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

[Specht77]

Hello, I started using ThrottleStop at around 2012 when my Asus k43e would get stuck at 800mhz

Since then I used ThrottleStop in basically every system I had

My last computer had a 10750H and I tried to make a preset with most power saving possible settings I found, In that preset I set the iccMax to a pretty low number and Speed Shift to a pretty high number, since I don't usually do any intensive tasks while on battery

I recently bought a new laptop, that has a 11800H, I installed ThrottleStop 9.4.2 and tried making a similar preset for power savings in battery, I was able to disable turbo and to do undervolt (I modified the bios to be able to undervolt), but for some reason the SpeedShift option and the iccMax doens't seem to do anything, I tried several values and didn't see any difference

I accept suggestions on what to do in this situation

Edit: I'm using Windows 11, don't know if that has something to do with the issue

 

[samot]

Hi,

I have a similar question.
I´ve noticed that on my 11800h the Icc Max option seems funky: if i apply even a small change to the slider, it seems that the effect of that is of setting it to the maximum value, and then reseting it to the original value (105 in my case) doesn´t do anything.

Example: on battery (which disregards any performance mode selection on my laptop´s control center), running CB23, my 11800h is capped at 20w. With a -80mv undervolt this translates to a maximum multiplier of 19x. If i touch the Icc Max slider, suddenly the cpu is able to reach 35w and a 26x multiplier. After this i can set the IccMax to any value, including 1, but without any effect.

Adding to the oddness: plugged in, selecting the lower performance mode on my laptop´s control center caps the cpu wattage to 35w, which then running CB23 results as expected in a maximum wattage of 35w and a maximum multiplier of 20x. Wtf?
So, it would seem logical that i´d like it to reach a steady 26x multiplier as when on battery. It turns out that messing with the Icc Max slider in this situation accomplishes nothing!

@unclewebb , do you have any clues? (ended up getting here after NBR died)

 

[unclewebb]

Post some screenshots if you see something unusual. I might see something in ThrottleStop that is not setup correctly. Include a screenshot of Limit Reasons if your CPU is throttling and running at less than its rated speed.

do you have any clues?

I do not own or have access to any similar 11th Gen mobile CPUs for development or testing purposes. That means I have never sat down and used ThrottleStop on an 11800H. How the IccMax slider works or whether it works on an 11800H is a mystery to me. Make sure you remember to adjust IccMax for both the CPU core and the CPU cache.

laptop´s control center

If you are running ThrottleStop along with a laptop's control software then there is likely going to be some interference. There is no way to predict what program will be in charge of your CPU at any moment in time.

for some reason the Speed Shift option

The Speed Shift values are suggestions to the CPU. ThrottleStop is still writing the same Speed Shift values to the same Speed Shift registers as it always has. This has worked correctly since the 6th Gen CPUs introduced Speed Shift Technology. It seems like some of the newer processors can ignore some of the Speed Shift suggestions that ThrottleStop is sending to it. Without hardware, I cannot test what is going on or make any changes to ThrottleStop to try and improve things.

 

[samot]

Hi unclewebb, long time overdue reply....

Post some screenshots if you see something unusual. I might see something in ThrottleStop that is not setup correctly. Include a screenshot of Limit Reasons if your CPU is throttling and running at less than its rated speed.


I do not own or have access to any similar 11th Gen mobile CPUs for development or testing purposes. That means I have never sat down and used ThrottleStop on an 11800H. How the IccMax slider works or whether it works on an 11800H is a mystery to me. Make sure you remember to adjust IccMax for both the CPU core and the CPU cache.


If you are running ThrottleStop along with a laptop's control software then there is likely going to be some interference. There is no way to predict what program will be in charge of your CPU at any moment in time.

There isn´t any kind of throttling happening other than "EDP Other" when running on battery. That´s to be expected. I´ve found out that on battery the laptop has a current limit of ~16A. I suppose it´s so to not over stress the battery?

Regarding your 3rd point: if i don´t have the laptop´s control software installed(which is handy, fan curves, etc..), even TS is unable to set power limits above 75w, but lower is fine.

