Thinkpad T480 - Throttlestop questions

[chunky_lafunga]

Hi all (and particularly @unclewebb )

I've been using Throttlestop for over 18 months now and have done a lot of reading during that time to get to a comfortable range of settings. Thanks unclewebb for your work and previous responses on other forums.

Laptop

Laptop: Thinkpad T480
CPU: i5-8250u
GPU: MX150
Windows power options: 'High performance'
Windows version: Windows 10 version 2004 / OS Build 19041.630 / Experience 120.2212.31.0

I have a couple of questions at this point I'm hoping to get help on;

1) PROCHOT incident

I've recently started gaming more (Team Fortress 2). In the last week during play, the computer froze (no BSOD) and became very slow to respond. I managed to exit out of TF2 and opened up Throttlestop to see that 'PROCHOT 97c' was ticked. When I unticked it, it would get ticked back on a few seconds later.

A reboot resolved this, however as I said it's happened twice in one week so I'm wondering what happened to cause this PROCHOT box to keep ticking itself on. I've attached my logs, it happened around 1:56pm-1:58pm from memory, sorry I can't be more specific.

2) GPU undervolting

Currently I have the Intel GPU undervolted to -50.8 (and iGPU unslice box is greyed out by default, although I can change that with the 'unlock adjustable voltage' box). Would you suggest that I;

1. just change the Intel GPU back top 0, because it's not worthwhile undervolting? Or
2. should I unlock adjustable voltage for iGPU and change that to -50.8 too?

3) CPU undervolting

Everything I had previously read suggested that CPU Core and CPU Cache should be undervolted by the same amount. However more recently I read that CPU Core and CPU Cache should be undervolted to different values. Could you please clarify for me?

4) Upgrading Throttlestop version

I'm currently on version 8.74 and can see the latest stable is 9.2. What is the best suggested way to upgrade and retain all my existing settings? I am also utilizing RwDrv.sys so want to make sure I upgrade correctly.

Sharing my existing settings with you below;

 

[unclewebb]

   DATE       TIME    MULTI   C0%   CKMOD  CHIPM   BAT_mW  TEMP  GPU     VID   POWER
2020-11-19  13:56:00  33.89   29.2  100.0  100.0        0   81    69   1.0100   23.8
2020-11-19  13:56:01  33.83   33.3  100.0  100.0        0   79    69   1.0184   25.5
2020-11-19  13:56:02  33.85   24.4  100.0  100.0        0   78    69   1.0447   22.6
2020-11-19  13:56:04   6.93   49.3  100.0  100.0        0   71    68   0.4813    5.5
2020-11-19  13:56:05   4.00   49.9  100.0  100.0        0   71    68   0.4867    3.8
2020-11-19  13:56:06   4.00   53.1  100.0  100.0        0   70    68   0.4818    3.7
2020-11-19  13:56:06   4.00   49.5  100.0  100.0        0   69    65   0.4873    3.7
2020-11-19  13:56:07   4.00   49.2  100.0  100.0        0   69    65   0.4873    3.7
2020-11-19  13:56:08   4.00   50.5  100.0  100.0        0   68    65   0.4867    3.7
2020-11-19  13:56:09   4.00   48.9  100.0  100.0        0   67    63   0.4872    3.7

1) PROCHOT incident

The log file shows that at 13:56 (1:56 PM) the CPU multiplier goes from 34 down to 4. That means your CPU instantly started running at about 12% of its rated speed. You are definitely going to feel a punch in the gut like that. The log file also shows that there was no reason for this. Your temperatures, CPU and GPU were great. No need for this excessive slow down.

4) Upgrading ThrottleStop version

You should have upgraded to ThrottleStop 9.2 a while ago. It is easy. Download the new version and copy the new ThrottleStop.exe into your ThrottleStop folder. Done. Upgrade complete. You can delete the RwDrv.sys file and all of the WinRing0 driver files. These two drivers have been replaced with a safer driver to keep the antivirus folks happy.

