Attempting i9-13900hx stable undervolt to try and reach 5ghz.

[Garueri]

Hello, Im Garu
I´ve got an alienware m16r1 recently and not to my surprise the i913900hx cant reach further than 3.6 ghz on P cores when in full load (CBR23)
So, Im trying to dream that I have a very golden chip thats definitely going to work at -1000mv making me reach my dream frequency, which is 5ghz, and I shall bring you along this journey

here are the current settings

So, for now im trying too limit 140w but I think it goes 150 and doesn´t care for some reason
I´ve also disabled 8 of the E cores in the hopes to increase thermals or to leave more wattage for my higher frequencies, does it make a difference? I was also thinking about offing all the E cores to try and reach 8/16 5ghz on a laptop
Its fixed at 4,4ghz because anything beyond this throttles on cyberpunk and I hate the 3 digits...
I was using -125mv for both core and P cache because I was told so by some yt guy, but I just read what uncle said about going further on core so I was very happy to see that
Btw if I were to use the E cores how do I undervolt the E cache? How agressive can I be?

EDIT: Forgot to ask
How do you guys deals with the BSOD´s? I´ve got some after some ´´questionable´´ undervoltages and Im scared that one of them is killing my windows, forcing me either a clean install or midia stuff but everytime I got blue screened I ran sfc /scannow and it did catch stuff 4/5 times I got screened. Any tips on how to deal with that? The odds for windows corruption are low to not matterable?

 

[Garueri]

Bad news, I can live super fine with -150 core -125 cache but for some reason, if I let the laptop for hours and hours open, at some point I think it enters hybernation and crashes. Does someone have issues like this?

I guess the journey ends here guys.
with -150 core and -120 cache now i still throttle at 4ghz and cant go further
For some reason hwinfo and hwmonitor show different offsets for voltage. Also the FIVR window sometimes flashes to zero and immediately goes back to the offset, maybe im not get insane results because my offset is actually not applied? Im using alienware m16r1, could the command center be interfering? Or did I already get expected results?

 

[unclewebb]

hwinfo and hwmonitor show different offsets for voltage.

HWMonitor has been using the wrong voltage formula for years. It divides the value in the voltage register by 1000 when it should be dividing that value by 1024. HWiNFO and ThrottleStop are both doing this conversion correctly.

the FIVR window sometimes flashes to zero and immediately goes back to the offset

The FIVR monitoring table accurately reports the offset voltages in real time. If you see a momentary value of 0, some other software on your computer is changing the voltages.

could the command center be interfering?

Definitely. It is never a good idea to have two different programs writing different values to the voltage control register. Use command center or use ThrottleStop. If you run both of them together, there will be conflicts like this.

When using only ThrottleStop, try running the TS Bench 960M test. This used to be a good test on older gen CPUs. If this test reports any errors, your cache voltage needs to be increased.

I think it enters hibernation and crashes

It might. Did you try checking both Sleep Defaults boxes in ThrottleStop? This resets the voltage register to 0 offset the moment before a computer enters Sleep mode and restores the offset voltage after resuming. This can help fix some crashes during sleep or hibernate issues. This probably will not work correctly if you have command center running on your computer.

If your computer is throttling during Cinebench, open up Limit Reasons and post a screenshot that shows ThrottleStop and Limit Reasons while the CPU is fully loaded. Is it a power limit throttling problem or something else? Did you try checking the MMIO Lock box in the TPL window? Many Dell laptops use an embedded controller (EC) to set low ball power limits. That might be why your CPU is throttling.

 

[Garueri]

HWMonitor has been using the wrong voltage formula for years. It divides the value in the voltage register by 1000 when it should be dividing that value by 1024. HWiNFO and ThrottleStop are both doing this conversion correctly.


The FIVR monitoring table accurately reports the offset voltages in real time. If you see a momentary value of 0, some other software on your computer is changing the voltages.


Definitely. It is never a good idea to have two different programs writing different values to the voltage control register. Use command center or use ThrottleStop. If you run both of them together, there will be conflicts like this.

When using only ThrottleStop, try running the TS Bench 960M test. This used to be a good test on older gen CPUs. If this test reports any errors, your cache voltage needs to be increased.


It might. Did you try checking both Sleep Defaults boxes in ThrottleStop? This resets the voltage register to 0 offset the moment before a computer enters Sleep mode and restores the offset voltage after resuming. This can help fix some crashes during sleep or hibernate issues. This probably will not work correctly if you have command center running on your computer.

View attachment 315065

If your computer is throttling during Cinebench, open up Limit Reasons and post a screenshot that shows ThrottleStop and Limit Reasons while the CPU is fully loaded. Is it a power limit throttling problem or something else? Did you try checking the MMIO Lock box in the TPL window? Many Dell laptops use an embedded controller (EC) to set low ball power limits. That might be why your CPU is throttling.

