Turbo time limit not has no effect - i9-13900hx

[DoDe]

Hi,
I have a strange situation.
For i9-13900hx when I set the PL1=140W and PL2=190W the turbo time limit set at 58 seconds takes exactly as it is set.
But in case I set the PL1=55W and PL2=157W (as the intel specification) the turbo time limit set at 28 seconds takes less then 10 seconds.
Both settings are tested in cinebench.
What am I missing?

 

[unclewebb]

The turbo time limit is not a written in stone kind of number. You will not always get the exact number of turbo boost seconds that you request. Sometimes you might and sometimes you might not. The exact algorithm that Intel uses to determine the turbo boost duration has never been publicly documented. Someone at Intel must know but I have never found any info about this.

I once read a document that compared turbo boost to water in a bath tub. Running the CPU well beyond what PL1 is set to will likely empty the turbo boost tub faster. Full turbo boost will not last as long compared to only running the CPU just a few watts beyond PL1.

That is all I know. Intel is pretty good at hiding their secrets. You need to be a big company and sign an NDA to try and get any info out of them. I am just an individual programmer so I do not have access to any of the good stuff.

turbo time limit set at 28 seconds

For many years 28 seconds was the official recommended value for many Intel CPUs. When they started struggling to keep up with AMD, they bumped the default value up to 56 seconds. This allowed them to be more benchmark competitive. Cinebench R23 completes very quickly so the CPU can run full turbo boost for the entire benchmark.

I think I have had my 14900HX briefly up to somewhere close to 260 Watts before it got hot and started thermal throttling. As long as the CPU is not overheating, I do not think you have to be too conservative with the power limits. No one knows how long the 13th and 14th Gen HX processors will last. If you are running the latest microcode versions, 0x12B or 0x12C, your CPU should be OK for a long time.

Here is an easy way to update the microcode if your laptop manufacturer has not yet updated their BIOS versions.

[INTEL]-How To Update Your Microcode for Intel HX 13/14th Gen. CPUs Laptops/Mobile Easily.

Let me start with a small background info... Since we know all Intel's fiasco about what is happening with Raptor Lake/Raptor Lake Refresh which is Intel's codename for the 13th and 14th generations of Intel Core processors, Most OEM/Vendors doesn't want to provide BIOS update that includes...

www.techpowerup.com

 

[DoDe]

Thanks for the explanation.

I think I have had my 14900HX briefly up to somewhere close to 260 Watts before it got hot and started thermal throttling.

How did you achieve 260W on the mobile CPU since the power brick is 330W and there is a voltage regulator which gives an alarm above 190W for PL2?

Here is an easy way to update the microcode if your laptop manufacturer has not yet updated their BIOS versions.

I'll check it but I thought the problem with 13th and 14th CPU's was related to the desktop once.

I would like to share with you my throttle stop log related to the cinebench stress. Would you be kind to check it for eventual problems? For my little knowledge it should be fine but an expert check is always better.

 

[DoDe]

Attached are my settings, logs and cinebench score.
What is worrying me is that during ther multi core cinebench test I get the CPU clock around 4640 MHz. It starts from 5420 MHz and when it sucks more watts it decreses to 4640 MHz. I don't know if this is correct and if it's based on the turbo bost settings.
The cinebench test was done with the fans set on maximum speed. I do not own a cooler.
Cinebench score: 33986 (I also hit 34400)
CPU Core=150,4
CPU P / E Cache=100,6
IccMax=255,75
PL1=140W
PL2=190W
TTLL=56 seconds

For the daily usage I set the PL1=55W, PL2=157, TTL=28 seconds. I also disable TURBO since there's no gain in gaming and the thermals are low. For example in Cyberpunk with everything on ULTRA, DLSS on I get somewhere around 68-77'C.

Should I be worried?

