matter what. My BIOS OC settings is completely the same as before. All core boost monitored in bios 4600 MHz which what should also be in windows and not 5100! like what it is right now.

 

I tried updating drivers, but that didn't help. I don't really want to reinstall windows again and keep troubleshooting forever so I hope that someone here knows what controls the CPU boosting behaviour on Windows (driver or whatever) so that I can try reinstall it, or try playing around it. I tried installing easy tune (my mobo OC software) to play with core ratios a bit, but that didn't help. I also tried Intel XTU, but that also didn't help. I can make the system stable if I set all cores to 4.6 GHz, but that drops single core performance by quite a margin. I want the smart boosting of the CPU

[leclod]

Which mode are you in in power management ? (I've got Normal)

Some relevant Bios pictures might be useful.

"I checked  what is going on" How ?

Edited January 20 by leclod

[ki8aras]

to be fair, you are using an ES cpu and complaining it's not working correctly

 

[Needfuldoer]

4 hours ago, Islam Ghunym said:

I am running a QTJ1 (coffe lake I9 ES) 8 P-cores no E-cores on a Gigabyte H310 Motherboard which allows for Overclocking.

Motherboards with entry level chipsets are generally designed to a price point, and one of the first places they cut corners value engineer is CPU power delivery. 

 

What happens if you set the BIOS back to defaults?

[Islam Ghunym]

5 hours ago, leclod said:

Which mode are you in in power management ? (I've got Normal)

Some relevant Bios pictures might be useful.

"I checked  what is going on" How ?

The only thing that changed is the Windows. BIOS settings were my all time saved profile the same. I use many softwares that reports and monitor clock speeds. Do you want me to count them? It doesn't matter how.

[Islam Ghunym]

4 hours ago, ki8aras said:

to be fair, you are using an ES cpu and complaining it's not working correctly

 

Yes!, I got an Engineering Sample and I expect it to work correctly. I didn't buy a chip that doesn't work correctly... The chip was working correctly from years. I knew what I was buying when I did so. If I am having a problem now, it doesn't mean because I have an engineering sample. It is too late to label the issue this way.

[Islam Ghunym]

56 minutes ago, Needfuldoer said:

Motherboards with entry level chipsets are generally designed to a price point, and one of the first places they cut corners value engineer is CPU power delivery. 

 

What happens if you set the BIOS back to defaults?

PL1 is only 45W. there is no problem to mention here

[unclewebb]

7 hours ago, Islam Ghunym said:

My BIOS OC settings is completely the same as before.

Your CPU supports Speed Shift Technology. Windows 11 23H2 seems to assume that if Speed Shift is available then the BIOS must have enabled it whether it is enabled or not. Try enabling Speed Shift in the BIOS and see if that makes any difference. 

 

I would disable Windows 11 core isolation memory integrity and anything else VBS related. These features prevent software, including Windows 11, from reading information from your CPU. This might be why your CPU is not being setup correctly. If Windows 11 cannot read the turbo ratios that the BIOS has set then it might just set all turbo ratios to 51. 

 

https://beebom.com/how-disable-virtualization-based-security-vbs-windows-11/

 

After VBS is disabled, reboot and try running,

ThrottleStop 9.6

https://www.techpowerup.com/download/techpowerup-throttlestop/

 

Post screenshots of the FIVR and TPL windows. The FIVR window will likely show that the staggered turbo ratios you set in the BIOS are not being used. If VBS is disabled, you should be able to use ThrottleStop to set the turbo ratios however you like. I am not sure if you can also do the same thing with Intel XTU. I do not know if XTU will properly support your ES processor. ThrottleStop should work OK. 

[Islam Ghunym]

2 hours ago, unclewebb said:

Your CPU supports Speed Shift Technology. Windows 11 23H2 seems to assume that if Speed Shift is available then the BIOS must have enabled it whether it is enabled or not. Try enabling Speed Shift in the BIOS and see if that makes any difference. 

 

I would disable Windows 11 core isolation memory integrity and anything else VBS related. These features prevent software, including Windows 11, from reading information from your CPU. This might be why your CPU is not being setup correctly. If Windows 11 cannot read the turbo ratios that the BIOS has set then it might just set all turbo ratios to 51. 

