ThrottleStop undervolt is great, how to make it even better (Lenovo T480 Intel Core i5-8350U @1.9ghz)

[Natcuber]

I started using Throttlestop a few days ago, and it's been awesome! My R23 Cinebench multi-core scores went from around 3300 to 3900! I'm just wondering what else I can improve to get even more of my cpu. Also, is it worth undervolting my iGPU since I don't have a dGPU? All tips are appreciated! This is what I have noticed from looking while running TSBench so far:

1. Before I started undervolting, my peak wattage was 33w, but now it only reaches 29
2. My PKG power usually goes like this in TSBench 960M: 3.6ghz @29w for until about 9 seconds, then down to 3.3-3.2ghz @23-24w for the rest of the duration. I have the power limit controls set at 33, but it never reaches that. Also, my cpu does reach its max temp, but doesn't seem reduce its speed (temps are stable at 97)
3. After somewhere around -125mV of core undervolt, there doesn't seem to be any more reduction in voltage in VID. I kept the cache undervolt at -100mV cuz it'll crash
4. I noticed that my clock speeds are heavily affected by the current CPU temp when I start the bench. If I start at about 48c or higher for the 960M TSBench, then instead of the speeds listed in No.2, it will drop down to 3.1-3ghz.


Here are my current settings:

 

[Lew Zealand]

First off, it sounds like you've already done 95% of what you can with TS so you're in good shape.

If you're gaming on that iGPU (sorry if you are) then undervolting it will certainly help in any gaming sutiation. I used to do this on my NUCs when they were my "gaming" PCs (LOL). However for everyday use an iGPU undervolt won't do much of anything. If you're using QuickSync to convert videos it will probably help keep temps down too, but QS taxes the GPU so really test that UV to make sure it's stable.

I recently started testing Throttlestop in my i5-8259u NUC again and you'll need to undervolt both the GPU and the iGPU Unslice to see any actual voltage reduction. It's easiest to just undervolt both by the same amount. I got to about -0.080v before chickening out but all silicon tolerates this differently. For instance, I can only UV my CPU by about -0.090v to remain stable while you seem able to go lower.

 

[unclewebb]

is it worth undervolting my iGPU

Try undervolting the Intel GPU and iGPU Unslice equally. Only you can decide if this is worth doing or not. If you have a separate Nvidia GPU then undervolting the Intel GPU will probably not make much of a difference. Here is a -75 mV undervolt for the Intel GPU and iGPU Unslice.

but doesn't seem reduce its speed

Intel thermal throttling is subtle. CPUs are designed to run as fast as possible. If your CPU reaches 97°C it will slow down. Do not expect to see a big drop in MHz if you are right on the edge of thermal throttling. The CPU is literally adjusting its speed hundreds of times per second during a thermal throttling episode. The goal is always maximum performance without the CPU exceeding the thermal throttling temperature. Turn on the ThrottleStop Log File option so you have a record of your CPU performance. It will show any thermal throttling and it will show how much the MHz declines. ThrottleStop is one of the few monitoring tools that can accurately track this data.

somewhere around -125mV of core undervolt, there doesn't seem to be any more reduction in voltage in VID

Most Intel CPUs sync the core and cache voltages internally. Using slightly more core offset voltage compared to the cache offset voltage can help when running some software like Cinebench but you will reach a point where going further will make no difference. If you reached that point of diminishing returns at -125 mV for the core, why do you have the core offset at -229 mV? That big of a difference is probably not accomplishing anything.

If I start at about 48c or higher for the 960M TSBench, then instead of the speeds listed in No.2, it will drop down to 3.1-3ghz.

If you start a benchmark when your CPU is already warm, you will have less thermal headroom. It will heat up and start thermal throttling sooner compared to if you run the same benchmark when the CPU is cold. That seems like common sense when you think about it.

I prefer checking the MMIO Lock box and I like to set IccMax for both the core and the cache to the max, 255.75.

 

[Natcuber]

Try undervolting the Intel GPU and iGPU Unslice equally. Only you can decide if this is worth doing or not. If you have a separate Nvidia GPU then undervolting the Intel GPU will probably not make much of a difference. Here is a -75 mV undervolt for the Intel GPU and iGPU Unslice.

