Throttling Asus ROG Strix (14900HX & 4070)

[Letty]

My laptop experiences such a picture: 14900HX goes up to 70-80C in games and spikes up to 90-99 ocassionaly for milliseconds (then comes back to 70-80) with 1-2% (sometimes 6-8% but rarer) throttling. For example, when i am opening a map or starting some loading stuff in a game. No freezes and I don't notice any slow downs in the game. But. It happens either at stock settings or throttlestop.
It looks like that (100% CPU load was a stress test with 0% throttling after setting ThrottleStop, after it was Cyberpunk 2077):

My ThrottleStop looks like that. It is very rough and currently I am not about a fine tuning, because somehow even if I set PL1 and PL2 to such ridiculous limits for 14900HX it still hits 90C+ and throttles. I have tried to set the Prochot offset to 100C, no changes. Seems like the CPU throttles even at 91C.

As you can see, Max Power was 83W during the game, but CPU still hits 90+ and throttles in game moments. How come? Shoud I reduce my 14900HX to 15W and make him a Celeron to avoild it? xD Probably I miss something out.
GPU is running 75C during the game, as far as CPU sits upon the same cooling pipe it should be close to GPU temp. Is it even real to prevent the CPU from spiking up with such a little but throttling?
Any thoughts?

 

[unclewebb]

How come?

A 14900HX in a laptop is like having a funny car engine in a passenger car. The cooling has to be perfect.

I am not familiar with your Asus laptop model. Some Lenovo Legion laptops with the same 14900HX can run in the 150W to 175W range before they inevitably trigger thermal throttling. Your level of cooling is about half of that. I am not sure if the heatsink Asus used on your laptop is in the same league as what Lenovo is using. I do think it should be able to dissipate more heat than what it is presently capable of.

If it was my laptop, I would open it up and replace the thermal paste with some Honeywell PTM 7950. It is possible that the person working on the assembly line that day botched the installation of the heatsink. This is a known issue. I would not trust the local shop to try and fix this. You have no idea what level of experience they have or what thermal paste they will use. It is usually best to do this critical maintenance procedure yourself so you know it has been done properly.

GPU is running 75C during the game

Nvidia GPUs tend to run cooler compared to Intel CPUs in the same laptop. This might only mean that Intel does a better job of positioning their multiple core temperature sensors on the hot spot of each core. Trying to compare the peak temperature of a Nvidia GPU vs an Intel CPU does not really tell you anything useful.

 

[Letty]

150W to 175W range before they inevitably trigger thermal throttling

The weird thing is - it doesn't work that simple in my case. I can run a test at 120W with no throttling for hours.
1) A liquid metal here from the factory. Both the CPU and the GPU.
2) A steady load goes well. So the CPU may throttle at 80W in some apps but can be calm at some 120W steady load.
3) And the most interesting thing: the speed of fans is the same when throttling happens or doesn't happen in these scenarious (like 4500 - 4000 - 5600). This laptos has 3 fans.
4) You know, something might have worked out if I saw a dose-dependent effect. Like I set 120W = 20% throttling. I set 80W = 5%. Yet nope, the change is minimal. 1-5% anyway.
5) I didn't compare the chips but I meant that while GPU heats up the same pipe up to any dergree - a CPU goes up to the temp

 

[unclewebb]

Some laptops use an embedded controller EC to set separate turbo power limits compared to the MSR and MMIO power limits that ThrottleStop has access to. Whether the Nvidia GPU is active or not can be used to determine how much power is available for the CPU to use.

liquid metal

That is what I thought. It is still possible that someone did a poor job applying it.

https://www.reddit.com/r/ASUSROG/comments/17cs9fm

So the CPU may throttle at 80W in some apps but can be calm at some 120W steady load.

When testing, is the GPU active at 80W? The extra heat from the Nvidia GPU might cause CPU thermal throttling even when CPU power consumption is lower compared to a 120W CPU only stress test like Cinebench.

 

[Letty]

It is still possible that someone did a poor job applying it.

Sure. Anyway watching other laptop reviews thay have close temps in games. 70-80C. Or even higher. And they also spikes up to 90C+ for milliseconds. I just don't know if they have throttling there.

When testing, is the GPU active at 80W? The extra heat from the Nvidia GPU might cause CPU thermal throttling even when CPU power consumption is lower compared to a 120W CPU only stress test like Cinebench.

