Encountered weird behaviour while undervolting my i7-8665U

[hkgute99]

So I have been using Throttlestop for undervolting for a while now since its so much better than XTU (and latest versions don't even support my processor).
Since past 6 months, I had been using Throttlestop since Dell has provided such an ineffective cooler with the Latitude 5500, that it used to thermal throttle at almost anything.
Earlier I just used to disable the Turbo Boost, but then the laptop performance just drops significantly.
Thankfully, undervolting had made my thermals and package power usage much better and its definitely a lot better now.

Since past 6 months, this has been my voltage offsets.


Earlier just going above -110mV on both made my system unstable and TS Bench would show errors, so I never bothered tweaking more since I was more than satisfied.
But past week I finally tweaked it more. I found out that if I reduced the cache offset from -110mV to just -100mV, the core can go upto -190mV while still remaning 100% fully stable.
This behaviour is something I haven't seen anywhere before.
So is this normal and should be expected?

Also is having such a big difference between core and cache voltage offsets normal?

 

[unclewebb]

This behaviour is something I haven't seen anywhere before.

Using different offset voltages was popular with 8750H and 9750H processors. I just did a write up this morning in a different thread that explains what is likely happening.

Undervolting Intel i7-10750h on MSI GE75 Raider

Hi all, I'm new to the forum and sorry for my English but I don't speak it very well and I help myself with Google Translate :p. I have read many threads about Throttlestop and undervolting and thank you for learning a lot. I have a GE75 Raider with Intel 10750h. I'm not interested in energy...

www.techpowerup.com

The bottom line is that if you get better performance or better temperatures by using different voltages then continue using different voltages. Use Cinebench R23 when testing your results.

Also is having such a big difference between core and cache voltage offsets normal?

Yes. Whatever works best is normal. You will reach a certain point where going any further with the core offset will be ignored by the CPU. You can probably adjust the core slider to -1000 mV. You might still be 100% stable because most of that difference is being ignored by the CPU. Most owners of U series processors have never tested using different voltages.

 

[hkgute99]

So I was experimenting with a few more settings. I raised my PL1 from 28 to 45 and PL2 from 51 to 60. Also time was increased from 28 to 640.
However the cinebench R23 score dropped. And it isn't a small drop, it is a drop of more than 200 points.

Note that before this change, there were many more power limits appearing in the Limits tab. PL2 isn't even reached, as max draw was about 50W.

 

[unclewebb]

You have to check the MMIO Lock box. Your MMIO power limit is set lower than the MSR power limit. Increasing the MSR limits will be ignored when the MMIO limits are lower. The lowest power limit wins control of the CPU. It is easier to manage only the MSR limits. The MMIO Lock box disables the secondary power limits.

No need to increase the turbo time limit to 640 seconds. Do you really need more than 10 minutes of turbo boost? Your laptop cooling system cannot handle that. The default 28 seconds is more than enough turbo time.

Try setting Power Limit 4 to a value of 0.

Your CPU does not support overclocking. You can set the turbo ratios back to the default values that are listed. U series CPUs do not support an all core overclock.

 

[hkgute99]

Do I need to lock the PL4 as well? What exactly will lock do?

 

[hkgute99]

So I ran some tests and finally found some success.

Firstly I had reduced the turbo time to 28.
As you can see Max power draw was 50.5 W. Score did increase, but still not to the level it was earlier. (4139 vs 4343 (earlier))
But then I spotted that for some weird reason, MMIO wasn't locked (I swear I had ticked that thing), so I just locked it and re-ran Cinebench R23.
And it worked, now got a score of 4506 vs 4343, which was previous best. So that MMIO lock thing did work.
I also figured out what the lock does, it basically ensures that the manually set value will be locked when shut down/restart etc.

After the second test (the one in which score was 4506, sorry I don't have screenshots for it), the max power draw was 49.9W. Also the in the Limits tab, everything was same as the one in the screenshot, except BD PROCHOT wasn't present for CPU and GPU, which I guess is a good thing.
However, TVB has now started popping up, and this wasn't there until I had raised my PL1 limits.