Two more things i found out in the meanwhile:
- without TS, running a benchmark (10min CB23 runs, for example) and let´s say with PL2=40w and PL1=35w, i have higher clock speeds when PL1 kicks in (2350mhz->2700mhz). But, if i set PL2=35w PL1=35w, after some time the cpu clocks get lower (2600mhz->2300mhz). Wtf?
If, on the other hand, i use TS and unclamp PL1, when PL2=40w PL1=35w the resulting clocks are different (2800mhz->2700mhz), and when PL2=35w PL1=35w there is no change in clocks (2600mhz).
Seems something is wrong with my Bios/Ec, right? I´ve already reflashed it, btw.

- also, this has nothing to do with TS, but maybe you can point me in the right direction - while doing 10min CB23 runs i noticed that after some time the cpu clocks would drop by 60/70mhz. This would drop my scores from ~14150 to ~13750. Looking at TS main window C0% values, before the drop they read 100% on all cores, but after the drop they read ~99.5%.
Looking up Windows 11´s event viewer i found out that this cpu clock drop happens precisely when a Windows Defender related service is installed (MpKslDrv.sys), always ~10 minutes after boot. Picked up another laptop which has Windows 10 installed, and the same event happens but no clock speed drop occurs. Did a separate Win11 install on my "faulty" laptop, in case something is wonky on my original install, but this behaviour persists. Maybe it has something to do with Windows Defender on Win11? I´ve search the web to see if anyone has the same complaint, but haven´t found anything.

 

[unclewebb]

"EDP Other" when running on battery

That sounds normal. Lots of manufacturers do what they can to avoid any battery fires. This type of throttling might be somewhat excessive but at least there is a legit reason for it.

even TS is unable to set power limits above 75w

Some laptop manufacturer's software likely has access to the EC power limit that ThrottleStop does not have access to. In this situation, if you need to run beyond 75W then I guess you have to run their software to do it.

The power limit Clamp options determine whether the CPU will go below its base frequency when it needs to power limit throttle. When Clamp is checked, the CPU will run slower than base frequency to try and keep power consumption right at 35W max. If you do not check Clamp, the CPU should only slow down to base frequency. Power consumption might be higher than your requested 35W when running at base frequency. There does not seem to be anything wrong with your BIOS. Your results depend on how you have the Clamp option set. Most BIOS versions do not give you access to the two Clamp options so if you are not running ThrottleStop, the PL1 Clamp option is usually checked and the PL2 Clamp option is usually clear. If you really want to make sure that power consumption does not exceed 35W then you need to check both Clamp options.

the cpu clocks would drop by 60/70mhz

Post a screenshot of the TPL window so I can see how you have that setup. Run Cinebench again and post a screenshot of the main ThrottleStop window and the Limit Reasons window while your CPU is throttling. A slight drop in C0% when fully loaded usually confirms some sort of throttling. Limit Reasons might show something in red. You should also run a log file to try and catch what is going on. Not exactly sure what is happening but some more data might answer a few questions.

long time overdue reply...

I will be back in April to check for your reply.
Just joking.

 

[samot]

Thanks unclewebb!

The power limit Clamp options determine whether the CPU will go below its base frequency when it needs to power limit throttle. When Clamp is checked, the CPU will run slower than base frequency to try and keep power consumption right at 35W max. If you do not check Clamp, the CPU should only slow down to base frequency. Power consumption might be higher than your requested 35W when running at base frequency. There does not seem to be anything wrong with your BIOS. Your results depend on how you have the Clamp option set. Most BIOS versions do not give you access to the two Clamp options so if you are not running ThrottleStop, the PL1 Clamp option is usually checked and the PL2 Clamp option is usually clear. If you really want to make sure that power consumption does not exceed 35W then you need to check both Clamp options.


Post a screenshot of the TPL window so I can see how you have that setup. Run Cinebench again and post a screenshot of the main ThrottleStop window and the Limit Reasons window while your CPU is throttling. A slight drop in C0% when fully loaded usually confirms some sort of throttling. Limit Reasons might show something in red. You should also run a log file to try and catch what is going on. Not exactly sure what is happening but some more data might answer a few questions.