Inactive. Unlucky I can't put such a load upon the CPU and GPU simultaneosly. I've tried - they both are just receiving less W to balance it out.

Well. Anyways, the last questiong is does it harm the CPU? Occasional 90C+ hits with a few throttling?

 

[unclewebb]

does it harm the CPU?

The purpose of thermal throttling at 100°C is to make sure the CPU does not suffer any temperature related degradation. Intel has always claimed that under 100°C is a "safe operating temperature". Anything over that is not guaranteed by Intel to be safe. Your laptop shows PROCHOT Offset is set to 5 so it will start thermal throttling a little prematurely at 95°C. That should reduce any temperature related risks even further.

The main problem with 13th and 14th Gen desktop CPUs is voltage related degradation. The jury is still out on the very similar HX processors that are in the laptops. I would be more worried about high voltage than high temperatures.

Your ThrottleStop screenshot shows very high VID voltage. Are you using ThrottleStop to undervolt the CPU? You should be doing that if possible. The next TS version will allow V/F Point voltage adjustment of the HX and K series CPUs. This might be a great way to keep the max VID voltage in check.

 

[Letty]

The main problem with 13th and 14th Gen desktop CPUs is voltage related degradation.

The microcode update has come in august/october.
"Two months later, in August, Intel released the 0x129 microcode update, which limited VID requests by the CPU to 1.55 V. Finally, in October, Intel announced that it had discovered the root cause of the majority of CPU failures and released microcode 0x12B. This microcode update prevents the processor from improperly requesting elevated voltages during idle and light-load states."

Are you using ThrottleStop to undervolt the CPU?

I've tried but no luck. Tried to lower the overall CPU Offset voltage, tried to decrease Cores Ratio.
So currently I don't use anything cause it makes no difference yet.
Voltage is jumping from 0.700 to 1.400 ocassionally.

 

[unclewebb]

I've tried but no luck.

Can you explain that? Most people who undervolt their 14900HX CPUs have good results. What sort of settings did you try and what were the results? Did you undervolt both the P core and the P cache? Did you adjust the mV Boost value to this same value?

microcode update prevents the processor from improperly requesting elevated voltages

Call me a sceptic. I am not yet convinced that the microcode update will completely solve the degradation issues. I have already heard about a user running the latest microcode from day one and they are already having stability issues compared to when their CPU was new and first installed. I think it is still too early to know how well the microcode update really works long term.

The 13th and 14th Gen seem fragile. I would undervolt and I would also consider lowering the 1 and 2 core active multiplier. If you set all of the turbo multis to 52, the CPU will not have to jump so high with the voltage. This will sacrifice some light load performance going from 58 to 52 but realistically, no one is ever going to notice a performance difference when lightly loaded surfing the internet or watching a YouTube video. Your CPU might thank you later.

 

[Letty]

Did you undervolt both the P core and the P cache? Did you adjust the mV Boost value to this same value?

I haven't gone that deep. As far as I know, decreasing Watts automatically decreases voltage (P = I * V if I am not wrong) so I decided to start from there with P1 and P2.
I will try other things tomorrow.

Also, when I make settings in Throttlestop I can't properly reboot the laptop. It stucks upon a logo. So i need to shut it down with the power button and only then the laptop finally is able to boot the system. This doesn't happen with Intel Extreme Tuning Utility and Asus Armory Crate. A huge factor for me, but probably can handle it if something really useful comes out.

 

[Letty]

I am not yet convinced that the microcode update will completely solve the degradation issues.

Probably need to monitor max voltage levels for a day or two. But need to acknowledge what level is exactly a red line to find out if the CPU crosses it.

 

[unclewebb]

when I make settings in Throttlestop I can't properly reboot the laptop

What changes are you making?

 

[Letty]

Todays settings:
1) Cores 1 and 2 ratio 58 -> 52
2) CPU Core -125 Voltage Offset (Adaptive)
3) CPU P Cache -125 Voltage Offset (Adaptive)
4) CPU E Cache -125 Voltage Offset (Adaptive)
5) mv Boost 150

No reboot problems this time. The reboot problems were after changing PL1 and PL2. No such changes this time.
Maximum VID i've seen 1.41 for a millsecond. At stock maximum VID was 1.52 or something. Is it good enough?

