BSOD if I touch the pcache offset voltage in throttlestop (i9-13900hx lenovo pro 7i)

[aby]

I'm completely stable in p95 small fft @ -145.5 CPU core offset in TS, but I get a BSOD if I UV the p cache offset voltage at all. I've tried lowering the min-max cache ratio to 8-38 (on the right side of fivr) but I still get a BSOD. Any clue as to what the issue could be?

Attached are pics of my settings. Tysm for any help.

 

[unclewebb]

Have you tried using the mV Boost feature? This boosts the voltage when the CPU is running at 800 MHz. That might help stability when trying to adjust the P cache voltage. Try setting this to 150. This will help balance the core undervolt you are using while still allowing the CPU to undervolt at full MHz.

 

[aby]

Have you tried using the mV Boost feature? This boosts the voltage when the CPU is running at 800 MHz. That might help stability when trying to adjust the P cache voltage. Try setting this to 150. This will help balance the core undervolt you are using while still allowing the CPU to undervolt at full MHz.

View attachment 342531

Thank you so much for the reply. I'll try this out when im back from work and give you an update.
Forgot to say: wouldn't setting icc to max in TS generate way more heat? The lenovo legion discord says to keep it at stock and gizmosliptech kept it on default as well on his final UV settings as seen in this video @ 53:52

 

[unclewebb]

wouldn't setting icc to max in TS generate way more heat?

If a CPU only needs 200 Amps of current to run at full speed, setting the IccMax limit to 200 will allow the CPU to run at full speed. Increasing the IccMax limit to 300 or 400 or 500 will not change anything or make any difference. The CPU was already running at full speed. Increasing the IccMax limit is not going to make it run any faster. Same speed, same power consumption. The CPU temperature will be exactly the same.

If you do not feel comfortable increasing IccMax then leave it as is.

If -100 mV is not stable for the core and cache, go down in steps of 25 mV. Try the bare minimum like -5 mV for the core and the cache. Try lowering the cache max speed further. This was the most common solution for this problem in the 12th Gen CPUs. If you cannot find what is causing this problem then I would leave the cache offset voltage at 0 mV and call it a day. No one seems to know why using ThrottleStop to adjust the P cache voltage is a problem for some CPUs but not others.

 

[aby]

If a CPU only needs 200 Amps of current to run at full speed, setting the IccMax limit to 200 will allow the CPU to run at full speed. Increasing the IccMax limit to 300 or 400 or 500 will not change anything or make any difference. The CPU was already running at full speed. Increasing the IccMax limit is not going to make it run any faster. Same speed, same power consumption. The CPU temperature will be exactly the same.

If you do not feel comfortable increasing IccMax then leave it as is.

If -100 mV is not stable for the core and cache, go down in steps of 25 mV. Try the bare minimum like -5 mV for the core and the cache. Try lowering the cache max speed further. This was the most common solution for this problem in the 12th Gen CPUs. If you cannot find what is causing this problem then I would leave the cache offset voltage at 0 mV and call it a day. No one seems to know why using ThrottleStop to adjust the P cache voltage is a problem for some CPUs but not others.

Ok I set the mV boost feature to 150 and I've passed 2 hours on occt with the small FFT option (-75.2 uv on both the core and pcache) so it definitely helped. I tried dropping the core and pcache offset to -80 and I get a bsod after 5 mins.

When you refer to the 'cache max speed' you're referring to the min-max ratio which is 8-48 by default, correct? Which number should I drop the min/max ratio down to?

 

[unclewebb]

I set the mV boost feature to 150

You can try going even higher than 150. Maybe increasing this to 200 will allow you to reliably use a larger undervolt which can improve your full load temperatures. Using this feature to increase the lower end of the voltage curve can improve stability when undervolting.

so it definitely helped

Thanks for doing some testing. Very few people understand or use the mV Boost feature. I am always trying to sell and promote this useful feature.