View attachment 356411


Intel thermal throttling is subtle. CPUs are designed to run as fast as possible. If your CPU reaches 97°C it will slow down. Do not expect to see a big drop in MHz if you are right on the edge of thermal throttling. The CPU is literally adjusting its speed hundreds of times per second during a thermal throttling episode. The goal is always maximum performance without the CPU exceeding the thermal throttling temperature. Turn on the ThrottleStop Log File option so you have a record of your CPU performance. It will show any thermal throttling and it will show how much the MHz declines. ThrottleStop is one of the few monitoring tools that can accurately track this data.


Most Intel CPUs sync the core and cache voltages internally. Using slightly more core offset voltage compared to the cache offset voltage can help when running some software like Cinebench but you will reach a point where going further will make no difference. If you reached that point of diminishing returns at -125 mV for the core, why do you have the core offset at -229 mV? That big of a difference is probably not accomplishing anything.


If you start a benchmark when your CPU is already warm, you will have less thermal headroom. It will heat up and start thermal throttling sooner compared to if you run the same benchmark when the CPU is cold. That seems like common sense when you think about it.

I prefer checking the MMIO Lock box and I like to set IccMax for both the core and the cache to the max, 255.75.

When I do the benches at slightly higher temperatures, the slowdowns I mention happen before it hits the PROCHOT temp. It seems to me as if it would choose a lower clock for the rest of the bench just because it started little warmer. Also does TSBench bench gpus? Otherwise I'll use Geekbench to test my iGPU undervolting.

 

[unclewebb]

The only purpose of the TS Bench is so you have a convenient tool to put a load on the CPU cores and create some heat. I would not use the timed results to compare to anything. It does not load the GPU so find another benchmark for that.

the slowdowns I mention happen before it hits the PROCHOT temp.

Check the Log File box when testing and attach a log file so I can have a look. There is always a reason for throttling. Did you check the MMIO Lock box like I suggested? If the CPU is not thermal throttling then it might be power limit throttling or current limit throttling. I can spend all night guessing but looking at some log file data is a lot easier.

Did you check PROCHOT Offset and Lock PROCHOT Offset in the Options window? Some Lenovo laptops can randomly change the thermal throttling temperature. Checking those two boxes can prevent that from happening.

 

[Natcuber]

I've checked the Lock MMIO and two PROCHOT settings you suggested. I also upped the ICCMax of the core and cache to the max. I did 3 benches, first two are 960M, and the third is 120M. I also did a 7680M a while after that.
Some things to note:
1. I switch my power profiles to a battery saver profile with 200 Speed Shift EPP in between benches so that my CPU will cool faster.
2. I've also noticed that the MMIO values have disappeared for some reason.
3. For some reason my CPU doesn't hit the PROCHOT temp of 97 the first 3 benches, but the red PROCHOT word turns up as you can see.
4. Thermal throttling on the last bench keeps it a more or less stable 3.05ghz at 97c.
I'm also wondering, is it important to turn of memory integrity in BIOS? I believe I still have it enabled.

Edit: I did one more TSBench

 

[unclewebb]

the MMIO values have disappeared for some reason

Checking the MMIO Lock box eliminates the MMIO power limits. Those power limits are no longer in control of your CPU after checking that box. Intel CPUs do not need multiple sets of redundant turbo power limits all fighting against each other.

When More Data is not checked, ThrottleStop will sample the CPU core and package temperatures only one time per second. When the CPU temperature is rapidly changing, it is possible to trigger thermal throttling without ThrottleStop recording the peak core temperature in its monitoring table. If you clear the PROCHOT 97°C flag before you start testing and this flag lights up red during a test, that confirms that the hottest spot on the core definitely reached 97°C and the CPU really was thermal throttling. It might have only reached this temperature for a few milliseconds. When this happens, the log data might not show 97°C as the max. If the log file shows TEMP, the CPU was thermal throttling and some temperature sensor on the CPU must have reported 97°C at some time during the one second monitoring interval.

When you check the More Data box, ThrottleStop will sample the CPU temperature sensors approximately 8 times per second. This might help you "catch" the CPU reaching that magical 97°C temperature. Beyond that, there is no practical reason to use the More Data feature. It creates way "too much information".

During your log file testing, the early tests show that almost all of the throttling is being caused by the PL1 and PL2 power limits. This seems to happen when PL2 is at 29W. PL1 throttling seems to get triggered at approximately 21.7W. During the longer test at the end, PL1 throttling changes to TEMP thermal throttling. This happens when the CPU starts running at a consistent 97°C.