And I've tested Cyberpunk 2077 again. No FPS decline. The temps:
1) CPU cooled down from 72-76 to 66-71
2) GPU cooled down to from 73-76 to 68-71 (interesting, the shared pipe?)
3) AIDA64 Throttling Graph became green for the first time in this laptop's life. After playing cyberpunk it became red again. Google says, that a red color means throttling is here, it is just less than 1%. But it's weird, cause its always been red even under no load and 40C temp. So I guess that's not a big deal.
But finally it shows a stable 0% in the game. So improvement is definitely here.
Trying to get further improvements. -150 Voltage Offset unfortunately shuts the laptop down in the game. Trying to find the max stable value.

 

[Letty]

Ooookay. Here is the stable tuning:
1) Cores 1 and 2 ratio 58 -> 52
2) CPU Core -160.2 Voltage Offset (Adaptive)
3) CPU P Cache -140.6 Voltage Offset (Adaptive)
4) CPU E Cache -140.6 Voltage Offset (Adaptive)
5) mv Boost 150
6) PROCHOT offset 5

About the temp in Cyberpunk 2077: actually it hasn't changed. I thought there was decreasing to 68-69 but no, it is just a temporal shift until the laptop heats up. Still 72-74.
But
Haven't seen throttling on this settings yet.
However, at CPU Core 140.6 there was a few 1% throttling peaks but very rare. So I think it can happen even at 160.2 but i guess it's not the big deal anyway.
Also, the freq of the CPU in AIDA64 heightened up from 3.7 to 4.07 under the same 120 Watts.
VID never goes above 1.39. So i guess no need to undervold anything further.
I've put ThrottleStop to load at system start.

Thanks,

unclewebb​

 

[unclewebb]

@Letty
If you are ever bored, turn on the ThrottleStop Log File option before you play a game. That will record temperatures, performance and any reasons for throttling. I would trust this data more than anything else. Exit AIDA while logging data with ThrottleStop and Attach a log file to your next post. I am curious to see how your CPU is really running.

mv Boost 150

When your core offset undervolt is greater than -150 mV, do not hesitate to increase the mV Boost value further. Any value in the 150 to 200 range should be OK. This is only increasing voltage at low MHz which can help improve light load stability when trying to use a large undervolt.

Checking the Sleep Defaults boxes is another trick that can help improve stability when resuming from sleep. This resets the voltage offset registers to 0 just before entering sleep mode. It then automatically restores your previous offset values just after your laptop resumes from sleep. This feature can help smooth voltages during the sleep resume transition.

VID never goes above 1.39

I think 1.39 is a much more comfortable voltage to operate an Intel 13th or 14th Gen CPU at. Someone at Intel might think that 1.55 should be OK for long term use. My thinking is that somewhere around 1.40 or less has got to be better for the long term life of any CPU.

Still 72-74

Most users with 14900HX processors can only dream about temperatures like that when gaming. Those temperatures have me questioning the throttling data that AIDA is reporting.

Consider clearing the Thermal Velocity Boost box in the FIVR window. This dumb feature when checked causes the CPU to thermal throttle 100 MHz when the CPU reaches approximately 70°C. It is a useless feature for most users especially when your temperatures are fine. What Intel calls Boost is actually throttling in disguise.

Try checking the Lock PROCHOT Offset box in the Options window if this feature is not already locked. Checking this will prevent any software from randomly changing the throttling temperature. Some Lenovo laptops used to do this.

Is it good enough?

Your results look great.

 

[Letty]

Ooough. The game and AIDA were fine, but got a BSOD in Cinebench. So I need to reach perfect stability.

When your core offset undervolt is greater than -150 mV, do not hesitate to increase the mV Boost value further.

Yea, I've tried to set mv boost up to 300. Seems like I barely can move P and E cache beyond 140 anyway. 140 is the last stable point here.

Someone at Intel might think that 1.55 should be OK for long term use.

At stock a maximum i've seen is 1.517. The intel's fix was about VID 1.55+ limit. Seems like my CPU is not from the list / the microcode works or I just haven't seen such moments.
So yea. Intel does not even consider as a problem anything below 1.55. Do the 12th gen and older gens live under 1.4V because we know they are okay?

Those temperatures have me questioning the throttling data that AIDA is reporting.

Got you some data attached below.