When you refer to the 'cache max speed'

The Cache Ratio Min / Max variables control your cache speed. Another thought just popped into my head. Perhaps the default Min ratio of 8 is too low for some CPUs. Perhaps increasing the Min value might help improve stability. You could try increasing the Min to 16 or some other magic number.

As for the Max value, ideally you want to run the cache as fast as possible without it causing a BSOD. For some CPUs, reducing the Max value by 1 or 2 or 3 might be a way to improve stability without reducing performance any meaningful or noticeable amount.

I tried dropping the core and pcache offset to -80 and I get a bsod after 5 mins.

Now that you have found a way to undervolt the cache without it crashing, you can try leaving the cache at -75 mV or -70 mV and try undervolting only the core some more to -100 mV or -125 mV. Is that stable? Does it offer any benefit such as a lower full load temperature compared to leaving these two voltages set equally? Using a slightly larger undervolt on the core compared to the cache used to be a sweet spot for some older Intel CPUs. For some CPUs, staggered undervolt requests do not accomplish anything. The CPU might decide to just ignore the higher core offset request value.

I have never played around with a 13900HX so I cannot say what will work best for your CPU. It is all trial and error. Have fun playing and post anything new you learn.

 

[aby]

You can try going even higher than 150. Maybe increasing this to 200 will allow you to reliably use a larger undervolt which can improve your full load temperatures. Using this feature to increase the lower end of the voltage curve can improve stability when undervolting.


Thanks for doing some testing. Very few people understand or use the mV Boost feature. I am always trying to sell and promote this useful feature.


View attachment 342586

The Cache Ratio Min / Max variables control your cache speed. Another thought just popped into my head. Perhaps the default Min ratio of 8 is too low for some CPUs. Perhaps increasing the Min value might help improve stability. You could try increasing the Min to 16 or some other magic number.

As for the Max value, ideally you want to run the cache as fast as possible without it causing a BSOD. For some CPUs, reducing the Max value by 1 or 2 or 3 might be a way to improve stability without reducing performance any meaningful or noticeable amount.


Now that you have found a way to undervolt the cache without it crashing, you can try leaving the cache at -75 mV or -70 mV and try undervolting only the core some more to -100 mV or -125 mV. Is that stable? Does it offer any benefit such as a lower full load temperature compared to leaving these two voltages set equally? Using a slightly larger undervolt on the core compared to the cache used to be a sweet spot for some older Intel CPUs. For some CPUs, staggered undervolt requests do not accomplish anything. The CPU might decide to just ignore the higher core offset request value.

I have never played around with a 13900HX so I cannot say what will work best for your CPU. It is all trial and error. Have fun playing and post anything new you learn.

No dice. I tried dropping the max ratio to 36 without luck even when I maxed out the icc (dropped it to 220 now). Setting mv boost to 200 didn't help either.

If I set the pcache offset to 0 then I'm stable at a -145.2 uv on the core however for some weird reason. I can't get it anywhere higher than -75.2 if I also uv the pcache. Pic is my current settings.
No idea how gizmosliptech managed to get a uv that high on the core and pcache and be stable.
I should also add: the lenovo legion pro 7i I bought was a refurbished model. Could that impact uv potential on the cpu?

 

[unclewebb]

Undervolt potential is completely random. There are always going to be some CPUs that undervolt much more compared to someone else's CPU. The ability to undervolt is neither good or bad. Undervolting less just means that Intel did a better job of setting the default voltage table so you do not need to undervolt as much.

What is actually better? Leaving the core and cache offsets set to the same values or setting the cache to 0 mV and setting the core to -145.2 mV? Does one or the other give better temperatures or better performance when running Cinebench or 3D Mark?

ThrottleStop lets you set these voltage request values individually but not all CPUs work the same. Sometimes an undervolt will only work correctly if the core and cache offset undervolt are set equally. That is why you need to do some consistent testing to prove if your undervolt request is actually doing something useful. Reducing the Cache Ratio Max to 36 might reduce performance in some apps or games that are heavily cache dependent. Do not sacrifice too much performance if you do not have to.

