CPU downclocks after restarting from sleep mode

[MicrobeMossberg]

Dear all,

I've noticed lately that while using ThrottleStop, the CPU clock speeds tend to downclock after a restarting from sleep/hibernation.

I've clocked my CPU 5.0 GHz on P and 3.6 GHz on E.

As you can see, after the restart, the downclock sets in (only in the P cores as you can see). In the worst case, it caps all cores at 2 GHz and there is no way for me to reset the clock speeds to the way it was before, except through a reboot, which is tedious unfortunately.

Any ideas on how to prevent this from happening?

Thanks!

 

[unclewebb]

while using ThrottleStop, the CPU clock speeds tend to downclock after a restarting from sleep/hibernation

Does this problem only happen when using ThrottleStop?

As you can see

Your HWMonitor screenshot does not show me anything because most of the P cores are running at 4.8 GHz.

Post some screenshots of ThrottleStop that show the problem you are having. I need to see ThrottleStop screenshots while your CPU is stuck at a low speed. Run something simple like a few threads of the TS Bench test to put a load on your CPU. While this test is running, show me the main screen of ThrottleStop, the FIVR window, the TPL window and the Limit Reasons window. That information might provide me with some clues so I can make some suggestions.

Either ThrottleStop is not setup correctly or not working correctly or the BIOS is not setting up your CPU correctly after a sleep resume cycle. These kind of problems are common. Are you using the most recent BIOS version for your motherboard? Is SpeedStep or Speed Shift enabled or disabled in the BIOS? Some users like to disable things in the BIOS without fully understanding why some settings should be left enabled.

 

[MicrobeMossberg]

Does this problem only happen when using ThrottleStop?


Your HWMonitor screenshot does not show me anything because most of the P cores are running at 4.8 GHz.

Post some screenshots of ThrottleStop that show the problem you are having. I need to see ThrottleStop screenshots while your CPU is stuck at a low speed. Run something simple like a few threads of the TS Bench test to put a load on your CPU. While this test is running, show me the main screen of ThrottleStop, the FIVR window, the TPL window and the Limit Reasons window. That information might provide me with some clues so I can make some suggestions.

Either ThrottleStop is not setup correctly or not working correctly or the BIOS is not setting up your CPU correctly after a sleep resume cycle. These kind of problems are common. Are you using the most recent BIOS version for your motherboard? Is SpeedStep or Speed Shift enabled or disabled in the BIOS? Some users like to disable things in the BIOS without fully understanding why some settings should be left enabled.

Hi Unclewebb, thanks for your quick reply.

Indeed, the downclocking only occurs when using ThrottleStop.

As you can see, the Turbo Ratios are set at 50x. With a fresh restart, they do register and this is also observed by either TS or HWmonitor.

After reopening the laptop after it has been through battery mode + hibernation, the clock speeds cap at factory clocks (4.8P and 3.4E) or worse, at 2 GHz all core which is the base clock speed (that's a bit more rare). Note the discrepancy where the P is downclocked but the E cores still run at 3.6 GHz.

Just ran a 960M TS Bench

The TS UI during bench:

I have the latest BIOS version, just confirmed that yesterday. Not sure about the SpeedShift option in the BIOS, but will recheck in case I missed it.

Thanks!

Just to confirm, there is no SpeedShift/SpeedStep option in the BIOS. I am assuming this is enabled as TS is able to manipulate these

 

[unclewebb]

To try and troubleshoot this problem, I would start by using only a single profile in ThrottleStop. Open the Options window and clear the two check boxes in the top right corner. Clear the AC Profile and Battery Profile check boxes. You can try using these later after you figure out what is going on with the reduced CPU speed during sleep resume.

It is possible that your BIOS is not setting the Speed Shift Min and Max values correctly after a sleep resume cycle. Try checking the Speed Shift box in the TPL window. Set the Speed Shift Min and Max values appropriately. If 50 is your maximum turbo ratio then set Speed Shift Max to 50. Checking this box tells ThrottleStop to send the Speed Shift Min and Max values to the CPU after a sleep resume cycle. I always check the Speed Shift box just to make sure.