 

https://beebom.com/how-disable-virtualization-based-security-vbs-windows-11/

 

After VBS is disabled, reboot and try running,

ThrottleStop 9.6

https://www.techpowerup.com/download/techpowerup-throttlestop/

 

Post screenshots of the FIVR and TPL windows. The FIVR window will likely show that the staggered turbo ratios you set in the BIOS are not being used. If VBS is disabled, you should be able to use ThrottleStop to set the turbo ratios however you like. I am not sure if you can also do the same thing with Intel XTU. I do not know if XTU will properly support your ES processor. ThrottleStop should work OK. 

I am on it, thx for the info. However I have disabled VBS and all these security stuff after installing windows immediately. Windows also seems to be able to see my bios settings (core ratio settings), but as you mentioned there is some weird behaviour regarding speed shift. I tried enabling it before and also disabling it, but it didn't change the clock behaviour as it should. The CPU was behaving the same. I will try to set speed shift value manually from throttlestop and see what happens. Just to mention I am ok with all kind of power saving settings (all set to on with me), but not the speed shift as it hurts system latency. I also don't enable ring down bin as I set uncore more than all core ratio -3, but I will see what happens and report here

[unclewebb]

1 hour ago, Islam Ghunym said:

not the speed shift as it hurts system latency

When Speed Shift is enabled, use the Windows High Performance power plan. This automatically sets Speed Shift EPP to 0. This tells the CPU to run at full speed regardless of load. There should be no additional latency when you do this. The Balanced power plan will slow the CPU down when lightly loaded. This can increase latency. 

 

It sounds like the turbo ratios in the FIVR window are causing the problem. Try pushing the Defaults button on the left bottom side of the FIVR window. This will try to read the default turbo ratios from the CPU. This should work with an ES processor.

[Mark Kaine]

9 hours ago, Islam Ghunym said:

Yes!, I got an Engineering Sample and I expect it to work correctly. I didn't buy a chip that doesn't work correctly... The chip was working correctly from years. I knew what I was buying when I did so. If I am having a problem now, it doesn't mean because I have an engineering sample. It is too late to label the issue this way.

easy fix: install previous windows version.  you cant expect software being optimized for "engineering samples", especially not forever.

 

(i think rufus can do this, good luck )

 

 

[Islam Ghunym]

17 hours ago, Mark Kaine said:

easy fix: install previous windows version.  you cant expect software being optimized for "engineering samples", especially not forever.

 

(i think rufus can do this, good luck )

 

 

That was a verified possible solution, but I installed the 23H2 again and I am re-doing the core OC. I must share here what I found so it can hopefully be useful.

 

What caused the instability with the newer OS isn't a compatibility matter or being an engineering sample, but rather an updated way of how windows uses the CPU and the proper thing to blame here is my OC. the newer OS 23H2 could give lower CPU latency and higher memory bandwidth on Aida64 (about 1-7% improvements in each field) and even better single core performance (1-2% out of margin of error). being limited to 45W means the CPU won't boost enough to show off multi core instability. I used that to my advantage on older versions of windows to get higher performance on lighter workloads while staying stable on heavier workloads due to power limit. For why I am not increasing the power limit to 120W which should be fine for my motherboard is that this PC is running on battery power so efficiency matters a lot. The newer OS allowed the CPU to do better and boost better on the same power limit which in it's turn caused instability.

 

This is the whole story I believe. It took some real time to verify and come up with all of that so I think I should share it to fill curiosities.

 

Now I am just re-doing the core OC

[Mark Kaine]

ah, that's interesting,  glad i could help a little. 

[Bagzie]

10 hours ago, Mark Kaine said:

ah, that's interesting,  glad i could help a little. 

Not very nice qouting your solution and then giving himself the solution tick but atleast it's fixed.

[Islam Ghunym]

11 hours ago, Bagzie said:

Not very nice qouting your solution and then giving himself the solution tick but atleast it's fixed.

Sorry about that, but re-doing the OC was a better solution so it makes sense to mark my post as the solution and quoting his one as it was a possible solution.