 

[aby]

Undervolt potential is completely random. There are always going to be some CPUs that undervolt much more compared to someone else's CPU. The ability to undervolt is neither good or bad. Undervolting less just means that Intel did a better job of setting the default voltage table so you do not need to undervolt as much.

What is actually better? Leaving the core and cache offsets set to the same values or setting the cache to 0 mV and setting the core to -145.2 mV? Does one or the other give better temperatures or better performance when running Cinebench or 3D Mark?

ThrottleStop lets you set these voltage request values individually but not all CPUs work the same. Sometimes an undervolt will only work correctly if the core and cache offset undervolt are set equally. That is why you need to do some consistent testing to prove if your undervolt request is actually doing something useful. Reducing the Cache Ratio Max to 36 might reduce performance in some apps or games that are heavily cache dependent. Do not sacrifice too much performance if you do not have to.

-75.2 on both core and pcache gives a slightly higher score on cinebench.
Still it's really weird how I can get a heavy uv on the core to be stable if I don't mess with the pcache at all. I wonder if anybody else (with the same cpu) ends up being the same situation

 

[unclewebb]

Keep in mind that the offset voltage values that ThrottleStop sends to the CPU are only voltage requests. It is impossible to know how much voltage the core or cache are actually getting. With older CPUs it was possible to set the core slider to -1000 mV. If the cache slider was left at 0 mV, the vast majority of that core offset request was ignored by the CPU.

I wonder if anybody else

Most users are too busy playing games. No time to test or share results. Your CPU is unique. It is one of the first ones that I have heard about that needs some mV Boost so you can play around with the P cache offset voltage without crashing. Good job proving that works.

Lenovo has recently released an updated BIOS for your laptop that forces Undervolt Protection to always be enabled with no option in the BIOS that works to turn this off. Do whatever you can to avoid this recent BIOS update. I think the last good BIOS will have the number 42 in its file name. The new one is version 44. There are a few more letters in the BIOS name. I am not 100% sure if that update and BIOS name applies to your laptop model. Just avoid updating the BIOS and you should be OK.

https://www.reddit.com/r/LenovoLegion/comments/1br0s7s
Some people on reddit do not know what they are talking about. Read through all of it to get a better understanding of what is going on.

This next link gets to the point much quicker.
https://www.reddit.com/r/LenovoLegion/comments/1bvpnhu

 

[aby]

Keep in mind that the offset voltage values that ThrottleStop sends to the CPU are only voltage requests. It is impossible to know how much voltage the core or cache are actually getting. With older CPUs it was possible to set the core slider to -1000 mV. If the cache slider was left at 0 mV, the vast majority of that core offset request was ignored by the CPU.


Most users are too busy playing games. No time to test or share results. Your CPU is unique. It is one of the first ones that I have heard about that needs some mV Boost so you can play around with the P cache offset voltage without crashing. Good job proving that works.

Lenovo has recently released an updated BIOS for your laptop that forces Undervolt Protection to always be enabled with no option in the BIOS that works to turn this off. Do whatever you can to avoid this recent BIOS update. I think the last good BIOS will have the number 42 in its file name. The new one is version 44. There are a few more letters in the BIOS name. I am not 100% sure if that update and BIOS name applies to your laptop model. Just avoid updating the BIOS and you should be OK.

https://www.reddit.com/r/LenovoLegion/comments/1br0s7s
Some people on reddit do not know what they are talking about. Read through all of it to get a better understanding of what is going on.

This next link gets to the point much quicker.
https://www.reddit.com/r/LenovoLegion/comments/1bvpnhu

is there a threshold that I shouldn't raise past on the mV boost feature?

 

[unclewebb]

is there a threshold that I shouldn't raise past on the mV boost feature?