As you can see, no throttling happening, and it happens exactly when that Windows Defender service is installed. This same service event happens in Windows 10 but there´s no C0% drop.
Sorry for obscuring the HWiNFO graph with the time info box. The clock speed there is a straight 4156mhz line. Also, forgot the log.

Regarding the other weird behaviour where i have to unclamp PL1, here are some pictures (also notice the voltage graph):
Without TS PL1=35w PL2=35w

With TS PL1=35w PL2=35w and PL1 unclamped

Without TS PL1=35w PL2=40w

With TS PL1=35w PL2=40w and PL1 unclamped

TS also doesn´t report higher than 35w when i unclamp PL1.

 

[unclewebb]

forgot the log

I would much rather see a TS log compared to all of those graphs. I like seeing the numbers. Next time you are bored, exit HWiNFO and only run ThrottleStop with the Log File option checked when Cinebench testing. If you notice anything unusual when testing, keep track of what time it happened at.

I would check the MMIO Lock box in the TPL window when testing. The MMIO power limits also have Clamp options. ThrottleStop might not duplicate your Clamp requests to both the MSR and MMIO power limits. I only use the MSR power limits and I get consistent results when using them.

In the FIVR window do you have Thermal Velocity Boost or V-Max Stress checked?

TS also doesn´t report higher than 35w when i unclamp PL1.

Show me a ThrottleStop screenshot of this after you lock the MMIO power limits. The Clamp option only determines whether the CPU will run slower than its base frequency when power limit throttling. If TS and your control center software are both writing different values to the same power control register then the results might be unpredictable.

Edit - I just did some testing of the HWiNFO Core Clocks data. When a CPU is power limit throttling, the graphed data is not even close to the level of accuracy that ThrottleStop provides in the log file. HWiNFO will show +/- 400 MHz swings while ThrottleStop is more in the +/- 10 MHz range.

The Effective Clock data is only useful if the CPU is 100% loaded but even then it still does not track the CPU multiplier as accurately as ThrottleStop does. Avoid using those graphs to determine anything. The HWiNFO power limit data and graphs look OK.

Edit - ThrottleStop Limit Reasons shows that you need to exit HWiNFO when using Limit Reasons. The reason for this is HWiNFO clears the throttling data out of the CPU before ThrottleStop can report it. With HWiNFO off, you might see some boxes turn yellow when stress testing. This is the first sign when you are right on the edge of throttling.

 

[samot]

I would much rather see a TS log compared to all of those graphs. I like seeing the numbers. Next time you are bored, exit HWiNFO and only run ThrottleStop with the Log File option checked when Cinebench testing. If you notice anything unusual when testing, keep track of what time it happened at.

I would check the MMIO Lock box in the TPL window when testing. The MMIO power limits also have Clamp options. ThrottleStop might not duplicate your Clamp requests to both the MSR and MMIO power limits. I only use the MSR power limits and I get consistent results when using them.

In the FIVR window do you have Thermal Velocity Boost or V-Max Stress checked?


Show me a ThrottleStop screenshot of this after you lock the MMIO power limits. The Clamp option only determines whether the CPU will run slower than its base frequency when power limit throttling. If TS and your control center software are both writing different values to the same power control register then the results might be unpredictable.

Thanks again.
Here is the log and the screenshot. This regarding my first issue (C0% not reaching 100% after some time). I´ll do more tests on the second issue (higher clocks when PL1 kicks in) with HWiNFO disabled and with a TS log.
No, TVB and V-Max Stress are unchecked.

No throttling happening. And as before, before that defender service starts C0% is at 100% and i get ~14150 in CB23, after it starts C0% won´t reach 100% and CB23 scores get down to ~13800. This doesn´t happen on Win10, which has the same service event. In both os´s, windows defender is with it´s default settings.

Again, thanks for your time.

 

[unclewebb]

I had a quick look through your log and noticed these big drops in C0% happening about every 56 seconds. At first I thought I was on to something so I ran Cinebench R23 on my computer. I noticed that this happens every time Cinebench finished one image, resets itself, and starts working on the next image. These drops in C0% and power consumption are obviously normal.

