[SOLVED] 13900HX only getting 1600 single core on Cinebench R23. Any ideas?

[raywdude]

Edit - Perhaps your AVX Offset setting is part of the problem. Set that to 0 when testing.

@carlos1172 I think we have our winner

 

[Dr. Dro]

View attachment 301651
Final settings maybe? Just testing for stability now.

It's mind boggling how fast that 13900HX CPU is. I just don't see how a laptop is going to handle those thermals, though. 300 W is well within the realm of high-end desktop cooling. I think you'll have a better experience if you loosen the pedal a little bit, try and keep the CPU under 90 full load so you have some headroom for ambient temperature changes over the day - and over seasons.

 

[carlos1172]

@carlos1172 I think we have our winner

I've settled on this

It's mind boggling how fast that 13900HX CPU is. I just don't see how a laptop is going to handle those thermals, though. 300 W is well within the realm of high-end desktop cooling. I think you'll have a better experience if you loosen the pedal a little bit, try and keep the CPU under 90 full load so you have some headroom for ambient temperature changes over the day - and over seasons.

Yep, that's what I did with liquid metal and 185W PL. stays at about 90C.

 

[unclewebb]

I've settled on this

Do you still need to sleep resume to access the 56 multiplier? Was your AVX Offset setting the problem?

If you are using MMIO Lock, you do not need to check Sync MMIO. I would set the turbo time limit to the default 28 seconds. Both power limits are the same so this probably does not matter. Seeing the default 28 second number helps soothe my OCD.

I still like setting Power Limit 4 to a value of 0.

 

[Fizban]

2075 is what I get for single-core on myt 13980HX with no undervolting/tweaking of any sort. Had it undervolted previously, but led to instability so just reverted to stock.

 

[carlos1172]

Do you still need to sleep resume to access the 56 multiplier? Was your AVX Offset setting the problem?

If you are using MMIO Lock, you do not need to check Sync MMIO. I would set the turbo time limit to the default 28 seconds. Both power limits are the same so this probably does not matter. Seeing the default 28 second number helps soothe my OCD.

I still like setting Power Limit 4 to a value of 0.

Sleep Resume unnecessary when using the regedit things that @raywdude provided, and Legion Toolkit. I also put AVX Offset to 0, so I'm not sure which fixed the problem really.

Done and done.

PL4 I set to 227 which equates to 185w for some reason. Otherwise it would spike every now and then and I didn't like that.

Do I set Speed Shift Max to 255 or 0? @unclewebb

These Lenovo machines (and some other OEMs) include drivers that gimp performance for some reason. Here's how to disable them,

1. Open regedit
2. Go to Computer\HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\ACPIVPC
3. Change the value of Start to 4 (which is Disabled). Take note of the original value if you wish to revert it later.
4. Go to Computer\HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\ipf_cpu
5. Change the value of Start to 4 (which is Disabled). Take note of the original value if you wish to revert it later.
6. Restart your PC.
7. Go into BIOS and change "Performance Mode Setting" to Extreme to prevent the EC from overriding the PL1/PL2 settings.
8. Open HWInfo and monitor the CPU and GPU power usage and available power limits. Make sure the CPU is able to pull PL1 140 / PL2 190

Vantage also has a bunch of background processes that can mess with your single thread performance. You can download Legion Toolkit from Github and use the in-app options to Disable Vantage Service and Hotkeys (Toolkit can replace that functionality).

You don't need to use Lock MMIO after you do these tweaks and avoid the weird freezes on reboot/resume from sleep.

Also if you are using *Custom* performance profile, use *Performance* instead as Custom introduces some weird stuttering that occurs frequently.


Amazing, this one simple trick improved my CPU single and multi-core performance by another ~5% on top of already great performance. Nice.

How do I use performance profile. I am getting stuttering in games. I make a profile then select high performance?

 

[raywdude]

Sleep Resume unnecessary when using the regedit things that @raywdude provided, and Legion Toolkit. I also put AVX Offset to 0, so I'm not sure which fixed the problem really.

Glad it helped!

How do I use performance profile. I am getting stuttering in games. I make a profile then select high performance?

Use either Lenovo Vantage or Legion Toolkit to switch to Performance mode.

 

[carlos1172]

Glad it helped!


Use either Lenovo Vantage or Legion Toolkit to switch to Performance mode.

Oh you meant Performance mode as in the laptop's Performance, Quiet, or Balanced, not Windows power plan?

 

[raywdude]

Oh you meant Performance mode as in the laptop's Performance, Quiet, or Balanced, not Windows power plan?

Correct

 

[carlos1172]

Correct

That fixed my stuttering, thank you!

 

[EsliteMoby]

2075 is what I get for single-core on myt 13980HX with no undervolting/tweaking of any sort. Had it undervolted previously, but led to instability so just reverted to stock.

