• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Help Tweaking 8550u In ThrottleStop

Falcorion

New Member
Joined
Dec 8, 2018
Messages
9 (0.03/day)
Hello,

I was wondering if anyone who has experience with undervolting and squeezing max performance out of an 8550u could have a look at my settings. I have been running stable with these settings for a while now. Temps have been pretty good since I already repasted as well (might go back with better paste and new thermal pads later for better thermal headroom) and i was wondering if there is anything I should be using or changing that I haven't yet or maybe I have done something i shouldn't as well. Either way, please let me know your thoughts! Screenshot images attached.

Capture.PNG Capture3.PNG Capture2.PNG

Thanks
 

unclewebb

RealTemp Author
Joined
Jun 1, 2008
Messages
1,379 (0.34/day)
Overall your settings look like you know what you are doing. You do mention that, "temps have been pretty good", but your screenshot shows the CPU is reaching the maximum thermal throttling temperature.

The Intel specified default thermal throttling temperature for an 8550U is 100°C but it looks like the manufacturer of your laptop decided to reduce that to 97°C. The check mark in the PROCHOT 97°C box confirms that there has been some thermal throttling. You can click on this box to clear that data which is stored in the CPU. Definitely try to improve your thermal paste job. These CPUs can handle high temperatures but for consistent performance, it is best to avoid any thermal throttling.

Is there any reason the Turbo Boost Short Power Max box is not checked? Below that where the Speed Shift box is, the Min is set to 1 and the Max is set to 128. The Max value should be set to your Maximum multiplier which is 40. Will this make any difference? Probably not.

There is rarely any need to have the BD PROCHOT box checked. The CPU will still thermal throttle whether this box is checked or not. Leaving BD PROCHOT checked allows other sensors on your motherboard or in the power supply to throttle your CPU. Your laptop might not use this feature.

On the main screen you can check Speed Shift - EPP. For maximum performance, try adjusting the EPP value from 128 to 0.

The best thing you can do is turn on the Log File option and then use your computer. The log file will keep track of your CPU's performance and temperatures and will show any throttling problems. Exit ThrottleStop so it can finalize the log file.

When idle without any web browser etc. open, what C0% is reported. A big problem I see is that many people have too much junk running in the background. Somewhere around 0.5% in the C0 state when idle is a good number to shoot for.

 

Falcorion

New Member
Joined
Dec 8, 2018
Messages
9 (0.03/day)
Thanks unclewebb, you're the best.

I went through your recommendations and now have it running consistently at higher frequencies on full load. I also adjusted the Turbo Boost Power Limits a bit to dial back the temps from peaking high enough to trigger PROCHOT. I am not sure what my reasoning was behind having turbo boost short power max not checked but it is definitely better with it on now. I will definitely be getting some Thermal Grizzly Kryonaut to put on there and swap out the heat spreader pads with some Arctic ones instead and see how much that helps too.

I'll come back to this thread with my results after all that!
 

Biosys

New Member
Joined
Jun 28, 2019
Messages
2 (0.04/day)
Hello Falcorion (or unclewebb, if you have time), maybe one of you are still watching this thread, didn't want to start a new one: I have just gotten a Lenovo X1 Carbon with an i7 8550u and trying to undervolt using ThrottleStop - I am getting some crazy results. The short of it is that it seems (?), I am undervolting the CPU core by 400 mV (?), cache by 105 mV, (intel GPU, Unslice, System Agent 50mV, but I havent even really tried those ones). Do you think it's for real? I checked with HWINFO64, it seems to be true, but I am hard pressed to believe this. At the beginning, before undervolting, I was getting constant thermal throttling, the laptop seems to throttle at 91C. After undervolting, I only have TDP throttling. Even using Cinebench r20, now the temp barely goes above 80. The score in Cinebench increased from around 1180-1200 to 1450-1500.
At this level the laptop is rock solid - any change on cache undervolting will result in crashes, but I can still increase the CPU core undervolt. However, I am sort of doubtful of these values :). What do you think? Is this for real or something is messing up the numbers? Should I keep pushing or just be happy - I am perfectly happy with the current results :-D in a thin light notebook???
cheers


st1.gifst2.gifst3.gifhw64.gif
 

unclewebb

RealTemp Author
Joined
Jun 1, 2008
Messages
1,379 (0.34/day)
The values you enter into ThrottleStop are sort of like voltage requests. ThrottleStop allows you to request whatever you want. If the CPU does not understand your offset voltage request, it will ignore it or it will ignore the part that it does not understand.