   DATE       TIME    MULTI   C0%   CKMOD  BAT_mW  TEMP    VID   POWER
2022-03-13  13:39:20  42.00  100.0  100.0       0   82   1.0325   74.6
2022-03-13  13:39:21  42.00  100.0  100.0       0   82   1.0325   74.6
2022-03-13  13:39:23  42.00  100.0  100.0       0   82   1.0325   74.4
2022-03-13  13:39:23  42.00  100.0  100.0       0   82   1.0325   74.4
2022-03-13  13:39:24  42.85   56.5  100.0       0   66   1.0599   42.8
2022-03-13  13:39:25  42.00   95.9  100.0       0   80   1.0325   65.3
2022-03-13  13:39:26  42.00  100.0  100.0       0   82   1.0325   74.8
2022-03-13  13:39:27  42.00  100.0  100.0       0   82   1.0325   75.0
2022-03-13  13:39:29  42.00  100.0  100.0       0   83   1.0325   75.2
2022-03-13  13:39:30  42.00  100.0  100.0       0   82   1.0299   75.3

The C0% drops in your log file from 100.0 to between 99.0% and 99.5% are definitely not normal.

   DATE       TIME    MULTI   C0%   CKMOD  BAT_mW  TEMP    VID   POWER
2022-03-13  13:44:26  42.00  100.0  100.0       0   85   1.0299   75.7
2022-03-13  13:44:27  42.00  100.0  100.0       0   85   1.0325   75.6
2022-03-13  13:44:28  42.00  100.0  100.0       0   86   1.0299   75.9
2022-03-13  13:44:29  42.00  100.0  100.0       0   85   1.0299   76.1
2022-03-13  13:44:30  42.00   99.6  100.0       0   85   1.0299   75.9
2022-03-13  13:44:31  42.00   99.3  100.0       0   85   1.0299   76.5
2022-03-13  13:44:32  42.00   99.4  100.0       0   85   1.0299   76.3
2022-03-13  13:44:33  42.00   99.2  100.0       0   85   1.0299   76.4
2022-03-13  13:44:34  42.00   99.4  100.0       0   85   1.0299   76.4
2022-03-13  13:44:35  42.00   99.1  100.0       0   86   1.0299   76.5

The CPU multiplier in your log file is steady at 42.00 so that confirms that there is no throttling. The CPU speed is rock steady. HWiNFO reports any drop in C0% as a drop in the Effective Clock. This is just a different way to report that something is not quite right.

When this drop happens, open up the Task Manager and see if you can kill the MpKslDrv.sys service. Does your C0% go back up to 100.0%? It is possible that this service is working at a higher priority and is interrupting Cinebench more compared to this same service in Windows 10. You could try using the Task Manager to increase the priority of Cinebench. That might prevent this from happening.

You have definitely found some sort of odd Windows 11 bug but I do not know how to fix this. Your CPU is not throttling its MHz so I do not think ThrottleStop can do anything to solve this problem.

If you wait 5 or 10 minutes and run Cinebench again, does C0% ever return to 100.0%? I remember a very similar problem about a year ago and once this was triggered, there was no way to get back to 100.0% without rebooting. Performance in Cinebench, etc. remained below normal. Also try doing a sleep resume cycle when this happens.

 

[dnm_TX]

I wonder what is the point to set PL1/PL2 to 120 if(looking at the main window) PKG Power goes up to 75-ish W only?

 

[unclewebb]

Setting the PL1 and PL2 power limits high helps ensure that they will not limit CPU performance. There is nothing wrong with setting these limits higher. It is like a game of limbo. No one ever complains if the bar is set too high. Most desktop boards have an Unlimited feature that sets the two power limits to the max, 4095. You will never have to worry about these limits causing any power limit throttling.

 

[dnm_TX]

True and True....but when it comes to temperature on those "slim and light" laptops and you see it's already at 88 °C with 78 W,i'd personally worry more about that then the power throttling. Just my 2 cents,didn't mean to create argument.

 

[samot]

True and True....but when it comes to temperature on those "slim and light" laptops and you see it's already at 88 °C with 78 W,i'd personally worry more about that then the power throttling. Just my 2 cents,didn't mean to create argument.