There is a third unique set of turbo power limits that only some computers use. This third set is managed by an embedded controller (EC). It looks like Lenovo has programmed these limits to lower values than what ThrottleStop lets you request. The MSR PL1 and PL2 power limits that you have set to 33W seem to be ignored. The MMIO power limits are disabled so the only thing left is the EC power limits. If you want to do some testing you can try using ThrottleStop to set the MSR power limits sky high. If you are still seeing MSR PL1 power limit throttling at approximately 21.7W, I guess the EC power limits have been set lower. There might be a BIOS setting that lets you toggle these lower power limits on or off.

What BIOS version are you using? Lenovo might have added these lower power limits to the T480 after an update. It is also possible that the T480 with an i5-8350U has always been set to these values by the EC. My daughter had an 8th Gen Lenovo C930. It was wonderful. It did not have any EC power limits enforced so the sky was the limit. If you ever get lucky and find a laptop with no power limits, avoid any BIOS updates. Your computer is working great so leave it as is.

is it important to turn off memory integrity in BIOS?

The memory integrity setting is a Windows setting that needs to be disabled so ThrottleStop can access the CPU's voltage control register. Undervolting appears to be working correctly so this must already be off. If there is a separate memory integrity setting in the BIOS, I would disable it. I am not sure what it does but I am 99.9% sure that it is not necessary.

 

[Natcuber]

My BIOS version is N24ET76W (1.51 ). Supposedly it was released this year, so it's pretty new. Also, my BIOS has no voltage settings, EC settings, or anything fancy. It's quite plain and simple. It also turns out that the memory integrity being on was a mistake. I had confused it with execution prevention. Would you recommending flashing to an older version of BIOS to see if it unlocks the power limit?

I also readjusted my undervolt settings. Right now I'm seeing CPU to throttle at about 22.2w with 3.2ghz. I remember seeing my CPU holding 23-24w in some benches before. I'll try to see if I can find a setting that does that. Would there be any reasons for a CPU to hold back even if it has the PL headroom other than temps?

 

[unclewebb]

Would you recommending flashing to an older version of BIOS

How unhappy are you with how your laptop is running now? There is always a risk when flashing the BIOS, especially to an older version, that something might go wrong.

Would there be any reasons for a CPU to hold back

CPUs can throttle for a variety of reasons. Most of your throttling is being caused by power limits that are set conservatively. If you find a way to remove these power limits, then your CPU will run hotter and will start thermal throttling instead of power limit throttling. You will be no further ahead. That is why I think it would be foolish to flash an older BIOS version. There is next to nothing to be gained and if something goes wrong, your laptop will become as useful as a brick.

 

[Natcuber]

I'm pretty happy actually. I just like to maximize everything. But if there are risks, I'll stay with this. I've now managed to get my CPU to throttle above 23w at almost 3.3ghz instead. Thank you for all the help and for developing this wonderful software!

 

[Natcuber]

Does ThrottleStop support older CPUs like Sandy Bridge and Arrandale? I want to try and undervolt some old laptops which run pretty hot.

 

[unclewebb]

Does ThrottleStop support older CPUs like Sandy Bridge and Arrandale?

ThrottleStop supports older CPUs going back to the Core 2 Duo era. Unfortunately you need a 4th Gen Haswell or newer to access FIVR voltage control. ThrottleStop was very useful when overclocking and maximizing performance of the 1st Gen Core i mobile CPUs such as the 920XM. Here is a screenshot of TS 3.00 from the archives. It was posted 13 years ago! TS has been around for a while.

Pushing an i7 920xm to new levels by increasing the CPU core voltage

This post will quickly describe a way to increase (and decrease) the core voltage of the CPU, I have to advise anyone against doing this unless you're aware of all the risk that is involved. First of all a quick summary of how the voltage regulation works in the recent Intel CPUs, if you want to ...

www.techinferno.com

 

[Natcuber]

How unhappy are you with how your laptop is running now? There is always a risk when flashing the BIOS, especially to an older version, that something might go wrong.


CPUs can throttle for a variety of reasons. Most of your throttling is being caused by power limits that are set conservatively. If you find a way to remove these power limits, then your CPU will run hotter and will start thermal throttling instead of power limit throttling. You will be no further ahead. That is why I think it would be foolish to flash an older BIOS version. There is next to nothing to be gained and if something goes wrong, your laptop will become as useful as a brick.

I increased my PL1 and PL2 limits to 45w and now the power throttling is gone. It runs at full speed until it starts thermal throttling. Now all that's left is for me to get thermal paste and maybe swap the cooling pipes for better temps.