Your FIVR screenshot shows that Speed Shift EPP is set to 128. Try switching to a different Windows power plan that sets this to a lower value. There should be a slider available in the system tray to adjust this. Most computers running Windows 10 used to set this to 84. An EPP value of 128 can reduce maximum performance when the CPU is fully loaded.

The ThrottleStop screenshot you posted above shows the CPU at 97°C and it also shows HOT in red. That means thermal throttling is in progress. In that situation, your CPU is running slower than its full rated speed for a reason. The CPU is too hot.

After you get Speed Shift checked, do a sleep resume cycle and run a more leisurely TS Bench load like a 4 or 8 Thread test. Show me a picture of the main screen of ThrottleStop while that test is running. Open the Limit Reasons window so I can also see that info in your screenshot.

In the TPL window I would set Power Limit 4 to a value of 0. This can help with some EDP OTHER throttling issues.

Your cooling is good for about 135W so having the long term MSR PL1 power limit set to 135W makes sense. Having MSR PL2 set to 215W seems like more heat than your cooling system can manage. The turbo time limit is set to 12288 seconds which is more than three hours of turbo boost at up to 215W. Your cooling system can most definitely not handle that. Most laptop cooling systems need a much more conservative time limit value. Somewhere between 8 and 28 seconds at the higher PL2 power limit makes sense.

VR CURRENT lighting up red is not a good sign. The voltage regulators might not be able to fully power your CPU when it is overclocked. Watch for this type of throttling to see if it happens after first booting up or if it happens after a sleep resume cycle. Some voltage regulators might start to perform poorly if they get too hot.

If you are able to solve your problem when using a single ThrottleStop profile then you can go back and try using multiple profiles. The FIVR Overclock box should be checked in each profile.

 

[MicrobeMossberg]

Hi Unclewebb,

Thanks for the extensive reply.

I did a bit of tweaking with everything you've said. Turned off the default profiles:

I've set the Speed Shift to 50 as you said. Interestingly, what it did was cap the max clocks at 3.9 GHz (even though they're set at 5 GHz), as you can see:

What I assume is happening is that the Max value for the Speed shift is multiplied by 0.8x or so (isn't that a bit weird?), resulting in the previous clock speed cap. I've set it to 75, and it seems to have registered and there is no speed limitation in this particular case:

Even after doing this, it could not prevent the eventual downclock after hibernation.

I did however, find the sequence of settings that lead to the downclock after hibernation.

When I hibernate using setting 3 (base factory clock speeds (4.8P, 3.4E), it will downclock after reopening if I go back to setting 1, which is the OC setting (5P, 3.6E).

When I hibernate using setting 1, which is the OC setting (5P, 3.6E), it will preserve the max clock speed cap after reopening, even if I switch back and forth between different settings.

Setting 3 (factory clocks, 4.8P, 3.4E) is this:

Whereas Setting 1 (5P, 3.6E) is this:

BTW, also did a 8 thread bench with the following results: some of the warnings are not there any longer.

Any ideas on why it would downclock after reopening when hibernating with the base factory clock speeds?

Thanks!

 

[unclewebb]

It looks like Intel changed how the Speed Shift Max variable should be set in their 12th Gen CPUs. It was nice and simple with 10th and 11th Gen CPUs. Set Speed Shift Max to 50 and the CPU would use the 50 multiplier.

With your 12th Gen CPU I would consider setting this to the recommended Max value which ThrottleStop shows is 85. You should be able to use the Turbo Ratio settings to control the maximum multiplier so it does not go sky high.

The FIVR monitoring table is still showing a Speed Shift EPP value of 128. This can interfere with maximum performance and maximum CPU speed. Is there a Windows performance slider in the system tray? Have you tried setting that all the way to the right for maximum performance? Does moving this slider back and forth make any difference to the EPP value that ThrottleStop reports?