There is some debate in various forums about whether requesting a larger offset for the core compared to the cache makes any difference or whether these should always be set equal. Intel XTU only lets you set these 2 registers equally. Some users are convinced that going higher on the core really does make a difference during testing. If this does work, it also seems like there is a certain point where going and making a really crazy request is ignored. HWiNFO just reports what ThrottleStop is requesting. It does not confirm how much of that request is being used or ignored.

For an example, if -100 mV Core and -100mV Cache is your default undervolt, perhaps -150mV Core and -100mV Cache is better. Maybe -200mV Core and -100mV Cache is even better yet but at some point, maybe -300mV Core or -400mV Core and -100mV Cache might not show any improvement compared to the previous setting. I do not have any documentation from Intel that confirms what would be best. Maybe using different offsets does not change the actual voltage but perhaps it is tricking the processor into thinking that it is using less voltage. This trick might allow more turbo boost or less throttling. It is still a mystery.

To try to prove what is going on, you will need to do lots of testing. I would start by using something consistent like the TS Bench test to put a constant load on the CPU. While a long 1024M bench is in progress, leave the Cache offset voltage consistent and only adjust the Core offset voltage. See if there is any difference in CPU speed or CPU temperature. There is no problem doing voltage adjustments while the CPU is loaded. It would be great to see someone prove this once and for all. There are still lots of doubters. According to Intel, these should be set equally. On older CPU generations this seemed to be true but on some of the 8th Gen, maybe not. Maybe independent adjustments are best.

In terms of performance, I would increase the Turbo Boost Long Power Max way beyond 25 and I would check the Disable and Lock Turbo Power Limits feature. Doing those two things on my daughter's Lenovo C930 made a huge difference to maximum performance. Temperatures went through the roof but performance was much improved.

Here is a CPUz benchmark to compare to.


Not too bad in Cinebench either after some ThrottleStop tweaking.


If you want to create a custom Lenovo version of ThrottleStop, just download and rename the next image to logo.png and copy that into your ThrottleStop folder.


No power limit throttling. Just temperature throttling when it gets too hot.
 
Last edited:

Biosys

New Member
Joined
Jun 28, 2019
Messages
2 (0.04/day)
Thank you for the super quick reply - yes, its much clearer now. Btw I started with XTU and somehow the numbers didnt seem to stick. At most I could do about 80mV on core/cache with XTU, but the values reset after a restart and temp values were inconsistent. So then I switched to ThrottleStop, but I was using the benchmark test in XTU. XTU was reporting the mV request up to around 130mV core change and then it just ignored the values I put in. I was considering that the two programs may clash, and uninstalled XTU (it was a hassle to get it installed at all).

I will have to test this further, but I can tell that at least for a while it really seemed that the core undervolting made a difference - I would say probably up to about 150 mV or so, as I was doing Cinebench after every round of 10 mV change. After that it just seemed like nothing was happening. OK, will definitely do more elaborate tests and I will update you! Probably it will take me a few weeks to get back to this, busy with projects :-(.

I was doing a lot of oc'ing in my times on ABIT motherboards, but I have to admit that in the past 10 or so years I didnt't have time for it (apart from tweaking my son's computer). So it was fun trying to do some undervolting, but things have definitely changed :) and my vocabulary & knowledge needs an update :). Once again, thanks for the reply & your comments on the forum helped a lot to get going with ThrottleStop!
Will try your recommendations and a test on core/cache.
Best,
 

unclewebb

RealTemp Author
Joined
Jun 1, 2008
Messages
1,379 (0.34/day)
At most I could do about 80mV on core/cache with XTU, but the values reset after a restart
That is the biggest problem with Intel XTU. Values get reset and you are never sure if your undervolt is working or not. People everywhere have mentioned this problem for years but as far as I know, it has still not been fixed. Installing and trying to uninstall XTU can also be painful so I just don't bother testing it anymore.