 

[misha1350]

I increased my PL1 and PL2 limits to 45w and now the power throttling is gone. It runs at full speed until it starts thermal throttling. Now all that's left is for me to get thermal paste and maybe swap the cooling pipes for better temps.

I'd rather not push it that hard. 33W for the PL2 and 30W for PL1 should already be fine. If you also run the CPU this hot all the time, consider increasing the PROCHOT offset from 3 to 15 to make sure it starts thermal throttling at 85C, because you're at the point of diminishing returns anyway with the amount of power you feed the already undervolted CPU with.

Above 85C is when the silicon starts to degrade much faster due to thermals. You should be extra cautious when pushing the CPU that hard, even if it's undervolted. Do you know how much of a pain it is to replace a soldered CPU on a Lenovo laptop? The ThinkPad T480 is a fantastic piece of tech - don't kill it just now, it can serve you for another 5 years. And use Honeywell PTM7950 to repaste the CPU, along with some good thermal pads

 

[Natcuber]

I'd rather not push it that hard. 33W for the PL2 and 30W for PL1 should already be fine. If you also run the CPU this hot all the time, consider increasing the PROCHOT offset from 3 to 15 to make sure it starts thermal throttling at 85C, because you're at the point of diminishing returns anyway with the amount of power you feed the already undervolted CPU with.

Above 85C is when the silicon starts to degrade much faster due to thermals. You should be extra cautious when pushing the CPU that hard, even if it's undervolted. Do you know how much of a pain it is to replace a soldered CPU on a Lenovo laptop? The ThinkPad T480 is a fantastic piece of tech - don't kill it just now, it can serve you for another 5 years. And use Honeywell PTM7950 to repaste the CPU, along with some good thermal pads

I set it to that temp, but it only reaches 33w at max. It only usually reaches prochot when doing benchmarks. Generally it stays below 80c.

 

[x3oo]

Checking the MMIO Lock box eliminates the MMIO power limits. Those power limits are no longer in control of your CPU after checking that box. Intel CPUs do not need multiple sets of redundant turbo power limits all fighting against each other.

When More Data is not checked, ThrottleStop will sample the CPU core and package temperatures only one time per second. When the CPU temperature is rapidly changing, it is possible to trigger thermal throttling without ThrottleStop recording the peak core temperature in its monitoring table. If you clear the PROCHOT 97°C flag before you start testing and this flag lights up red during a test, that confirms that the hottest spot on the core definitely reached 97°C and the CPU really was thermal throttling. It might have only reached this temperature for a few milliseconds. When this happens, the log data might not show 97°C as the max. If the log file shows TEMP, the CPU was thermal throttling and some temperature sensor on the CPU must have reported 97°C at some time during the one second monitoring interval.

When you check the More Data box, ThrottleStop will sample the CPU temperature sensors approximately 8 times per second. This might help you "catch" the CPU reaching that magical 97°C temperature. Beyond that, there is no practical reason to use the More Data feature. It creates way "too much information".

During your log file testing, the early tests show that almost all of the throttling is being caused by the PL1 and PL2 power limits. This seems to happen when PL2 is at 29W. PL1 throttling seems to get triggered at approximately 21.7W. During the longer test at the end, PL1 throttling changes to TEMP thermal throttling. This happens when the CPU starts running at a consistent 97°C.

There is a third unique set of turbo power limits that only some computers use. This third set is managed by an embedded controller (EC). It looks like Lenovo has programmed these limits to lower values than what ThrottleStop lets you request. The MSR PL1 and PL2 power limits that you have set to 33W seem to be ignored. The MMIO power limits are disabled so the only thing left is the EC power limits. If you want to do some testing you can try using ThrottleStop to set the MSR power limits sky high. If you are still seeing MSR PL1 power limit throttling at approximately 21.7W, I guess the EC power limits have been set lower. There might be a BIOS setting that lets you toggle these lower power limits on or off.

What BIOS version are you using? Lenovo might have added these lower power limits to the T480 after an update. It is also possible that the T480 with an i5-8350U has always been set to these values by the EC. My daughter had an 8th Gen Lenovo C930. It was wonderful. It did not have any EC power limits enforced so the sky was the limit. If you ever get lucky and find a laptop with no power limits, avoid any BIOS updates. Your computer is working great so leave it as is.