You can do whatever you want. If you think any setting in ThrottleStop is going to make your CPU blow up, don't do it. Making any adjustment is always at your own risk and expense.

At default settings, Intel lowers the VID voltage close to the minimum necessary to be stable when running at 800 MHz. When you are using an offset undervolt, this reduces the VID voltage further at 800 MHz. This can easily cause instability or a BSOD if the voltage gets too low. Using mV Boost can help prevent that instability while still allowing you to offset undervolt reliably when your CPU is running at full speed.

The mV Boost value only increases the voltage at the very low end of the MHz curve. If the minimum VID voltage is typically 0.7000 V (700 mV) and you use mV Boost to increase that by 300 mV to 1.0000 V, that is not going to hurt anything. That VID voltage is still well within the Intel spec.

I have only ever tested mV Boost on my 10850K. Normal VID voltage when the CPU is locked to 800 MHz is approximately 0.6900 V. With mV Boost set to the max, 499, the VID voltage increases to 1.1294 at 800 MHz. That is still a lot less than usual. At 5000 MHz the typical VID voltage is about 1.3500. The mV Boost value is not used when the CPU is running at full speed.

The maximum for mV Boost adjustment was deliberately capped to a maximum of 499. I have always tried to make sure that any new feature is well thought out and thoroughly tested. mV Boost should be 100% safe no matter what value you set this to. ThrottleStop has been around for 15 years and continues to have an excellent reputation. I would like to keep that.

 

[icarium_jag]

Bit of an unrelated note, but what might be the reason fo mV Boost option in FIVR windows showing as 'mV Boost @ 5000 MHz'?
I've been fiddling with my mini desktop's undervolt settings, and this voltage boost when the computer is idle/lightly loaded might be just what I need to prevent bsod's on Windows desktop (the undevolt is stable when under heavy load).

My processor is i9-11900KB, a quite obscure Tiger Lake processor only used in Intel NUC Beast Canyon extreme kit. I've been using Throttlestop a lot to improve and finetune this processor, and it has worked great aside from mVBoost @ setting showing this strange MHz number.

 

[unclewebb]

mV Boost @ 5000 MHz

This sometimes happens if VBS is enabled. Post a screenshot of the FIVR window so I can have a look.

 

[icarium_jag]

This sometimes happens if VBS is enabled. Post a screenshot of the FIVR window so I can have a look.

 

[unclewebb]

i9-11900KB, a quite obscure Tiger Lake processor

It is possible that your CPU does not support the mV Boost feature or it might just be reading the MHz value from the CPU incorrectly. Have you tried using mV Boost?

To test if this applies additional voltage at 800 MHz, open the TPL window, check Speed Shift and set Speed Shift Min and Speed Shift Max both to 8 which should lock the CPU to 800 MHz. Have a look at what VID voltage is reported. Open the FIVR window and set mV Boost to 100 or 200. Press Apply and see if the VID voltage value increases.

You can do the same test with the CPU locked to the 45 multiplier. Use the Windows High Performance power plan so there is no interference from Windows. mV Boost should not add any voltage when the CPU is at 4500 MHz. mV Boost should only add voltage when the CPU is at 800 MHz.

 

[icarium_jag]

It is possible that your CPU does not support the mV Boost feature or it might just be reading the MHz value from the CPU incorrectly. Have you tried using mV Boost?

To test if this applies additional voltage at 800 MHz, open the TPL window, check Speed Shift and set Speed Shift Min and Speed Shift Max both to 8 which should lock the CPU to 800 MHz. Have a look at what VID voltage is reported. Open the FIVR window and set mV Boost to 100 or 200. Press Apply and see if the VID voltage value increases.

You can do the same test with the CPU locked to the 45 multiplier. Use the Windows High Performance power plan so there is no interference from Windows. mV Boost should not add any voltage when the CPU is at 4500 MHz. mV Boost should only add voltage when the CPU is at 800 MHz.