I was considering that the two programs may clash
When running both programs at the same time, ThrottleStop is usually in control of the voltage. XTU ignores when other programs make changes to the voltage registers. It just assumes that it is the one and only program on the planet that can set these voltages. I would not trust anything that XTU says while using ThrottleStop at the same time. Uninstalling XTU was a wise choice.
 

Alkerion

New Member
Joined
Jul 21, 2019
Messages
3 (0.09/day)
HI !

I've a Lenovo C930 with a 8550U cpu.

As my Laptop was really overheating I've decided to give a try to ThrottleStop.

I've followed the tutorial but I've some doubts regarding the undervolting.

I've set the CPU down to -500mV, CPU Cache -110mV, GPU -115mV.

Seems also hard to me to believe that the PC still works at -500mV, so I think TS is just not working.

Even at -500mV the CPU is going up to 3.4Ghz and temp rise to 95°C quite immediately and it start to throttling after 1s, slowing to ~2.4GHz and even slower.


I also have no idea how you reach a C0 of 0.3 as mine never go below 3, and I've nearly nothing installed on the PC (antivirus and very few other things requiring no CPU while idle).
 

unclewebb

RealTemp Author
Joined
Jun 1, 2008
Messages
1,379 (0.34/day)
Seems also hard to me to believe that the PC still works at -500mV, so I think TS is just not working.
TS definitely works. Set the CPU Core and CPU Cache under volt to the same values. Scroll up this page and read my June 28th post.


Some people claim to have had success by using a slightly higher CPU Core offset compared to the CPU Cache. This might only be possible on some CPUs (8750H) and not others. They have also found that beyond a certain point, any additional CPU Core offset will be ignored. I do not have access to an 8750H to prove this one way or the other. On the 8550U I was using, I do not remember observing any benefit using different offsets so I set the CPU Core and CPU Cache offsets equally.

For the Intel GPU, you usually need to have the iGPU and the iGPU Unslice undervolted equally. If you only undervolt the iGPU, this setting will probably be ignored.

requiring no CPU while idle
What antivirus program are you using? The C0% data that ThrottleStop reports is a very accurate measurement of what your CPU is doing. If you cannot get the average below 3% when idle, you definitely have an inefficient program or two running in the background.



It only takes one poorly written program to screw things up. After installing the Adobe Creative Cloud on my daughter's C930, things went to crap real fast. Open up the Task Manager click on the Details tab and then click on the CPU heading. Have a good look there for what is running on your computer.

What benchmark are you running that is instantly sending your CPU up to 95°C? The screenshot I posted shows that the TS Bench 64M test was 83% complete before thermal throttling kicked in. Your room temperature might be significantly higher or your heatsink might not be attached as well. If it was my laptop, I would replace the thermal paste and I would make sure the heatsink was square, snug and making full contact with the CPU.
 

Alkerion

New Member
Joined
Jul 21, 2019
Messages
3 (0.09/day)
Hi Unclewebb,

Thanks for your reply, and for TS too.

Here are my actual settings :
2019-07-22_21-51-58.png
2019-07-22_21-53-02.png
2019-07-22_21-54-08.png

I ran TS Bench 1024 and obtained a score of 189.3 with Performance settings. score 9.9 with 64.

I've replaced the thermal compound, now it doesn't throttle with 64 TS Bench.

What sound strange to me is that I never encountered any BSOD during all my tests, that's why I was thinking undervolting wasn't working.

My biggest concern with this PC is overheating when using it while connected to AC plug, PC case temp rise up to 53°C.
Issue I hope to solve with TS.
With actual settings and new thermal compound I'm not at 47°C, better, but still too warm for the case.

Something strange is that in w10 power settings only Balanced is available, and lot of options are unavailable.

I use Bitdefender, with Dashlane, Logitech Option and SnagIt running on background (using 01 CPU sometimes).

CPU usage on idle show only these as using more CPU :
dwm.exe
System interupts
WmiPrvSE.exe
svchost.exe


Thanks and best regards
 

unclewebb

RealTemp Author
Joined
Jun 1, 2008
Messages
1,379 (0.34/day)
I've replaced the thermal compound, now it doesn't throttle with 64 TS Bench.
That is good to know. I should probably do the same but my daughter does not run any demanding software so I have not yet bothered.