The new ThrottleStop version has a new feature to deal with the excessive throttling that some Lenovo laptops are using. Open the Options window and on the right side is a value called PROCHOT Offset. Intel sets this to 0 by default. Your laptop set this to 3. That is OK. This value controls when your laptop will start to thermal throttle. Intel default thermal throttling temperature is 100°C. When this is set to 3, that reduces the thermal throttling temperature down to 97°C instead of 100°C.

The problem is that some Lenovo laptops are randomly setting the offset value to 35. This forces the CPU to go into full thermal throttling mode even though the CPU temperature is only 65°C (100°C - 35 offset). This is absolutely nuts. To prevent this from happening, set PROCHOT Offset to 3 and then check the Lock PROCHOT Offset box.

http://imgur.com/aD9WsML

Press OK. If you open the Options window back up, you should see a small Lock icon near this setting. That means the CPU has been locked to the proper thermal throttling temperature until the next reboot. This should take care of your premature thermal throttling problem.

It might be Lenovo Vantage or some other Lenovo software that is botching things.

2) GPU undervolting

If you are going to undervolt the Intel GPU then yes, you need to undervolt both the Intel GPU and the iGPU Unslice, probably equally. You have an Nvidia GPU so there is rarely going to be a significant load on the Intel GPU. The Intel GPU is not used at all when gaming so my opinion is that there is no point undervolting it. There is nothing to be gained and you might lose overall stability.

3) CPU undervolting

I have not done any voltage testing of 8th Gen U CPUs. The only thing I can recommend is give it a try. Use Cinebench R20 when testing.

MAXON Cinebench R20.0 Download

CINEBENCH is a real-world cross platform test suite that evaluates your computer's performance capabilities. CINEBENCH is based on MAXON's award-winn

www.techpowerup.com

See if setting the core to a bigger offset improves temperature or performance.

-100 mV cache and -100 mV core for a baseline
-100 mV cache and -125 mV core
-100 mV cache and -150 mV core

etc., etc.

With the H series CPUs, there is an improvement up to about -225 mV for the core and the CPU is still 100% stable. Definitely worth doing some testing.

Once you get the new TS version installed, turn on the Log File option and go play your game again. The new version includes more information in the log file so it can help if you are still having problems. Thanks for posting lots of pics and data. Anyone that does that deserves a thorough reply.

 

[chunky_lafunga]

You should have upgraded to ThrottleStop 9.2 a while ago. It is easy. Download the new version and copy the new ThrottleStop.exe into your ThrottleStop folder. Done. Upgrade complete. You can delete the RwDrv.sys file and all of the WinRing0 driver files. These two drivers have been replaced with a safer driver to keep the antivirus folks happy.

Done now, thanks for that.

The new ThrottleStop version has a new feature to deal with the excessive throttling that some Lenovo laptops are using. Open the Options window and on the right side is a value called PROCHOT Offset. Intel sets this to 0 by default. Your laptop set this to 3. That is OK. This value controls when your laptop will start to thermal throttle. Intel default thermal throttling temperature is 100°C. When this is set to 3, that reduces the thermal throttling temperature down to 97°C instead of 100°C.

Interesting

If you are going to undervolt the Intel GPU then yes, you need to undervolt both the Intel GPU and the iGPU Unslice, probably equally. You have an Nvidia GPU so there is rarely going to be a significant load on the Intel GPU. The Intel GPU is not used at all when gaming so my opinion is that there is no point undervolting it. There is nothing to be gained and you might lose overall stability.

Cool, I decided to leave Intel GPU alone and keep at 0 offset in that case.

See if setting the core to a bigger offset improves temperature or performance.

-100 mV cache and -100 mV core for a baseline
-100 mV cache and -125 mV core
-100 mV cache and -150 mV core

etc., etc.

So I went ahead and did the following;

• Upgraded to 9.2
• Locked PROCHOT Offset to 3
• Changed Cache offset to -100mV (-107 was becoming unstable since my original post)
• Changed Core offset to -200mV (I did it gradually from -100 --> -150 --> -200 testing each change on R20 and R23)

Here are the results

• CB R20 results: increased by ~15
• CB R23 results: Increased by ~20
• In-game temp reduced by ~11*C on average
• In-game power reduced by ~5w on average (keep in mind my Long/Short is 25w/29w so 5w reduction is huge.