So, here´s the story
I got the m16, played cyberpunk, experienced stutters.
Went out to look for what was causing it.
Found it. CPU hits 100c too often and then throttles to 2000 and something mhz, GPU demands more, gets bottlenecked.
Solutions: 1 - Cap core clock to something manageable, factory settings are 5.7ghz with 0.5 overvolt (insanity and unrealisticaly stupid)
2 - Repaste with liquid metal or better performing thermal paste (I probably have a bad aplication, core 5 and 6 for example have a 10c difference in temps when in load)
3 - Undervolt
Capping didnt bring me what I really wanted, it can only not reach 100c when at 3.9ghz in what I like to call ´´max game load´´ (around 75%) and still bottlenecks gpu by 5% at some scenarios
Repasting brings me risk to end my own machine due to inexperience, specially the dell inverted motherboards
So here I am trying undervolting for the first time

Curiously, bios version 1.6 from alienware had bios undervolting, still scared me because if my undervolt goes oops how do I even get to the bios again to offset? Im not sure if alienware was smart enough to bring the fail boot safe trigger thing to reset stuff in order to boot again, and Im not used to bios recovery through hard disk inside backup, https, battery CMOS or external storage.
They probably removed it after 1.6 due to people bricking their devices and returning the equipment on the 30 day thing you guys at US have.
First, before the unlocked bios I thought I just needed to throttlestop my way into haven, applied the offset and no changes in temps or clocks, then I found out I needed the FIVR to show my offset and got into reasons why it wouldn´t, first, locked CPU´s which wasnt my case, since im HX, then the virtualization stuff, did everything just to later find out alienware had locked undervolting. Finally the unlocked bios.
So since Im also scared to bios downgrade into 1.6 and brick my device I got myself an unlocked bios, pluged it through usb drive and it opens something called intel advanced menu with 95% things I can´t even read.
Inside it, I toggled something called cfg lock to disable, undervolt protection to disabled and I think overclock proction to disable too, someone told me it could interfere with the undervolt thing.
Then I tried undervolting and I think it worked? My clocks did go up by 400mhz but I didn´t test it correctly because I disabled 8 of the eff cores, which could be the cause?
I had an issue that after removing the AWcommand center my fans wouldnt ramp as much, but now I set my bios to ultra performance and I hope it makes my fans ramp
Do you know any third party that has fan control without messing with my voltages?

Now, here comes the testing:
Cinebench r23 load:
PL 1 = 157w
PL2 = 165w
Average P core clock: 3.5ghz
Average E core clock: 3.0ghz
Average power consumption: 105w total, 98w from the IA cores
Curiously, only 4 of my P cores are throttling, and the E cores gradually decrease in temperature, is this a sign of a bad paste?
Score: 25500k. Probably because the fans werent at absolute max since I got rid of AWCC, when at total max my best score was 27k

Now im going to undervolt 100 core and 100 cache on throttlestop
Applied, FIVR recons, HWi recons
Lets test again
OK actually got an increase
PL1, 2 and power basically the same
But avg P and E cores went to 3750 and 3250
Only 4 P cores throttled, again, indicates bad paste?
Score increased to 28300 now
So the undervolting is indeed working
Thats awesome, but it probably means I cant go further than 4ghz right?

As for the sleep thing, I just checked it in and am going to report later about
My limits yellows EDP other on CPU and RING and somethings TVB, my thermal velocity boost is checked

Uncle, this is incredible, could you explain me how undervolting just creates more results? I was told its super safe beyond crashing windows and just needing to roll it back but why doesnt amd and intel do this by default? Specially the higher ends with insane power draw

 

[PHVM_BR]

could you explain me how undervolting just creates more results?

On your laptop, undervolt delivers more performance (higher clocks) and does not reduce the temperature because the defined power limit is not reached because the temperature limit (100°C) is reached before.

For example, its power limit is 157W and its CPU throttles at 105W (100°C) with P-cores at 3.5GHz and E-cores at 3.0GHz.
With undervolt your CPU will continue to be throttled at 105W (100°C), but your clocks will be higher while consuming the same power.

If you limit the clocks to, example, 3.5GHz on P-cores and 3.0GHz on E-cores during Cinebench, your temperature with undervolt will stabilize below the 100°C it would be without undervolt.

Undervolt makes the processor more efficient, consuming less energy to perform the same task with the same performance (same clock).
Consuming less energy, in this case with limited clocks within your thermal headroom, the temperature will be lower.
Less Watts = less °C.

I wrote a topic about undervolt effect on my laptop.