Why are you using Clock Modulation with it set to 75%? I would either reduce the turbo power limits or reduce the turbo ratio limits if you want to slow your CPU down or reduce its maximum power consumption. Clock modulation is not a smooth way to slow a CPU down.

For maximum performance, I highly recommend checking the FIVR - Disable and Lock Turbo Power Limits feature. You will need to first download RwDrv.zip from Mega and copy the RwDrv.sys file into your ThrottleStop folder to use this feature.


Intel says their CPUs can safely run up to 100°C. In a small and light laptop, with this CPU fully unlocked, high temperatures will be the result. The CPU will thermal throttle if it needs to regardless of how you have ThrottleStop setup so no worries. If your laptop is burning through your lap then reduce your turbo power limits.

Windows 10 hides some of the power profiles but they are still there. To access the Windows High Performance power profile, open up a command prompt and type in this.

powercfg /s SCHEME_MIN

To go back to the Balanced profile, type in this command.

powercfg /s SCHEME_BALANCED

On a laptop, open the Control Panel - Power Options and at the lower left you should see Windows Mobility Center. This should also give you access to the High Performance power profile without having to play around with command prompt and the powercfg command.

I prefer using the High Performance power profile and then I can adjust the Speed Shift EPP value in ThrottleStop to control the CPU. If you are using the Windows Balanced profile, there is a good chance that Windows is in control of the EPP setting and what you enter into ThrottleStop for EPP is being ignored. Open up the FIVR window and have a look at what is being reported in the top right monitoring table for EPP. The EPP value in that table is what the CPU is actually using in real time. If this value does not agree with what you are requesting on the main screen of ThrottleStop, that means Windows is in charge of EPP. Do not bother checking or adjusting the Speed Shift EPP setting in ThrottleStop if Windows is in control of EPP.

If you did not see a BSOD when testing, that is a good thing. It means you have a CPU that you can probably undervolt a little further. Every CPU is unique. No one can tell you how far you can go unless you do some hands on testing. Reduce the CPU Core and CPU Cache equally in small steps if you want to go further. You might be getting close to the limit. Watch for errors in the TS Bench test. If you see any errors, you have gone too far so back off a little.

(using 01 CPU sometimes)
Look at the C0% numbers in the picture you posted compared to the picture I posted. If your laptop was idle, there is more going on in the background than needs to be going on. High C0% when supposedly idle can significantly reduce battery run time. Open up the C State data. Cores should be spending 99% of their time in the low power C7 state when your laptop is idle.
 

Alkerion

New Member
Joined
Jul 21, 2019
Messages
3 (0.09/day)
Hi !

I finally got my first BSOD with CPU settings at -150mV, and screen artifacts with -130mV on System Agent. GPU at -60mV and it's still running fine.

Thanks for the trick to recover high Perf, and you was right, in Balanced W10 set the EPP at 178, in High Perf it take the one I set in TS.

I've also removed the Clock Modulation settings and changed the Turbo ratio as suggested.

Now with my settings, in Performance mode, under TS Bench 256 there is no throttling until ~80% of the test, and I obtained a pretty good score of ~40.

Could you please let me know why the result is sometimes green, sometimes red ?

How to know if there is errors during TS Bench ?

Regards
 

unclewebb

RealTemp Author
Joined
Jun 1, 2008
Messages
1,379 (0.34/day)
I noticed that EPP was set to 178 in the monitoring table in one of your screenshots. That is too high. Windows typically sets this to 128 but on some CPUs, running some apps, a setting of 128 has proven to reduce maximum performance. I like setting EPP to 80 which is similar to how the Windows Balanced power profile worked in previous operating systems. The advantage with Speed Shift is that it gives better off idle performance when setup correctly.

In the TS Bench, a result displayed in green just means you have set a new record. A result in red means you did not set a record. Green is good. Red, not quite so good. If you have errors during this test, instead of a final time being displayed, the TS Bench result will show 5 Errors or however many it found. Some users have discovered that the TS Bench will start to show errors right around the time they have found a maximum safe under volt. Some users have thought that errors in the TS Bench can be ignored. A day or two later, they tell me about a BSOD and then I can say, "See, told you so!" Errors are always bad.
 
Top