Stability seems fine for now. You think I should try pushing to -225mV Core offset? What's the worst that could happen?

 

[unclewebb]

What's the worst that could happen?

Your laptop catches on fire and you burn down half the houses on your block. This is not likely to happen.

Your laptop might crash. If it does, reboot and reduce your undervolt.

At some point, when you keep pushing the core offset more and more, you will reach a point where there is no further improvement in performance or temperatures. For the H series CPUs, this happens somewhere around -225 mV core. You could request -500 mV and the CPU will automatically ignore the excess that it does not understand. The H series never seem to crash unless the cache has been set too high.

No fear. Go for -225 mV.

The drop in temps is fantastic. It takes a lot to get my attention. I have to admit, I am impressed. Hopefully you run a log file and I do not have to see your CPU limping along at 400 MHz. Not sure what Lenovo engineering was thinking when they dreamed up that feature. Kind of like creating a zero emissions vehicle by removing the engine.

 

[chunky_lafunga]

Your laptop catches on fire and you burn down half the houses on your block. This is not likely to happen.

Your laptop might crash. If it does, reboot and reduce your undervolt.

At some point, when you keep pushing the core offset more and more, you will reach a point where there is no further improvement in performance or temperatures. For the H series CPUs, this happens somewhere around -225 mV core. You could request -500 mV and the CPU will automatically ignore the excess that it does not understand. The H series never seem to crash unless the cache has been set too high.

No fear. Go for -225 mV.

Cool, I will give it a try shortly!

The drop in temps is fantastic. It takes a lot to get my attention. I have to admit, I am impressed. Hopefully you run a log file and I do not have to see your CPU limping along at 400 MHz. Not sure what Lenovo engineering was thinking when they dreamed up that feature. Kind of like creating a zero emissions vehicle by removing the engine.

Attaching my logs. It was in the wee hours of the morning (you can see I when to bed around 3am!) when I was doing all my adjusting + testing, so in the logs you may identify;

- Numerous Cinebench testing
- Testing gaming (TF2) stability

Hopefully you can tell one from the other, but logs are attached anyway for your perusal

 

[unclewebb]

@chunky_lafunga - I always like looking at some numbers. You could try bumping up your turbo power limits. Your cooling system is doing OK. Maybe 30W / 35W for long and short might be doable.

 

[chunky_lafunga]

@chunky_lafunga - I always like looking at some numbers. You could try bumping up your turbo power limits. Your cooling system is doing OK. Maybe 30W / 35W for long and short might be doable.

So I went ahead and changed to -215mV on Core. Ran a couple of Cinebench tests, nothing too different to -200 so left it at -215mV for the time being.

I was then using the laptop for general browsing and word/excel for about 30-40 minutes. No problems.

Decided to turn my EPP from 0 to 64 for no reason, as soon as I did that I got a BSOD.

Why that could be happening?

(logs attached, happened at 10:29am)

 

[unclewebb]

Why that could be happening?

Speed Shift allows the CPU to rapidly change speed, hundreds of times per second. When EPP is set to 0, the CPU runs at a constant high speed. It usually takes slightly more voltage for a CPU to be 100% stable when Speed Shift is enabled.

 

[chunky_lafunga]

Speed Shift allows the CPU to rapidly change speed, hundreds of times per second. When EPP is set to 0, the CPU runs at a constant high speed. It usually takes slightly more voltage for a CPU to be 100% stable when Speed Shift is enabled.

Got it. So when EPP is set to 0, Speed Shift is effectively "Off'?

And when it's anything other than 0, it's 'On' and I may need to reduce my undervolt if not stable above 0?

 

[unclewebb]

Speed Shift is effectively "Off'

That is correct. It would be best to find a voltage setting that is Speed Shift EPP compatible without a BSOD. Your voltage settings are too close to the edge of stability so back up a little.

 

[chunky_lafunga]

That is correct. It would be best to find a voltage setting that is Speed Shift EPP compatible without a BSOD. Your voltage settings are too close to the edge of stability so back up a little.