Well. My last stable config is
2) CPU Core -168.2 Voltage Offset (Adaptive)
3) CPU P Cache -140.6 Voltage Offset (Adaptive)
4) CPU E Cache -140.6 Voltage Offset (Adaptive)
5) mv Boost 250
This config endures 10 mins of Cinebench with 30029 points. But during this test AIDA and HWINFO show throttling. AIDA shows 0% red line (which mean there is <1% throttling, HWINFO shows different things 1-6%. But if I turn ThrottleStop off HWINFO may show even 100% throttling or any other number. Even if temps are low. How come I dunno.
AIDA Stress-test seems easier for my CPU (even green):

Is it safe to end up upon this config?

 

[Letty]

Tried to disable TVB - leads to instability but doesn't solve throttling.

An example of throttling numbers during cinebench.
AIDA <1%, HWINFO 3%. Where it comes from? Why it's different? Why when max core temp 87C? LoL.

Upd. Unlucky when I tried to make 2 cinebench tests in a row - got reboot on the second pass.
So again... Going backwards... I actually don't even know if I need such stability. Seems like in daily usage + games it is stable enough , so... Maybe I don't need hours lenght synthetic tests stability? Well... Okay. Going for at least 2 cinebench in a row.

 

[unclewebb]

What values did you set the turbo power limits to? The Cinebench Success log file you posted shows non stop power limit throttling during Cinebench. It looks like you set the PL2 power limit to 120 Watts and PL1 is set to 100 Watts. The CPU is running at about two thirds of the speed it could be running at during PL2 throttling. The CPU slows down further when the 100 Watt PL1 limit kicks in. It is much more difficult to achieve a stable undervolt when the CPU is being forced to throttle like this. Just for a reference point, some of the Lenovo Legion laptops with 14900HX processors are running Cinebench at up to 180 Watts. Good cooling makes a big difference to the performance of the 14900HX CPUs.

AIDA <1%, HWINFO 3%

Those are completely meaningless numbers. It is not thermal throttling that is causing the problem. It is power limit throttling that is holding you back. With exceptional cooling, a 14900HX can run at up to 5200 MHz during a Cinebench stress test. Your CPU has slowed down to 3200 MHz. That is definitely a lot more significant than 1% or 3%.

CPU MHz while playing Cyberpunk is another problem. The CPU temperature is low because the CPU is running so slow. What Windows power plan are you using. I am not familiar with Asus power management software. Whatever economy oriented power plan your computer is using, try to find a higher performance power plan to use. If you cannot find the Windows High Performance power plan in the Windows Power Options window, try checking the High Performance box on the main screen of ThrottleStop. Run an updated log file while gaming after you do this.

A Cyberpunk load is nowhere near as stressful compared to a Cinebench stress test. A 14900HX should be able to run much faster when playing. A fast CPU equals reduced latency and smoother game play.

 

[Letty]

What values did you set the turbo power limits to?

Haven't set anything. It is by default.
I use a Best Performance windows plan.

The max power 14900HX is capable of is 157W. Where did 180W come from? Maybe there is a mobile model of Intel with such power... Dunno. Whether 14900 or 13900 specs say 157W max.
I know the CPU is running at 120W but I thought it is okay, cause we are trying to make less heat...
And even at PL1 100 and PL2 120W I've got 30000 points in cinebench. Isn't it enought for games? Oo
I've checked out the perfomance of 14900HX. The desktop version 14900K is only 30% faster. In fact this guy should to be a monster even at 120W.

So youre recommending to set it up to 157W anyway?

Finally got something really stable:
1) CPU Core -140
2) P Cache -125
3) E Cache -125
4) mv Boost 150
Four passes of Cinebench and 1 hour of the OCCT extreme test (but with 2 WHEA errors, which means I should add a little voltage, but if it is not critical...).
VID maximum was 1.405V.
I guess it looks good, doesn't it?

Btw about voltages.
Found the intel specs of the13th and 14th gens.
So even 1.55V way far below the max voltage... If an intel CPU can request V up to 1.72 you have to undervolt it down at least -320mv for guaranteed staying below 1.4V.
The 12th gen has the same max.
And where has 1.55V come from? Haven't found any official intel info where 1.55V was mentioned.
Well, i guess 1.4 is oversafe anyway...

 

[unclewebb]

The max power 14900HX is capable of is 157W.

That might be a theoretical maximum that the Intel wishful thinking department dreamed up.

Maximum Turbo Power is configurable by system vendor and can be system specific.