Thanks, I have to try this when I'll have some free time on hand. Do you think it likely i9-11900KB wouldn't support the mV Boost feature? The processor is mostly similar to 11th gen h-series processors, 11900h and 11800h being the closest equivalents.

 

[unclewebb]

Do you think it likely i9-11900KB wouldn't support the mV Boost feature?

I have never owned or tested an 11900KB so I do not know what features it has or does not have. You get to be the guinea pig. Hopefully your testing will prove if this works or not. I do not think any of the H series support this feature. The mV Boost box should be locked out in ThrottleStop when running on an H series CPU.

 

[aby]

I have never owned or tested an 11900KB so I do not know what features it has or does not have. You get to be the guinea pig. Hopefully your testing will prove if this works or not. I do not think any of the H series support this feature. The mV Boost box should be locked out in ThrottleStop when running on an H series CPU.

You can do whatever you want. If you think any setting in ThrottleStop is going to make your CPU blow up, don't do it. Making any adjustment is always at your own risk and expense.

At default settings, Intel lowers the VID voltage close to the minimum necessary to be stable when running at 800 MHz. When you are using an offset undervolt, this reduces the VID voltage further at 800 MHz. This can easily cause instability or a BSOD if the voltage gets too low. Using mV Boost can help prevent that instability while still allowing you to offset undervolt reliably when your CPU is running at full speed.

The mV Boost value only increases the voltage at the very low end of the MHz curve. If the minimum VID voltage is typically 0.7000 V (700 mV) and you use mV Boost to increase that by 300 mV to 1.0000 V, that is not going to hurt anything. That VID voltage is still well within the Intel spec.

I have only ever tested mV Boost on my 10850K. Normal VID voltage when the CPU is locked to 800 MHz is approximately 0.6900 V. With mV Boost set to the max, 499, the VID voltage increases to 1.1294 at 800 MHz. That is still a lot less than usual. At 5000 MHz the typical VID voltage is about 1.3500. The mV Boost value is not used when the CPU is running at full speed.

View attachment 342766

The maximum for mV Boost adjustment was deliberately capped to a maximum of 499. I have always tried to make sure that any new feature is well thought out and thoroughly tested. mV Boost should be 100% safe no matter what value you set this to. ThrottleStop has been around for 15 years and continues to have an excellent reputation. I would like to keep that.

I just tried setting the mV Boost to 499 earlier and my cinebench score decreased. What could the reason for that be?
Also

You can do whatever you want. If you think any setting in ThrottleStop is going to make your CPU blow up, don't do it. Making any adjustment is always at your own risk and expense.

At default settings, Intel lowers the VID voltage close to the minimum necessary to be stable when running at 800 MHz. When you are using an offset undervolt, this reduces the VID voltage further at 800 MHz. This can easily cause instability or a BSOD if the voltage gets too low. Using mV Boost can help prevent that instability while still allowing you to offset undervolt reliably when your CPU is running at full speed.

The mV Boost value only increases the voltage at the very low end of the MHz curve. If the minimum VID voltage is typically 0.7000 V (700 mV) and you use mV Boost to increase that by 300 mV to 1.0000 V, that is not going to hurt anything. That VID voltage is still well within the Intel spec.

I have only ever tested mV Boost on my 10850K. Normal VID voltage when the CPU is locked to 800 MHz is approximately 0.6900 V. With mV Boost set to the max, 499, the VID voltage increases to 1.1294 at 800 MHz. That is still a lot less than usual. At 5000 MHz the typical VID voltage is about 1.3500. The mV Boost value is not used when the CPU is running at full speed.

View attachment 342766

The maximum for mV Boost adjustment was deliberately capped to a maximum of 499. I have always tried to make sure that any new feature is well thought out and thoroughly tested. mV Boost should be 100% safe no matter what value you set this to. ThrottleStop has been around for 15 years and continues to have an excellent reputation. I would like to keep that.