So EPP really affects my results!

With EPP 'On' (e.g. 64), I can't get stable performance at anything beyond -99mV on Core and Cache! BSOD all over the place!

I know you suggested finding a suitable undervolt that gives me stability with EPP 'On', but I'd always used EPP = 0 (at least while plugged in). I was also only able to achieve my -200mV Core undervolt with EPP = 0.

So I guess the question, and I know this is all personal, but do I really need to find a EPP>0 undervolt, or not really, since I prefer 0 while plugged in?

Currently what I'm thinking is:

• Plugged in: (tested and stable)
  • -200mV Core / -99mV Cache
  • EPP = 0
  • 'High Performance' enabled
• Battery: (tested and stable)
  • -99mV Core / -99mV Cache
  • EPP = 180
  • 'Balanced' enabled

 

[unclewebb]

I can't get stable performance at anything beyond -99mV on Core and Cache!

This is usually a sign that you have set the cache too high. I have a similar 8th Gen U and the cache is not stable at anything beyond -80 mV. If you reduce the cache undervolt a little (-90 mV or -80 mV), that might improve stability and allow you to increase the core significantly. Some TS Bench tests, light load and full load, can help uncover any cache errors. If you stay with -99 mV, make sure the TS Bench is not reporting any errors.

I prefer 0 while plugged in

Same here. Every time someone wants to run their laptop at full speed when lightly loaded, someone jumps into the conversation to say what a bad idea it is. Of course they have never done any testing but they heard that this was bad 10 years ago so it must still be bad. If you do some testing, you will probably find that it is not bad at all. Here is my fav test. Huge change in CPU speed and huge change in VID voltage made zero difference to power consumption or CPU temperature.

Why is that? What old school users do not realize is that when a mobile CPU is idle, the individual cores can spend 99% of their time in the low power C7 state. That means they are already at 0 MHz and 0 volts. Whether the rest of the CPU is running fast or slow makes an insignificant difference to overall power consumption or heat. If you want to run your CPU fast all the time, go ahead. It is OK to do that. I prefer a responsive CPU.

So EPP really affects my results!

Thanks for sharing your results. Use whatever combination of settings that work best for you.

One last trick you can use is you can adjust the Speed Shift Min value in the TPL window higher. The minimum CPU speed is usually 400 MHz. Perhaps if you change Speed Shift Min to 8 or higher, that can help prevent the CPU from dropping so low which might allow you to maintain your -99 mV cache voltage without it crashing. ThrottleStop is not recommended for anyone with OCD. Too many variables to play with. Testing can consume a lot of time!

 

[chunky_lafunga]

This is usually a sign that you have set the cache too high. I have a similar 8th Gen U and the cache is not stable at anything beyond -80 mV. If you reduce the cache undervolt a little (-90 mV or -80 mV), that might improve stability and allow you to increase the core significantly. Some TS Bench tests, light load and full load, can help uncover any cache errors. If you stay with -99 mV, make sure the TS Bench is not reporting any errors.

I think what you're saying might also prove true for me too. I'm happy with a -90 or -80 cCche undervolt if it allows me to go back up to that -200 range on Core, and potentially keep those gaming temps down by 10*c!

I've never really known how to use TS Bench. Can you suggest how I could use it in my scenario (what priority, threads, size, MHz)?

 

[unclewebb]

TS Bench

For priority I would leave it at Normal. After that, change whatever.

Setting Threads to 1 or 2 is a good light load test. Use 8 Threads for a full load test.

Selecting a bigger Size allows the test to run a longer amount of time. You can push the Stop button or Stop logo at any time to stop a test.

The Random option might be a useful test for you. It uses Speed Shift Technology to continuously vary the CPU speed.

No matter what test you pick, you need to be able to finish it without any errors being reported.
You will see the number of errors reported if your voltage is not stable.

 

[chunky_lafunga]

For priority I would leave it at Normal. After that, change whatever.

Setting Threads to 1 or 2 is a good light load test. Use 8 Threads for a full load test.

Selecting a bigger Size allows the test to run a longer amount of time. You can push the Stop button or Stop logo at any time to stop a test.