So I have a few questions here,
1) Why is TVB (terminal velocity boost) limit kicking in after PL1 and PL2 have been raised?
2) What does PL4 stand for and does (sorry I am a technical nerd and I love to learn more)? From what I understand PL4 is the absolute power limit that CPU can handle without blowing up. So putting a 0 means effectively we are disabling it.
So if I put 82 (default value) back into there, there should be no effect since my CPU isn't going to reach that power limit anyways.
3) Should I enable clock mod at 100%. I read some old posts which suggests that some old CPUs benefit from it. So is intel Gen 8 CPUs affected by enabling/disabling it?
4) I have found that putting Speed Shift EPP to 80 gives better performance than putting it at 60. I have seen in old posts that you even told some people to put it at 0. So why is performance decreasing on putting it at 60 vs 80?
5) What exactly does pressing Turn On/Off button and putting ThrottleStop in Monitoring mode do? I found that whenever ThrottleStop was in Monitoring mode and the button says Turn On, Cinebench gives better scores. Maybe its coincidence or does it actually affect something?
6) Are there any other settings that I can change to get better performance out of this thing. There are many options that I haven't even touched (I don't want to ruin my laptop). It would be great if you could guide me.

Sorry for the long post. I apologize for it. But I will appreciate if you could answer all of my queries.

 

[unclewebb]

1) TVB throttling is typically triggered when a CPU reaches about 70°C. When you increase the power limits, you create more heat and this triggers TVB throttling in Limit Reasons. Clear the Thermal Velocity Boost box in the FIVR window to get rid of this throttling.

2) I believe PL4 is a current limit. Setting PL4 to a value of 0 tells the CPU to ignore this limit so it should not cause any throttling. A CPU will only flow as much current as it needs so disabling this limit has never caused any problems that I have ever heard about.

3) 8th Gen CPUs do not seem to ever use Clock Modulation throttling. It is still possible for these CPUs to use this control method but I cannot remember any instances of that off hand. When this is not checked, turn on the Log File option and run a log file. Leave that on while gaming or doing whatever you do. When finished testing, look in the log file. If the CKMOD column always shows 100.0 then you do not have a clock modulation throttling problem. That means you do not need to check the Clock Mod box.

4) The difference in performance when setting Speed Shift EPP to any value between 1 and 84 should not be significant. Set it to any value you like. Whatever works best for you is OK. I prefer to let Windows manage the Speed Shift EPP value. When Speed Shift EPP is not checked in ThrottleStop, when you change to different Windows power plans you should see the EPP value change in the FIVR monitoring table.

5) The Turn On Turn Off button controls the sending of Set Multiplier values and clock modulation values to the CPU. Your CPU uses Speed Shift Technology which replaced SpeedStep technology which the Set Multiplier feature used to use. If you are not using Set Multiplier and if you are not using Clock Modulation then there is no need to press the Turn On button. It is possible and likely that you get slightly better performance when this button is not pushed. ThrottleStop is slightly more efficient if it does not have to constantly monitor and write values to registers that are no longer being used.

6) Pretty hard to ruin a laptop by playing around with ThrottleStop. A million people a year download and use the program. Do a Google search. There are not a lot of reports of "ThrottleStop ruined my laptop".

Your overall settings look good. I always clear the BD PROCHOT box on the main screen. A CPU will still be able to thermal throttle if it ever gets too hot whether BD PROCHOT is checked or not. I prefer setting IccMax for both the core and the cache to the max, 255.75. This is just a limit. Setting limits sky high is like raising the limbo bar 20 feet high. Nice and easy to walk under.

Try using ThrottleStop 9.6. It has a few fixes so your PL4 power limit in the FIVR window should be reported correctly now.

Your 8565U has a 15W TDP limit. using ThrottleStop to run it at 50W is pretty darn good. 50W vs 15W makes for a much more powerful laptop. Many manufacturers locked these CPUs to 15W. Avoid ever updating the BIOS. A BIOS update could wreck your laptop. Do some more Google searching to find out how to block BIOS updates.

 

[hkgute99]

I never bios update since Dell had disabled undervolting since Version 1.6.3 of my laptop Dell Latitude 5500 due to Plundervolt. And they don't even allow downgrading below that. So even to use Throttlestop, I had to edit the bios variables or something, and since then I haven't updated the bios.
One of the few good things Dell does is that there is an option in BIOS to disable updating through windows update.

I have noticed that ever since I increased IccMax on both cache and core, I am able to undervolt more.

Few more questions.
1) I was recently reading a post where a person actually got higher Geekbench score on enabling Sync MMIO rather than Lock MMIO.
Before I wanted to test it out, I wanted to know the difference between the two options.

2) What does Power Balance and PP0 power limit do here?
3) Also even though I never play games on this system, I do occassionally use this laptop for playing old games and emulation stuff. So if I crank up the IccMax of iGPU, will that increase some performance? And will it be okay for both CPU and iGPU to have IccMax set to 255.75?
4) How much can you undervolt a iGPU? And how to test it for stability? I have seen people using MSI Kombustor for testing GPU stability, but I have also read that it doesn't represent real world scenarios? Can you help me with this one?