On the main screen of ThrottleStop you can also try selecting the High Performance box which should set Windows to the High Performance power plan. Does EPP change then? Switch back to the default Windows Balanced power plan if there is no change to EPP.

Hopefully you can lower the EPP value. Can you show me a screenshot of ThrottleStop running the TS Bench 8 Thread test while you have this throttling problem and while this test is in progress and the CPU is loaded?

I would set IccMax to the max, 255.75, for both the core and the cache for each ThrottleStop profile.

 

[MicrobeMossberg]

It looks like Intel changed how the Speed Shift Max variable should be set in their 12th Gen CPUs. It was nice and simple with 10th and 11th Gen CPUs. Set Speed Shift Max to 50 and the CPU would use the 50 multiplier.

With your 12th Gen CPU I would consider setting this to the recommended Max value which ThrottleStop shows is 85. You should be able to use the Turbo Ratio settings to control the maximum multiplier so it does not go sky high.

The FIVR monitoring table is still showing a Speed Shift EPP value of 128. This can interfere with maximum performance and maximum CPU speed. Is there a Windows performance slider in the system tray? Have you tried setting that all the way to the right for maximum performance? Does moving this slider back and forth make any difference to the EPP value that ThrottleStop reports?

View attachment 261891

On the main screen of ThrottleStop you can also try selecting the High Performance box which should set Windows to the High Performance power plan. Does EPP change then? Switch back to the default Windows Balanced power plan if there is no change to EPP.

Hopefully you can lower the EPP value. Can you show me a screenshot of ThrottleStop running the TS Bench 8 Thread test while you have this throttling problem and while this test is in progress and the CPU is loaded?

I would set IccMax to the max, 255.75, for both the core and the cache for each ThrottleStop profile.

Thanks for your reply,

I've set the Speed Shift to 85.

The Speed Shift EPP value is indeed unresponsive to any power plan setting, either in Windows or TS (Performance, Balanced, Battery), nor was there a performance slider. Perhaps that's a Windows 11 thing? As you can see:

What did respond was setting the Shift Speed EPP yourself to 0

Even with all of these adjustments, the downclock still happens post hibernation if I go into hibernation with factory max clocks (4.8P, 3.4E). But still no downclock occurs when I hibernate with OC settings (5.1P, 3.9E in this example)

This is quite odd..

 

[unclewebb]

Speed Shift EPP value is indeed unresponsive

Intel's latest 12th Gen CPUs have a new feature. There are two separate EPP variables and the CPU can choose to ignore any EPP requests from the user or EPP requests from Windows. The CPU gets to manage itself however it likes. That might be why changing Windows power plans no longer makes any difference.

Checking the Speed Shift EPP box in ThrottleStop allows you to change the EPP request variable but it appears that the CPU has decided to ignore this request. If Speed Shift EPP is actually at 0 and the CPU is using this EPP request information, the CPU multipliers would all be up over 40. The CPU would be running at full speed even when lightly loaded. That is how previous Intel CPUs since the 6th Gen have always worked. Your screenshot above shows lots of low multipliers when EPP is set to 0 so it appears that on your CPU, this no longer works.

I do not presently have access to a computer with a 12th Gen CPU running Windows 11. No plans to upgrade at the moment so no way for me to try and solve this problem if it is solvable. Your Intel CPU gets to decide what it thinks is the best compromise between performance and power savings.

the downclock still happens post hibernation

Can you show me a screenshot of ThrottleStop with Limit Reasons open while an 8 Thread - 960M TS Bench test is in progress? Can the lower clocks be seen in ThrottleStop immediately after you resume and start this test or does it take a minute or two for the clocks to be reduced? I am trying to figure out if the reduced clocks are related to the Speed Shift EPP variable that ThrottleStop has lost control of or if the reduced clocks are being caused by something else. It could be related to Windows 11 over writing one of the CPU control registers. If ThrottleStop is not aware of what Windows 11 is doing, ThrottleStop will not try to correct the issue.