The Random option might be a useful test for you. It uses Speed Shift Technology to continuously vary the CPU speed.

No matter what test you pick, you need to be able to finish it without any errors being reported.
You will see the number of errors reported if your voltage is not stable.

View attachment 176400

Got it.

So I tried a test but I am getting the time taken (17.421), rather than the # of errors in the area where you've got '4 Errors'.

Do I need to change something to see the # of errors?

 

[unclewebb]

Do I need to change something

No. It will only report errors if there are errors. If there are 0 errors it will report the time it took to complete the test.

Everything is OK.

 

[chunky_lafunga]

No. It will only report errors if there are errors. If there are 0 errors it will report the time it took to complete the test.

Everything is OK.

Awesome, I will keep Cache at around -90, and keep pushing Core and testing using TS Bench on Random.

Question, I notice if I hold down the < buttons on the FIVR page to change the undervolt (rather than dragging the slider), it stops at numbers like -90.8, -105.5, -120.1, rather than whole numbers like -90, -105, -120. Is there any particular logic behind that?

 

[unclewebb]

Is there any particular logic

No logic. It just adjusts it a fixed number of steps before stopping. Part of the programming language. I do not like that either but too lazy to reinvent the wheel.

 

[chunky_lafunga]

No logic. It just adjusts it a fixed number of steps before stopping. Part of the programming language. I do not like that either but too lazy to reinvent the wheel.

Makes sense!

A unrelated question on C states; I'm reading and learning your previous replies to other users on C states and what we should expect to see / be aiming for. I'm trying to get a baseline of where I am and what I may need to investigate / improve further.

Would you mind having a look at my C states below and let me know what you think, or any observations?

Current configuration

• Below is at idle
• I hardly ever see PKG Power drop below 3.8w

 

[unclewebb]

C states

Windows 10 and ThrottleStop are very efficient when set up correctly. Individual cores need to spend hardly any time in the C0 state working on the background tasks. The cores can spend over 99% of their time in the low power Core C7 state when a computer is idle.

Your results show that you have a significant amount of stuff running in the background. You have almost one entire core being wasted processing background tasks. Open the Task Manager, click on the Details tab and click on the CPU heading to organize the running tasks by CPU usage. See if you can find anything running in the background that really does not need to be running.

Autoruns can help you manage what tasks are automatically starting on your computer when Windows boots up.

Autoruns - Sysinternals

See what programs are configured to startup automatically when your system boots and you login.

docs.microsoft.com

O&O Shutup10 is another good tool to keep Windows 10 in check.

O&O ShutUp10++ - O&O Software GmbH

www.oo-software.com

 

[chunky_lafunga]

Windows 10 and ThrottleStop are very efficient when set up correctly. Individual cores need to spend hardly any time in the C0 state working on the background tasks. The cores can spend over 99% of their time in the low power Core C7 state when a computer is idle.

Your results show that you have a significant amount of stuff running in the background. You have almost one entire core being wasted processing background tasks. Open the Task Manager, click on the Details tab and click on the CPU heading to organize the running tasks by CPU usage. See if you can find anything running in the background that really does not need to be running.

Autoruns can help you manage what tasks are automatically starting on your computer when Windows boots up.

Autoruns - Sysinternals

See what programs are configured to startup automatically when your system boots and you login.

docs.microsoft.com

O&O Shutup10 is another good tool to keep Windows 10 in check.

O&O ShutUp10++ - O&O Software GmbH

www.oo-software.com

I'll certainly give Autoruns and Shutup10 a try.

In the mean time I had a look at Task Manager > Details.

To be honest the only concerning item is ironically 'System Idle Process' which appears to be using the most processor utilization!

 

[unclewebb]

System Idle Process

When your computer is idle, System Idle Process should show 99%. That is what you want to see. This data is telling you that your computer is idle.

Maybe Malwarebytes is keeping your CPU active. If you are curious, temporarily uninstall it and see if your C0 and C7 data changes significantly. I try to keep away from the seedy side of the internet so I have not used Malwarebytes or had any problems in years.