 

[unclewebb]

I have noticed that ever since I increased IccMax on both cache and core, I am able to undervolt more.

That sounds good. The original IccMax value was very low which could be restrictive.

Before I wanted to test it out

Do not wait for anyone's opinion before testing something out. There should be no noticeable performance difference between MMIO Lock and Sync MMIO. When you use Sync, it sets both the MSR and MMIO power limits to the same values. This setting wastes CPU cycles because ThrottleStop has to constantly monitor the power limit value in memory. If you Lock MMIO to a fixed value, ThrottleStop no longer has to monitor the turbo power limit memory location for any changes. You will need to reboot after MMIO is locked before the memory power limits can be changed.

higher Geekbench score

Some people that post their results are not very scientific when it comes to benchmark testing. It can be difficult to get repeatable benchmark results from many laptops even if you do not change anything from one run to the next. Some variation is normal. Many computers have over a thousand threads running in the background that are constantly starting and stopping. Hard to get repeatable results without making sure that Windows is as lean as possible.

Power Balance

If a CPU needs to power limit throttle, the Power Balance feature can be used to decide what should throttle first to reduce power consumption. Either the CPU will slow down first or the Intel GPU will be told to slow down first. This setting is fairly useless and is rarely used by anyone.

PP0 Power Limit

The main power limit is for the entire CPU package. The PP0 Power Limit is a power limit that only applies to the individual cores. The PP0 Power Limit should be left unchecked and set to 0. This limit is not necessary.

if I crank up the IccMax of iGPU, will that increase some performance?

Probably not. It never hurts to try something though if you are curious. I never use the Intel GPU so I have not done any significant testing of it.

How much can you undervolt a iGPU?

No idea. Just remember to undervolt both the Intel GPU and the iGPU Unslice equally.

Can you help me with this one?

You need to help yourself. Do some testing. Find out what works best for your computer.

 

[hkgute99]

For some reason, you removed Random frequency benchmark from TSBench since ver 9.6? I found that till version 9.5, random frequency option was much more capable of producing any possible errors than the fixed frequency option? Why did you remove that option? I just wanted to know.
It seems I had undervolted the cache a bit more and it became problematic, but testing with TSBench using the 960M test with fixed mhz didn't produce any errors. This resulted in laptop just BSODing just as it started. Everytime I restarted my laptop, it just started crashing. Good thing is that I have configured Throttlestop to only start with AC power, so I could easily log-in in a stable configuration. And after half an hour when it finally didn't crash soon enough, I was able to delete the Throttlestop.ini and fix those values.

 

[unclewebb]

you removed Random frequency benchmark from TS Bench since ver 9.6

Did you read the included ReadMe file? I am using TS 9.6 and I still have the TS Bench Random MHz feature.

- disabled the TS Bench - Random MHz feature when TPL - Speed Shift is not checked.

 

[hkgute99]

Oh thanks for pointing that out! I never actually read that file because I only replaced the older version of the executable with the newer version.

 

[unclewebb]

I never actually read that file

That is why I stopped bothering with documentation. The people that need documentation the most, never bother to read it.

 

[hkgute99]

I would actually love to have proper documentation. I think if you have proper documentation along with a recommended setting of each function, half of the questions on this forum will be answered by that alone (including mine).

 

[hkgute99]

So firstly, I tested out the performance difference when using Sync MMIO and Lock MMIO. It wasn't much, 4468 vs 4519 respectively.
Another thing I have noticed that ever since I have deleted the old throttlestop.ini and started a new one (because my computer was in bsod mode literally) a fwe days back, even though Thermal velocity boost is disabled in FIVR, I always see it in the Limits section. This time it even comes when on idle and temperature aren't even 70C.

 

[unclewebb]

4468 vs 4519

This is within the margin of error. If you run Cinebench 10 times and you do not make any changes, there is always going to be some variation in your scores. It is difficult getting repeatable results on any computer running Windows unless all of the background tasks can be controlled or their impact minimized. Many laptops with U series CPUs can change performance significantly between the first run after booting up and the second or third run depending on how long you wait between each test. Notebook Check used to include this info in their laptop test results. Sync vs Lock MMIO should make zero difference in any test that produces repeatable results.

Thermal velocity boost

On some CPUs, TVB might show yellow in Limit Reasons even when Thermal Velocity Boost is disabled in the FIVR window. I know this can happen but I am not sure why this happens. It is probably caused by a bug within the CPU. Every Intel CPU ever made comes with a long list of what Intel calls errata. Those are just the bugs they know about. Some of these bugs are fixed with microcode updates when possible. Many bugs are so insignificant that they are never fixed.