The memory integrity setting is a Windows setting that needs to be disabled so ThrottleStop can access the CPU's voltage control register. Undervolting appears to be working correctly so this must already be off. If there is a separate memory integrity setting in the BIOS, I would disable it. I am not sure what it does but I am 99.9% sure that it is not necessary.

i think there are modified lenovo bioses which open up oem settings. you can get them on badcaps forum. last time i tried them there were power settings.

 

[PHVM_BR]

Above 85C is when the silicon starts to degrade much faster due to thermals. You should be extra cautious when pushing the CPU that hard, even if it's undervolted.

So should I limit the CPU to up to 85°C because of silicon degradation?

Even if this happens, will the CPU's life cycle decrease from 15/20 to 10/12 years?

Long before the laptop will be scrap, besides the CPU which will be a museum piece...

 

[x3oo]

@Natcuber i found it: https://github.com/digmorepaka/thinkpad-firmware-patches/tree/master
there you can unlock the advanced menu with a bios patch. it will open up a lot of new settings.

 

[Natcuber]

@Natcuber i found it: https://github.com/digmorepaka/thinkpad-firmware-patches/tree/master
there you can unlock the advanced menu with a bios patch. it will open up a lot of new settings.

Thanks a lot! But before I flash my BIOS, I would like to know if there are any risks, and how much of an advantage it would be over simply using ThrottleStop?

 

[unclewebb]

I increased my PL1 and PL2 limits to 45w and now the power throttling is gone. It runs at full speed until it starts thermal throttling.

There does not seem to be any reason to flash the BIOS.

I would like to know if there are any risks

There is always a risk when flashing the BIOS. There is always the possibility of something going horribly wrong. You can turn your laptop into the equivalent of a brick. Your computer may not turn on after updating the BIOS. Is flashing the BIOS worth that risk?

how much of an advantage it would be over simply using ThrottleStop?

I see zero advantage. Why risk your computer if it is running great?

 

[Natcuber]

There does not seem to be any reason to flash the BIOS.


There is always a risk when flashing the BIOS. There is always the possibility of something going horribly wrong. You can turn your laptop into the equivalent of a brick. Your computer may not turn on after updating the BIOS. Is flashing the BIOS worth that risk?


I see zero advantage. Why risk your computer if it is running great?

Thanks for replying. There is something I want to ask actually.

I have two profiles for plugged in and on battery. The plugged in one is set to -110mv on CPU and cache with speedshift at 0. The battery profile is at -100mv for core and -95mv for cache and speedshift at 160. I noticed that when I plug-or unplug my laptop, there's a chance of a crash or BSOD happening in the next few seconds, especially if I'm doing anything CPU-intensive. Both profiles by themselves had been running perfectly fine. Is there any way to fix this?

 

[unclewebb]

Is there any way to fix this?

No there is not. A crash means the CPU needs more voltage. There is no point living life on the edge of stability. Using a Speed Shift value of 160 slows the CPU down. It is not unusual when you do this that the CPU will need more voltage to be stable. I would not use a Speed Shift EPP value any higher than 84. Making a CPU inefficient may not save any power. It just makes it a little more sluggish to get up to speed. This might be pointless if it is causing instability and limiting your undervolt.

Keep in mind that 8th Gen U series are already low power CPUs. To save power, reduce the amount of stuff running in the background.

 

[Natcuber]

No there is not. A crash means the CPU needs more voltage. There is no point living life on the edge of stability. Using a Speed Shift value of 160 slows the CPU down. It is not unusual when you do this that the CPU will need more voltage to be stable. I would not use a Speed Shift EPP value any higher than 84. Making a CPU inefficient may not save any power. It just makes it a little more sluggish to get up to speed. This might be pointless if it is causing instability and limiting your undervolt.

Keep in mind that 8th Gen U series are already low power CPUs. To save power, reduce the amount of stuff running in the background.

So you think it's the speed shift that's causing this? I'll try lowering it to see if it helps. Also, is there any reason that the crashes happen only when changing profiles, and not when using them?

Edit: I realized that the crash happens when being reconnected to power, not when switching to battery. Hope this info is useful.

 

[x3oo]

@Natcuber With Biosflash you can overclock the gpu.
but i dont remember other settings.

 

[Natcuber]

@Natcuber With Biosflash you can overclock the gpu.
but i dont remember other settings.

Overclocking the GPU? Boy that is tempting. I'll hold off for now, since my temps aren't that great in the first place, and it's risky.