That 78w reading isn´t right. I´m using a beta bios from a well known person (i suppose i can´t say more for now). On the stock bios i´m reaching ~100w on the same workload (CB23), same voltages and multipliers. The temps are the same. And i was using auto fans speeds, btw.

The CPU multiplier in your log file is steady at 42.00 so that confirms that there is no throttling. The CPU speed is rock steady. HWiNFO reports any drop in C0% as a drop in the Effective Clock. This is just a different way to report that something is not quite right.

When this drop happens, open up the Task Manager and see if you can kill the MpKslDrv.sys service. Does your C0% go back up to 100.0%? It is possible that this service is working at a higher priority and is interrupting Cinebench more compared to this same service in Windows 10. You could try using the Task Manager to increase the priority of Cinebench. That might prevent this from happening.

You have definitely found some sort of odd Windows 11 bug but I do not know how to fix this. Your CPU is not throttling its MHz so I do not think ThrottleStop can do anything to solve this problem.

If you wait 5 or 10 minutes and run Cinebench again, does C0% ever return to 100.0%? I remember a very similar problem about a year ago and once this was triggered, there was no way to get back to 100.0% without rebooting. Performance in Cinebench, etc. remained below normal. Also try doing a sleep resume cycle when this happens.

LatencyMon names the MpKslDrv.sys process as KSLD, with product description "Microsoft Malware Protection". On the Task Manager i don´t see anything named like that on the Processes, Details, and Services tab.

Only related entries are "Antimalware Service Executable" and "Antimalware Service Executable Content Process". These can´t be killed.

On the Task Manager, when running CB23 before the clock speed drop, the antimalware service is at 0.1/0.2% cpu utilization. After the service install the antimalware service rises to over 1.5% cpu utilization. When idling, before and after the drop, the utilization is the same, 0.1%.

Setting CB to a higher priority still doesn´t make C0% reach 100%, it´s the same as before, although i do get the scores that i should.

Even running TS Bench, C0% doesn´t get to 100%.

I´ve also tried waiting some minutes, as you asked, but no change. The same with a sleep resume cycle.

Only way to "correct" this, other than a reboot, is to disable Defender´s real-time protection. The clocks return to 4200mhz and C0% to 100%.

But, as i said before, with Win10 that same service install happens and there´s no reduction on clock speeds/C0% (and this with Defender´s real-time protection on). On a fresh Win11 install this problem exists. So, yeah, probably a Win11 bug.

Again, thanks!

 

[samot]

The power limit Clamp options determine whether the CPU will go below its base frequency when it needs to power limit throttle. When Clamp is checked, the CPU will run slower than base frequency to try and keep power consumption right at 35W max. If you do not check Clamp, the CPU should only slow down to base frequency. Power consumption might be higher than your requested 35W when running at base frequency. There does not seem to be anything wrong with your BIOS. Your results depend on how you have the Clamp option set. Most BIOS versions do not give you access to the two Clamp options so if you are not running ThrottleStop, the PL1 Clamp option is usually checked and the PL2 Clamp option is usually clear. If you really want to make sure that power consumption does not exceed 35W then you need to check both Clamp options.

I would check the MMIO Lock box in the TPL window when testing. The MMIO power limits also have Clamp options. ThrottleStop might not duplicate your Clamp requests to both the MSR and MMIO power limits. I only use the MSR power limits and I get consistent results when using them.

Show me a ThrottleStop screenshot of this after you lock the MMIO power limits. The Clamp option only determines whether the CPU will run slower than its base frequency when power limit throttling. If TS and your control center software are both writing different values to the same power control register then the results might be unpredictable.

I´ll do more tests on the second issue (higher clocks when PL1 kicks in) with HWiNFO disabled and with a TS log.

Ok, now to this issue.
For example, let´s say i want to limit (which sometimes i do for gaming) the cpu to 35w. 11800h base frequency is 2.3ghz.
So, PL1=PL2=35w, Cinebench R23:
PL1 35w (PL1 clamped): 27/28x PL2 35w: 27x/28x
PL1 35w (PL1 unclamped): 29/30x PL2 35w: 29/30x

I´ll attach the TS logs. They show there´s no difference in reported package power. It´s great to have your creation - it gives me the ability to unclamp PL1, something that isn´t available on my bios, which i´ll request to a certain modder, along with PL controls. Btw, the behaviours i´m describing in this post also happen in the stock bios.