 

[unclewebb]

   DATE       TIME    MULTI   C0%   CKMOD  BAT_mW  TEMP   NVIDIA GPU     VID   POWER
2025-02-23  17:09:03  50.26    8.5  100.0       0   66    1455    43   1.2991   53.5
2025-02-23  17:09:04  51.65   26.2  100.0       0   69    1455    43   1.2472   75.0
2025-02-23  17:09:05  49.92   49.5  100.0       0   85    1455    43   1.1603   99.6   PL2
2025-02-23  17:09:06  47.12   91.6  100.0       0   87    1455    43   1.1147  184.3   PL2
2025-02-23  17:09:07  37.00  100.0  100.0       0   90    1455    43   1.1165  190.0   PL2
2025-02-23  17:09:08  37.00  100.0  100.0       0   90    1455    44   1.1189  189.8   PL2
2025-02-23  17:09:09  37.00  100.0  100.0       0   91    1455    44   1.1152  189.8   PL2
2025-02-23  17:09:10  37.00  100.0  100.0       0   91    1455    44   1.1118  189.9   PL2

Here is a good example. When you go up to full load (C0% = 100.0), the CPU immediately starts PL2 power limit throttling right at 190 Watts. That makes sense. Your CPU is being forced to slow down so it does not exceed the 190 Watt PL2 power limit that has been set. If this test runs long enough, at some point the CPU will be forced to slow down further so it does not exceed the 140 Watt long term PL1 power limit.

Your CPU is only using the 37 multiplier and is already reporting 190 Watts of power consumption. I am running a very similar 14900HX and I am trying to maintain the full 52 multiplier during Cinebench testing. If the 37 multiplier requires 190 Watts then it is easy to see that the 52 multiplier is going to need close to 260 Watts.

How did you achieve 260W on the mobile CPU since the power brick is 330W and there is a voltage regulator which gives an alarm above 190W for PL2?

Some things are like magic. There are tricks that can be used to push ones hardware to the edge of self destruction.

The power consumption number that Intel CPUs generate is only an estimation. It can also be manipulated by changing the AC / DC loadline values. Its sole purpose is to manage the turbo boost function. No one knows how accurately this calculated number represents actual power consumption. I would not try to compare this CPU generated power consumption number to the power rating of a power brick. The Intel number may not be the same as measured power consumption.

Are you using the V/F Point feature yet? V/F Point 1 should be set to 150 for both the core and the cache. This has allowed some people with HX processors to reliably undervolt the P cache much further. With a 14900HX, I have both the CPU Core and P Cache set to the same -150 mV. I am not undervolting the E cache at all. I am not sure how much you can gain by undervolting the E cache. If you go too far it might cause instability without reducing power consumption or heat any significant amount. The Core and the P Cache are the two most important voltages to adjust.

IccMax=255,75

255.75 was the maximum IccMax value for the older 10th Gen processors. The max for 12th Gen and newer processors is 511.75. I set the Core, P Cache, Intel GPU and iGPU Unslice all to the new max, 511.75. This scares some people. Setting this limit sky high can help reduce some EDP throttling. It may not be a safe thing to do. I also set Power Limit 4 to the max, 1023. Doing stuff like this is good for full load Cinebench testing but is probably not necessary for day to day use.

I'll check it but I thought the problem with 13th and 14th CPU's was related to the desktop once.

13th and 14th Gen desktop and mobile CPUs are exactly the same. The mobile parts are less likely to have issues because the default power limits are set lower and cooling for the mobile parts is not nearly enough to push these CPUs to the max. A desktop setup with a triple fan 360mm AIO allows one to push these CPUs much further to even higher power values for extended periods of time.

Intel has not said too much about the mobile parts but there are already reports of some HX processors crashing during certain games just like the 13th and 14th Gen desktop parts do. In theory, if your CPU is not already damaged, using the updated 0x12B or 0x12C microcode should help reduce the possibility of any long term degradation and instability.

Should I be worried?

No worries. You are using ThrottleStop to fully control your CPU and run it at a very leisurely pace compared to what some people are doing with their laptops. Instead of disabling turbo boost, you could setup a profile in ThrottleStop for gaming that uses reduced turbo power limits. This is another way to keep temperatures down to a manageable level without sacrificing too much performance. Setting PL1 and PL2 to somewhere around 100 Watts might be good enough for many games.

 

[DoDe]

Are you using the V/F Point feature yet? V/F Point 1 should be set to 150 for both the core and the cache. This has allowed some people with HX processors to reliably undervolt the P cache much further. With a 14900HX, I have both the CPU Core and P Cache set to the same -150 mV. I am not undervolting the E cache at all. I am not sure how much you can gain by undervolting the E cache. If you go too far it might cause instability without reducing power consumption or heat any significant amount. The Core and the P Cache are the two most important voltages to adjust.