For the above test, hibernate at 4.8 GHz, resume, switch to your 5.1 GHz profile and then run the TS Bench test so I can see an example of the downclock. Show me two separate images of this test if the downclock is time related.

 

[MicrobeMossberg]

Hey Unclewebb, thanks for getting back to me.

Did a few tests after hibernating at 4.8 GHz.

First, the downclock (for the P cores, not the E) happens immediately post-hibernation, without even starting up a TS bench, as you can see:

I decided to do a TS bench run with this downclocked setting. Before even starting the TS bench, I get a few limit notifications:

On idle, I get the EDP OTHER flag for the ring (sometimes flaring up red) and GPU and PL2 for the core (eventhough the wattage used was 10-20 W max on idle). To be fair, I also get exactly these flags on the 4.8 GHz setting on idle. What would that mean?

Did a 8 thread 960M bench at the downclocked 5.1 GHz setting and it throttles immediately ofcourse due to a lack of an undervolt. Included some pics during the test:

As you can see, even during bench the clocks are capped at 4.8 Ghz, even though the E cores do reach their set turbo limits. Did a clear on min/max to be sure as well.

As far as I could tell, the switch to downclock happens immediately post-hibernation and not post-benchmark.

 

[P4-630]

That's one powerful CPU you got there crammed in a laptop, only if you can keep that sucker properly cooled.

 

[unclewebb]

Before even starting the TS bench, I get a few limit notifications:

Some yellow boxes in Limit Reasons are normal after first starting up or after resuming from sleep or hibernate. These flags are triggered within the CPU before the BIOS has a chance to fully setup the CPU correctly. You can ignore this info. It is meaningless. Red boxes in Limit Reasons while the CPU is loaded are what you need to watch out for.

The screenshot you posted with the TS Bench in progress shows that thermal throttling is the reason your CPU is downclocking. If your computer cannot run 8 threads of the TS Bench without overheating then try running a TS Bench - 4 Thread test. I need to see what ThrottleStop shows for the CPU multiplier when the CPU is loaded and it is not thermal throttling.

Turn on the ThrottleStop Log File option before you start doing any testing. When finished testing, exit ThrottleStop so it can finalize the log file. Attach a log file to your next post.

 

[MicrobeMossberg]

Some yellow boxes in Limit Reasons are normal after first starting up or after resuming from sleep or hibernate. These flags are triggered within the CPU before the BIOS has a chance to fully setup the CPU correctly. You can ignore this info. It is meaningless. Red boxes in Limit Reasons while the CPU is loaded are what you need to watch out for.

The screenshot you posted with the TS Bench in progress shows that thermal throttling is the reason your CPU is downclocking. If your computer cannot run 8 threads of the TS Bench without overheating then try running a TS Bench - 4 Thread test. I need to see what ThrottleStop shows for the CPU multiplier when the CPU is loaded and it is not thermal throttling.

Turn on the ThrottleStop Log File option before you start doing any testing. When finished testing, exit ThrottleStop so it can finalize the log file. Attach a log file to your next post.

Hey Unclewebb, thanks for your reply.

There is a difference between an imposed hard ceiling in your max clocks and the downclock that occurs due to throttling right? I'll post the logs.

What I also noticed is, when starting a full-screened game at factory settings (4.8 GHz), the downclock also occurs during gaming and after gaming. Exactly the same as the hibernation issue... odd isn't it?

 

[unclewebb]

Intel CPUs can throttle for a variety of different reasons. There are also a variety of different methods that can be used to impose a hard MHz ceiling on them. That is why I would like to see a log file and hopefully a 4 Thread TS Bench test. Thermal throttling when the CPU gets too hot is normal. Whatever else that might be going on needs to be investigated further.