Next, for the really odd part, PL1=35w PL2=40w, again Cinebench R23
PL1 35w (PL1 clamped): 30x/31x PL2 40w: 26x/27x
PL1 35w (PL1 unclamped): 30x/31x PL2 40w: 30x/31x

Nope, didn´t switch things around, both lines are correct, but they definitely don´t seem right to me on various accounts: PL1 vs PL2 resulting clocks, and these same clocks (@35w) when compared with the previous values when PL1=PL2=35w.

 

[unclewebb]

PL1 35w (PL1 clamped): 27/28x PL2 35w: 27x/28x
PL1 35w (PL1 unclamped): 29/30x PL2 35w: 29/30x

Power consumption in Intel CPUs is not measured power consumption. The formula used by the CPU uses the VID requested voltage to try and come up with an approximated power consumption number. This number is then used to control the turbo boost function.

It appears that the VID voltage is on average lower when the CPU is unclamped. Lower VID voltage means lower reported power consumption. This allows the CPU to speed up slightly to get back up to the 35W power limit. That is what seems to be happening but I cannot come up with any explanation as to why this is happening. As long as the CPU is operating above the base frequency, in theory, clamped or not clamped should not make any difference to CPU speed.

PL1 35w (PL1 clamped): 30x/31x PL2 40w: 26x/27x
PL1 35w (PL1 unclamped): 30x/31x PL2 40w: 30x/31x

At the start of both log files when the CPU is being capped to 40W, the CPU is running a good 400 MHz faster when it is unclamped. The VID voltage tends to be more consistent between these two log files so I cannot even use that as an excuse.

Is this increase in MHz confirmed by better Cinebench results when comparing unclamped to clamped?

Later in these log files when the 35W PL1 throttling starts, the Clamped log file shows that the CPU speeds up. The power limit drops from 40W to 35W and now both the clamped and unclamped are running at similar speeds. The power consumption data is a lot less consistent. It almost seems that the CPU might be constantly varying the power limit between 35W and 40W. It tries to go up to 40W, quickly runs out of turbo boost and then goes back down to 35W. It goes back and forth like this whereas before when both PL1 and PL2 were set to the same 35W, power consumption would stay very stable right at or just a hair under 35W. Intel says turbo boost is like water in a bath tub. After you pull the plug and all the water drains out of the turbo boost tub, you have to wait before the turbo boost reserve can be replenished.

You probably cannot even find an engineer deep inside Intel that would be able to explain why your CPU does what it does.

When I originally added Clamp control to ThrottleStop, I had no idea that sometime in the future, disabling Clamp would be a useful way to increase MHz. Hopefully some Cinebench numbers will confirm that the CPU really is speeding up when power limited and unclamped. Is it a bug at the CPU level? Probably. Will Intel ever fix this with a microcode update? Maybe but I doubt it.

 

[samot]

Yeah, the voltages are a lot more stable and lower when PL1≠PL2, specially when PL1 kicks in.

Yes, i definitely have better scores in Cinebench when unclamped in both cases (PL1=PL2 and PL1≠PL2), nothing that could be mistaken for margin of error.
So, what i´ve being doing when gaming and limiting the cpu to a set value, let´s say 40w, is choosing 40w for PL1 and 45w(can be any higher value) for PL2, unclamp PL1 and reducing the turbo time limit to 1s (yeah, one of these would suffice). It´s clear that for better performance, setting PL1=PL2 is a no go.

So, as you say probably a bug at the cpu level. Any chance of it being bios related or a weird communication issue between the laptop´s control center and bios/ec? Since i´m beta testing a bios, i was hoping this was something that could be corrected. Already passed on these findings but haven´t heard anything back yet.

Someday i´ll test this with the laptop´s control center uninstalled, and control PL1/2 with just TS.

Thanks unclewebb, i think that´s all for now!