Yes already have it at 150 for both Core and Cache. Thx for the E cache tip.

255.75 was the maximum IccMax value for the older 10th Gen processors. The max for 12th Gen and newer processors is 511.75. I set the Core, P Cache, Intel GPU and iGPU Unslice all to the new max, 511.75. This scares some people. Setting this limit sky high can help reduce some EDP throttling. It may not be a safe thing to do. I also set Power Limit 4 to the max, 1023. Doing stuff like this is good for full load Cinebench testing but is probably not necessary for day to day use.

I wonder what benefit is from having the Intel GPU and iGPU current set to maximum. Isn't less current producing less heat and use the gap for the CPU better? Or it's better because the CPU manages better the current if it's the same for all components?

No worries. You are using ThrottleStop to fully control your CPU and run it at a very leisurely pace compared to what some people are doing with their laptops. Instead of disabling turbo boost, you could setup a profile in ThrottleStop for gaming that uses reduced turbo power limits. This is another way to keep temperatures down to a manageable level without sacrificing too much performance. Setting PL1 and PL2 to somewhere around 100 Watts might be good enough for many games.

100 Watts is the sweet spot.

Two more things:

1. Regarding the Power Limit 4 value and the check box next to it. I still don't get it what is the main goal. Not so much information on the web and what's the best number,
2. The lock check box in the PL1 and PL2 values - what's the purpose? To lock the values so they do not exceed?

 

[unclewebb]

Checking the Lock box in the MSR Power Limit Controls section locks the MSR turbo power limits. Any other software that might be running on your computer that tries to make any changes to these power limits will not be able to. Any time a value is locked, the only way to unlock it is to reboot your computer. Doing a sleep resume cycle is sometimes enough to unlock the power limit register.

I wonder what benefit is from having the Intel GPU and iGPU current set to maximum.

The reason I recommend setting everything and anything to the max is because I do not want 10 different power and current limits interfering with the performance of my computer. I am not interested in the game of trying to set everything perfectly. When everything is set on the very edge of throttling, the room temperature might change a few degrees, the CPU starts throttling and a user might not realize what is happening. Games start stuttering and a user cannot figure out why everything was great yesterday and not so great today. Who wants that hassle?

The Lock box beside Power Limit 4 locks this power limit to whatever value you have set. One reason I prefer to lock things is that helps improve efficiency. ThrottleStop only has to look at one bit within the register to see if it is locked. If it is locked, ThrottleStop does not have to do anything else. It does not have to check if the value has changed because there is no chance that a locked register will ever change. Tiny little things like this add up. ThrottleStop only needs one tenth as many CPU cycles compared to what XTU needs to manage a CPU.

In the good old days there was a thermal limit and that was it. If a CPU core reached 100°C, it would thermal throttle to protect the CPU. That kind of throttling makes sense. I find all of the other throttling schemes dreamed up by Intel over the years to not be all that necessary. Perhaps Intel could hire DOGE next to clean things up at Intel. Luckily there is ThrottleStop so individual users can setup their CPUs however they like. If you are ever concerned that raising a limit could damage your CPU, do not do it.

I recently discovered that the Core and Cache IccMax settings do not actually do anything on my 14900HX. The IccMax value that the Intel GPU is set to acts like a master IccMax control. This value is what controls IccMax for the entire CPU. As long as this is set to the max, I can set any of the other IccMax registers to any value from 1.00 to 511.75 and it will not make any difference. I used to always recommend setting IccMax for the core and P cache to 511.75 but on my CPU, doing that is not enough. Intel GPU IccMax seems to be the only important one. That could just be a bug in my 14900HX or maybe Intel has done this deliberately with a microcode update to try and protect the CPU from degradation due to the IccMax current limit being set sky high.

what is the main goal

My primary goal is always max performance. The original reason I wrote ThrottleStop was to stop any throttling so games could run smoothly without any throttling. Low power consumption and low temperatures are good goals when trying to come up with some everyday 24/7 type settings.

 

[DoDe]

Thank you very much for the explanation. I appreciate your amazing job.
Is there any possibility to donate for this amazing app you have done?

 

[unclewebb]

Is there any possibility to donate for this amazing app you have done?

Thanks for the offer. I am doing OK at the moment so I always suggest helping someone in your local community. Buy a bum a sandwich. Someday that bum might be me.