The maximum during a full load Cinebench test can be much higher than 157W.

Intel's "New" Core i9-14900HX is Slower than the AMD Ryzen 9 7945HX3D in Nearly Every Game | Hardware Times

The Core i9-14900HX fares better than the 13900HX but is no match for the Ryzen 9 7945HX3D, failing to register any notable wins.

hardwaretimes.com

Not only is the Ryzen 9 7945HX3D faster than the Core i9-14900HX, but it’s also significantly more power efficient. In most games, the former doesn’t sip over 50W of power, while the 14900HX draws nearly 90W in the Cinebench R23 1T benchmark. In the multi-threaded test, the 14900HX records a power consumption of 223W, up from just 158.5W on the 7945HX3D.

When the thermal paste is perfect, the cooling system of the best laptops with these processors can start getting overwhelmed at about 180W. You end up with either power limit throttling depending on what values the turbo power limits are set to or you end up with thermal throttling if the cooling system is not capable of cooling a 14900HX at full speed and full power.

I thought it is okay, cause we are trying to make less heat...

Reduced power limits are OK if less heat is your goal. The power limits need to be increased beyond 100W when max performance is the goal.

Best Performance windows plan

That plan is typically a modified version of the Windows Balanced power plan. It is not the same as the Windows High Performance power plan that ThrottleStop lets you access.

but with 2 WHEA errors, which means I should add a little voltage

That is a good idea. Any WHEA errors when undervolting is usually a sign that you have reduced the voltage too much.

I like seeing what a CPU can do when all of the limits are at the max and I am outside with a cool winter breeze. I think the North Pole is one of the few places on Earth where the 14900HX can stretch its legs. Intel's updated Core Ultra HX processors should be out early next year. Similar performance to the 14900HX but power consumption and heat should be significantly reduced.

And where has 1.55V come from?

I think the 1.72V max voltage value in the datasheet proved to be too much. The latest BIOS fix for the fragile 13th and 14th Gen have dialed back the maximum voltage. I think the V-Max Stress box in the FIVR window gives users some control over when voltage throttling will begin. At least it used to before the recent updates.

Well, i guess 1.4 is oversafe

I prefer extra safe settings for daily use. Living on the edge is fun when benchmarking. Less voltage equals less heat. That is a feature that most laptop owners prefer.

 

[PHVM_BR]

The max power 14900HX is capable of is 157W

I estimate that the i9-14900HX needs something close to 250W to run Cinebench at 5.2GHz on all P-cores and 3,9GHz on all E-cores WITHOUT UNDERVOLT.

It is an unattainable power consumption for any laptop.

 

[Letty]

less heat is your goal

It is my goal, actually. Probably a 14900HX is overgear for my current needs but I like to have some space.

High Performance power plan that ThrottleStop lets you access

Thanks. I'll keep it in mind.

Any WHEA errors when undervolting is usually a sign that you have reduced the voltage too much.

Ended up with no errors with 2 configs
1) CPU core, P-cache, E-chache - everything -125mV
2) CPU core -150, P-cache -125, E-cache -102
No WHEA errors in both cases. Should I consider the second option as the best one?

Intel's updated Core Ultra HX processors should be out early next year

I think the time of fair hardware calculations is passing. Say goodbie to fair working-boys CPUs. AI. AI is coming.

The latest BIOS fix for the fragile 13th and 14th Gen have dialed back the maximum voltage.

I've just researched the internet a bit and found that either 12th or below gens or 13-14th gens in the datasheets have the 1,72V limit. I haven't found where 1.55V comes from but found that people are nervious about still getting VID above this mark even after the update.
So, somehow the12th gen runs with no problems with no such "1.55V" limit but 13th 14th can't do this.
Intel found out they had problems. And as a temporal decision probably set a limit. Who found out this is exactly 1.55V? Nevermind, let's say there is some limit. Then, in October they had come with an update which "fixes a root problem". No one exactly knows what they have exactly done there. Like they fixed something deeper and brought back an old "1.72" or whatever it is limit.
I don't deny Intel has messed up hard. I am just not really sure about what exactly happened there.
Well. Nevermind. I personally want to undervold my CPU whether for temp or for CPU lifespan.

 

[Valour549]

Have you read my guide?

If you read the whole thing, esp the Advanced Section, I think it will answer all your questions. I have an ASUS with 14900HX too.