Curious what undervolt value did you set before?

 

[unclewebb]

I'm not sure which fixed the problem really.

When you change a handful of settings at the same time, it is impossible to know what change is necessary and what changes did not do anything.

Checking Speed Shift in the TPL window and changing Speed Shift Max may not be necessary. If it is necessary, I would set Speed Shift Max to the recommended value. Setting Speed Shift Max to 255 like your screenshot shows should also work. Trying to set the Max value to 0 will probably slow your CPU down to a crawl. Don't be afraid to experiment.

 

[EsliteMoby]

IccMax controls the maximum amount of current going to the processor. When IccMax is set too low, this can cause EDP throttling.

Does increasing the IccMax value lead to higher CPU temperature as well? Should I bother with this setting for my 13980hx? The default is 240.

 

[unclewebb]

increasing the IccMax

If your CPU is EDP throttling and raising IccMax reduces throttling, that might create a little more heat.

Remember, it is just a limit. If your CPU does not require more current, setting IccMax sky high will not make any difference. Do some Cinebench testing to see if increasing IccMax is necessary.

 

[raywdude]

LMAO, disabling Thermal Velocity Boost took me from 2057 to 2090 on single core:

 

[unclewebb]

Thermal Velocity Boost

That is a fancy name for a throttling feature. Leaving that checked in the FIVR window tells the CPU to throttle 100 MHz at 70C. Kind of pointless.

 

[carlos1172]

What does it mean if EDP OTHER blinks red every now and then for RING? @unclewebb

My temps are usually in the 70-80 range now after liquid metal, but I'm thinking if the thermal putty on my VRMs and such are still in good condition, or since this is my first LM application, if there's any hotspots not reported on software on my CPU.

Also how come PROCHOT 100C turns red sometimes even though even with More Data option checked, max temp for any individual core never reaches 100?

I also get crashes in Warframe and other games every now and then. I've reset most of the settings except TPL to default.

Any ideas? No GPU overclock done either.

 

[unclewebb]

What does it mean if EDP OTHER blinks red

Intel does not publicly document the OTHER part of EDP OTHER. It could be anything. Usually this only goes red if one of the current limits is set low. ThrottleStop does not have access to the current limits that the embedded controller (EC) can set so EDP OTHER blinking red might be because of that. Does it only blink red in the RING column or does it blink red across all three domains simultaneously?

When Unlock Adjustable Voltage is not checked in the FIVR window, the IccMax value of 511.75 for the core and cache will not be sent or maintained within the CPU.

how come PROCHOT 100C turns red sometimes

There are multiple temperature sensors within the CPU package. On most CPUs, the highest possible temperature is usually coming from one of the cores but this is not always true. Keep in mind that these sensors are only accurate to +/- 5°C. One of the other package temperature sensors might start reporting that it is at 100°C even though it might only be at 95°C. This can trigger thermal throttling. The temperature sensors that Intel uses are not space shuttle quality. They are far from perfect. When PROCHOT turns ted on the main screen of ThrottleStop, that confirms that the CPU thermal throttled, even if the CPU only throttled for a few milliseconds.

I also get crashes in Warframe

Crashes while in game might be CPU related or they might be GPU or GPU driver related. Every time something new is released, it can take Nvidia a year to work out the bugs.

 

[carlos1172]

Intel does not publicly document the OTHER part of EDP OTHER. It could be anything. Usually this only goes red if one of the current limits is set low. ThrottleStop does not have access to the current limits that the embedded controller (EC) can set so EDP OTHER blinking red might be because of that. Does it only blink red in the RING column or does it blink red across all three domains simultaneously?

When Unlock Adjustable Voltage is not checked in the FIVR window, the IccMax value of 511.75 for the core and cache will not be sent or maintained within the CPU.


There are multiple temperature sensors within the CPU package. On most CPUs, the highest possible temperature is usually coming from one of the cores but this is not always true. Keep in mind that these sensors are only accurate to +/- 5°C. One of the other package temperature sensors might start reporting that it is at 100°C even though it might only be at 95°C. This can trigger thermal throttling. The temperature sensors that Intel uses are not space shuttle quality. They are far from perfect. When PROCHOT turns ted on the main screen of ThrottleStop, that confirms that the CPU thermal throttled, even if the CPU only throttled for a few milliseconds.


Crashes while in game might be CPU related or they might be GPU or GPU driver related. Every time something new is released, it can take Nvidia a year to work out the bugs.

I deleted the .ini file and restarted my computer, seems to have fixed the crashing issues. Will play more to be sure.

What are the likelihoods that my CPU has a hotspot not reported in temperatures, because of poor LM application? Is that just a myth to scare people away from using LM? If there was a hotspot not reported in temps, that means the CPU wont throttle right? And it would kill itself?

Though I know I really spread it across the whole CPU, and even applied to the cooler as well.