I just tried setting mV boost to 499 and my cinebench score actually decreased. What could the reason for that be? Currently testing 350 on the mV boost and a -115.2 UV

Also sorry for the newbie question but could you explain what Thermal Velocity Boost & V-max Stress do and if they should be enabled or not? Thank you.

 

[unclewebb]

What could the reason for that be?

Cinebench scores vary at the best of times, especially if a CPU is thermal throttling, power limit throttling or EDP throttling. It takes some work to get consistent results if you are living on the edge of throttling. Turn on the Log File option when testing and attach a log file so I can see if your CPU is throttling during Cinebench.

Checking the Thermal Velocity Boost box tells the CPU to throttle 100 MHz when it reaches approximately 70°C. Checking V-Max Stress tells the CPU to throttle at high voltage. I clear both those boxes. I hate throttling.

If using a high mV Boost value does not improve stability or offset undervolt further then you do not need to use a high value.

 

[aby]

Cinebench scores vary at the best of times, especially if a CPU is thermal throttling, power limit throttling or EDP throttling. It takes some work to get consistent results if you are living on the edge of throttling. Turn on the Log File option when testing and attach a log file so I can see if your CPU is throttling during Cinebench.

Checking the Thermal Velocity Boost box tells the CPU to throttle 100 MHz when it reaches approximately 70°C. Checking V-Max Stress tells the CPU to throttle at high voltage. I clear both those boxes. I hate throttling.

If using a high mV Boost value does not improve stability or offset undervolt further then you do not need to use a high value.

Attached is the log file.
Another thing that's a bit unrelated: Dropping the 'max' value in the Cache min/max ratio from 8:48 to 8:36 significantly increases my cpu benchmark score in both cinebench and timespy. Doesn't a higher 'max' value in the cache ratio typically result in greater performance? Or do I have this backwards?

 

[unclewebb]

The log file shows constant EDP throttling. This is reducing your CPU speed to only 3600 MHz. It is going to be impossible to get consistent or meaningful benchmark results when your CPU is severely throttling like this.

If slowing your cache gives better performance, it is probably because the power being saved allows your CPU cores to run faster.

Watch ThrottleStop with Limit Reasons open when running Cinebench. The throttling should be obvious.

 

[aby]

The log file shows constant EDP throttling. This is reducing your CPU speed to only 3600 MHz. It is going to be impossible to get consistent or meaningful benchmark results when your CPU is severely throttling like this.

If slowing your cache gives better performance, it is probably because the power being saved allows your CPU cores to run faster.

Watch ThrottleStop with Limit Reasons open when running Cinebench. The throttling should be obvious.

EDP Other does flash red. HWinfo however shows 'No' on Core Thermal Throttling when running Cinebench. Why the discrepancy?

I don't see any lock options ticked in TPL either

Sorry about the double post UW
Even when deleting the TS ini file and resetting everything to stock it still shows EDP Other in Red when running Cinebench. What's going on?

EDP Other does flash red. HWinfo however shows 'No' on Core Thermal Throttling when running Cinebench. Why the discrepancy?

I don't see any lock options ticked in TPL either

Sorry about the double post UW
Even when deleting the TS ini file and resetting everything to stock it still shows EDP Other in Red when running Cinebench. What's going on?

Update I've figured out the issue by asking on the Lenovo Legion discord:

At stock Iccmax you will get EDP throttling even with zero voltage offsets on the 13900hx. Setting Iccmax to the highest value fixes the issue but greatly increases my thermals to where it throttles and I get the indicator in limits. The laptop I bought was a refurb model so it's very much likely the paste job was done poorly. I ordered liquid metal which should arrive next week so I can repaste the cpu. I'll post another update after the paste job and running ts and cinebench again.

 

[aby]

Sorry for the late reply
Getting up to 35k on cinebench now after repasting. 8-48 cache ratio and mVboost @ 330. Iccmax @ 260
See attached log file.
Everything good, UW? pl2 and vr current are on red when running cinebench however.