In this case, any undervolt already reduces the temperature as the laptop can deliver all of the processor's performance without any undervolt.
Look:

i7-9750H: undervolt x power ratio

I have a Precision 7540 for more than 3 years and with undervolt it always delivered easily the maximum multicore performance of the i7. At the beginning of the year I changed its thermal paste for the Honeywell PTM 7950 phase change pad and decided to test its maximum cooling capacity and took...

www.techpowerup.com

 

[Intrepidity]

I just got a Legion 7 Pro with 13900HX, but the scores are surprisingly different to yours. Perhaps a repaste needed or something? No idea, but it can't be indended to be this large of a difference when running the same CPU. Not that I complain..

For comparison I just ran 10min CB23:
Wattage fluctuate between 140W and 161W
Temps max 85C. Seem to stabalize around 75C +/-
P-cores fluctuate between 4500MHz and 4800MHz while under load. Didn't pay much attention to E-cores.

(Win11. Single test 35k+)

(Started just before running CB23.)

 

[unclewebb]

P-cores fluctuate between 4500MHz and 4800MHz while under load.

Exit HWMonitor. Run ThrottleStop with Limit Reasons open. Are any boxes lighting up red while Cinebench is running? Find out the reason for throttling and make adjustments to correct the problem. Post a screenshot that shows ThrottleStop and Limit Reasons with Cinebench running in the background. Post screenshots of the FIVR and TPL windows so I can see the rest of your settings.

 

[Garueri]

I just got a Legion 7 Pro with 13900HX, but the scores are surprisingly different to yours. Perhaps a repaste needed or something? No idea, but it can't be indended to be this large of a difference when running the same CPU. Not that I complain..

For comparison I just ran 10min CB23:
Wattage fluctuate between 140W and 161W
Temps max 85C. Seem to stabalize around 75C +/-
P-cores fluctuate between 4500MHz and 4800MHz while under load. Didn't pay much attention to E-cores.

(Win11. Single test 35k+)
View attachment 315394

(Started just before running CB23.)
View attachment 315393

Your print shows that you have the i913900H and not HX, also, you do have a far more agressive undervolt, if your hwmonitor is not wrong like mine, you have -226mv whereas I have -150. But indeed, the pasting jobs from dell are all bad

Exit HWMonitor. Run ThrottleStop with Limit Reasons open. Are any boxes lighting up red while Cinebench is running? Find out the reason for throttling and make adjustments to correct the problem. Post a screenshot that shows ThrottleStop and Limit Reasons with Cinebench running in the background. Post screenshots of the FIVR and TPL windows so I can see the rest of your settings.

btw uncle, I did the sleep thing undo undervolt and it worked, now I can go into sleep without issues, thanks for the tips

 

[Garueri]

Coming back to reply
Undervolt results:

Unfortunately, 5ghz at full load was not possible because my throttle limit seems to be at 100w although my cpu goes up to 167w.

In addition, I apparently have a seriously bad application by dell because a single core of mine is currenntly reaching 100c, with nearby cores cooler by 7c degrees on average, biggest difference being 12c and lowest 4. I´ve searcheed a little and my best options are either the PTM950 something thermal paste that is dielectric (does not fry motherboards) and doesnt need repaste after time, or the thermal grizily new graphene pad thing. They both are not better than comon thermal pastes but come really close or equal, and as a positive side don´t seem to get dry overtime (I intend to use my laptop at least 3 years).
Im not going to do this now because my fan came with issues (strange riskrisk sound at low rotation =~45%) meaning they are going to need to repaste anyway, so I hope it gets fixed when they do so, if not, im going to take maters in to own hands.

However, with currently -175mv to core and -125 to P Cache I was able to reach 28.5k on cinebench, 1500 better than before while also increasing from 3.6ghz on P cores to 4ghz. Although not perfect sounding, I had a terrible issue in cyberpunk where the temps were too high and the clock would drop to base (2.8ghz) and bottleneck my 4070 causing serious stutter. Right now I can achieve higher clocks with lower consumption, so I don´t run into that problem. I´ve locked the clock speed at 4.6ghz because of my pride. I didn´t want to lose to a mere r7 5700x that I could easily made a portable ITX build for cheap with. Since none of games have downclocked at that frequency and also not bottlenecked my gpu, I was satisfied. Though about keeping it to 4ghz that I was sure I could full load without reaching 100c constantly but in some scenarios I had a gpu subutilizaiton. So decided to keep that way

Im extremly thankful for @unclewebb and @PHVM_BR (btw, are you also BR?) for giving me insight on how to do this, as they literaly solved my and dell´s problems for free.
Btw, I shall come back after messing a little with P core cache to see if it does any gains on thermals or speed
Also going to check Sys Agent and iGPU for produtivity use as the best I´ve gotten up until now has been only 4 hours

Thank you, everyone!