You can also try opening up the Resource Monitor. It can keep track of average CPU usage over time so might make it more obvious what is hard at work in the background on your computer.

 

[chunky_lafunga]

When your computer is idle, System Idle Process should show 99%. That is what you want to see. This data is telling you that your computer is idle.

Maybe Malwarebytes is keeping your CPU active. If you are curious, temporarily uninstall it and see if your C0 and C7 data changes significantly. I try to keep away from the seedy side of the internet so I have not used Malwarebytes or had any problems in years.

You can also try opening up the Resource Monitor. It can keep track of average CPU usage over time so might make it more obvious what is hard at work in the background on your computer.

@unclewebb, just revisiting this after a few months.

Why do I see two drastically different C7% states? Both circled in red below;

It would be great if you could explain this a bit more.

 

[unclewebb]

Why do I see two drastically different C7% states?

Good question. The table at the top shows the amount of time the individual cores are spending in the low power C7 state. The table at the bottom shows when the entire CPU package enters one of the low power package C states.

The problem is that many manufacturers are releasing laptops and the package C states are broken or have been deliberately disabled. Some do this to improve their disk benchmark scores or to cover up some other problem.

An 8th Gen U series processor should have no problem at all entering the package C8 state. Your screenshot shows that for your laptop, this is not the case. It can be something simple like one poorly written driver that is preventing this. These problems are very difficult for end users to track down. It is up to the manufacturers to put in some more effort. A laptop that is not able to enter package C8 when running on battery power is going to consume more power and will need to be plugged back in sooner compared to the same laptop with fully working package C states.

Here is an example of the exact same laptop, one with the package C states working properly and one without.
ThrottleStop works correctly. Many laptops do not.

Try using the latest version of ThrottleStop.

ThrottleStop 9.2.9

ThrottleStop 9.2.9 https://www.techpowerup.com/download/techpowerup-throttlestop/ New Features - added 10850K / 10900K support including a new Turbo Group access window. - updated the TS Bench and the C State window for the 10 core CPUs. - enabled Limit Reasons support for Comet Lake CPUs. -...

www.techpowerup.com

It has a new feature where it will report if your CPU is going into package C8, C9 or C10 states when in standby / sleep mode. For connected standby to work properly, use ThrottleStop to switch from the Windows High Performance power plan to the Windows Balanced power plan. My desktop computer does not use any of these deeper package C states when running or when in sleep mode. Package C2 is as low as it goes.

Here is an older version of ThrottleStop showing that connected standby is working correctly. The CPU is spending the vast majority of time in package C10. This saves power when your computer is on battery power but is not being used. Many users have infinite problems with connected standby not working correctly.

 

[chunky_lafunga]

Good question. The table at the top shows the amount of time the individual cores are spending in the low power C7 state. The table at the bottom shows when the entire CPU package enters one of the low power package C states.

That's pretty clear, thanks!

Although one question; if you look at my screenshot, you'll see that all 4 cores are 90%+ in C7 in the top table. So wouldn't that imply the entire CPU package is 90%+ in C7 too?

Try using the latest version of ThrottleStop.

ThrottleStop 9.2.9

ThrottleStop 9.2.9 https://www.techpowerup.com/download/techpowerup-throttlestop/ New Features - added 10850K / 10900K support including a new Turbo Group access window. - updated the TS Bench and the C State window for the 10 core CPUs. - enabled Limit Reasons support for Comet Lake CPUs. -...

www.techpowerup.com

It has a new feature where it will report if your CPU is going into package C8, C9 or C10 states when in standby / sleep mode. For connected standby to work properly, use ThrottleStop to switch from the Windows High Performance power plan to the Windows Balanced power plan. My desktop computer does not use any of these deeper package C states when running or when in sleep mode. Package C2 is as low as it goes.

View attachment 182241

Here is an older version of ThrottleStop showing that connected standby is working correctly. The CPU is spending the vast majority of time in package C10. This saves power when your computer is on battery power but is not being used. Many users have infinite problems with connected standby not working correctly.

Great to see that 9.2.9 is out, I went ahead and downloaded it;

I'll put it into 'Balanced' via TS and see what happens over time and report back.