 

[samot]

Hi @unclewebb,

So, regarding that issue of the cpu never reaching 100% C0% ~10 minutes after boot, when a Defender related service starts:
- restarting the "Antimalware Service Executable" solves the issue. Since my system (Win11 Pro) doesn´t allow me to do that, i have to use a program called Process Hacker (it´s quite similar to another one called Process Explorer). I´ve also read some people were able to tame some Defender issues by deleting the mpenginedb.* files located in \ProgramData\Microsoft\Windows Defender\Scans
- after that i spent a day playing around with a fresh Win11 install, initally blocking any kind of updates, and there were no issues at all. When i finally let the system do its updates the problem appeared again.
- on the next day i did another fresh install and even with all updates the problem didn´t return....i didn´t take note on which one, but this time there was one update less.

 

[unclewebb]

regarding that issue of the cpu never reaching 100% C0% ~10 minutes after boot

Thanks for sharing how you got to the bottom of this issue. Some people have noticed a drop in Cinebench performance after the first 10 minutes but they have a hard time figuring out why. I was starting to wonder if this was a bug at the hardware level. The majority of people use Windows Defender so it makes sense that this bug is related to a program that is running on almost everyone's computer.

 

[samot]

These are the updates that cause the issue:
KB2267602
KB5011493

 

[TheWolfLoki]

I have this same issue with my 10700k and Asus z590-e on Windows 11
I've read through UncleWebb's threads on OC.net and LTT Forum.

C0% maxes at ~99% or so after restart. Reducing effective clocks from 5,000Mhz to ~4905Mhz

Running Throttlestop 9.3 immediately fixes this for me, even after closing it.

Wondering if anyone has found a solution other than running throttlestop at boot, I am happy to see this fix for it but it's kind of a band-aid over an unknown root cause.

I checked for both of these updates (KB2267602, KB5011493), and while they show as installed under update history, I cannot uninstall them, even using powershell doesn't show them and can't force uninstall to test if removal stops the ghost-throttling we're seeing.

 

[samot]

Yeah, you´d have to do a clean install and block those specific updates. The thing is, regarding KB2267602 which are the defender security updates, you´d have to completely disable those using the group policy editor. But that´s not a good idea if you plan on sticking with Windows Defender.

While sorting this out i didn´t try TS 9.3... TS 9.4.2, which is what i´ve been using never fixed the issue.

Try WUMT (Windows Update Mini Tool) and see if you can uninstall KB5011493, KB2267602 can´t be uninstalled.

There´s still something weird going on with KB2267602. After sorting out the issue (and with the KB2267602 updates being done automatically), after a full week without the issue it came back randomly, and after another week it seems to have gone away. I´ve already complained to Microsoft but haven´t had any feedback.

Even weirder is that on two older laptops (one with a 2nd gen i5, and the other with a desktop i7 4790k) this doesn´t happen.

 

[unclewebb]

While sorting this out i didn´t try TS 9.3...

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.

 

[rethcirE]

Possibly not the solution for everyone but I used Mirkec's 'Remove Defender' in Win 10 to completely uninstall/abolish Defender (Caution: Irreversible!). - https://teamos-hkrg.com/threads/remove-windows-defender-version-4-1-teamos.72189/
Also - https://teamos-hkrg.com/threads/windows-10-toolkit-version-1-0-teamos.118134/

I previously got the downloads from the now defunct NBR forum so you may have to register there to use their download links. Works a treat though, bye bye Defender and related services/updates.

 

[TheWolfLoki]

I would prefer to keep Windows Defender intact, running TS 9.3 at boot is better for me than deleting Defender.

Just such an odd issue, WUMT cannot see either update as uninstall-able either even though they are in the "History", same behavior as powershell.
Maybe because I used a windows 11 image that is very recent and already had basic versions of those installed... IDK grasping at straws lol

Even weirder is that on two older laptops (one with a 2nd gen i5, and the other with a desktop i7 4790k) this doesn´t happen.

Do either of these pcs use Win 11?

 

[samot]

Do either of these pcs use Win 11?

Nope, Windows 10. But on my actual laptop with a 11800h this happens on both Win10 and 11.