I do not think TVB in yellow is reducing your performance any. Try running a Log File. Check the log file data to see if there are any TVB throttling messages. Check the speed the CPU is running at. Are you having any throttling issues because of this?

 

[hkgute99]

It actually does seem that TVB is engaging and inducing some throttling. As seen from the attached pic, it seems that just as it engaged, it reduced the temps to 55 and pushed down clocks. And the weird thing is it is engaging very randomly. I have attached the full log as well, and according to Intel, TVB should engage when a processor is above 70C, however if you inspect through the logs, I was well above 80C before and TVB never engaged.


The weird thing of all this is that IIRC, this behaviour wasn't there until I was using TS 9.5 . When I switched to 9.6, just as I had mentioned before I had just replaced the old executable with the new one and it still worked fine. It was only after I pushed undervolt a bit too much so I had to delete the throttlestop.ini and restart again. Only since then has this issue appeared.
Not saying that anything regressed between TS 9.5 and 9.6, but since restarting with a new ini file has produced this issue.

One more issue I have is I get EDP Other under RING limits section. It is not as often as TVB, but it is sometimes there. I had read some older posts as well in which you said to users to put IccMax to 255.75, and it should probably solve the issue. It has definitely reduced since then, but it still comes from time to time(like 5-10% of the time). EDP Other also used to come under CORE, but IccMax to 255.75 has eradicated that limit 100%.

 

[unclewebb]

Your CPU generated a single TVB throttling message during almost two and a half hours of log data. Why do you think this is important? I recommend turning your computer off and going outside once and a while to enjoy life. Problem solved.

it should probably solve the issue

Yes, set IccMax to 255.75 and set Power Limit 4 to a value of 0 and that will solve a lot of EDP throttling issues. It will not solve all EDP throttling issues. Not all throttling problems can be solved. You have to learn to live with some minor issues. Life is not perfect. Neither are computers.

 

[hkgute99]

Your CPU generated a single TVB throttling message during almost two and a half hours of log data.

I actually had turned on log file option and then forgot . Then when I remember after a few hours, I did a little stress test on it by running ADS-B Exchange - track aircraft live website on my browser. Its actually just a plane tracking website, but when you do a max zoom out, to load all the planes, it actually acts like a mild stress test. Many previous times my undervolt had survived cinebench but failed on this.
Coming back to the topic, before running the stress test, I also had Speed Step EPP at 150, because I was studying and I don't want fans to rev up just because of some stupid windows/Dell service taking up CPU.
Just before running the test, I had turned SpeedStep EPP to 0, and that's the reason in past 2 hours there wasn't even a single TVB event. And yes, TVB limit only happens when I am stressing the CPU, mostly the limits section remains empty.

 

[unclewebb]

my undervolt had survived Cinebench

Cinebench is not a great stress test when adjusting voltages. My CPU needs a lot more voltage to be 100% stable compared to how much voltage I need to be Cinebench stable. I find the TS Bench 960M test can uncover an unstable voltage far sooner than Cinebench can.

TVB limit only happens when I am stressing the CPU

Does the log file show this happening once every two hours or is there constant TVB messages in the log file even with Thermal Velocity Boost disabled in the FIVR window? During your previous log file the load on the CPU was not consistent. The load on the CPU decreased a couple of seconds later. The temperature dropped to 55°C because the load decreased. This had nothing to do with any TVB throttling.

Run a consistent load like Cinebench. Does TVB throttling show up in the log file then?

 

[hkgute99]

So this is insane.
I changed EPP to 0. Then opened Cinebench. The fact that just minimizing and maximising the app repeatedly just produced TVB is insane.
And this is the limits section after Cinebench. The score was 4483, so still in the margin of error area. My max score till now was 4534 (iirc).
But I can swear that ever since I had changed the PL1 and PL2 to 45W and 60W respectively, PL2 never occurred. And it probably shouldn't since the max power draw was 48.7 W and PL2 is set at 60.
And this change has only occurred ever since I upgraded to TS 9.6 or the new .ini from which I have started, either of the two.
Also for some reason PROCHOT 98 shows up in red even though it is in the 78 to 80 C range. Is that normal? I don't remember seeing that before.
Maybe I will try to delete TS altogether and start up totally fresh again. Maybe that can help....



I am sorry for posting again. But I found something.
So I have actually never even touched this Power Balance setting. Even though I had asked about it earlier, I never bothered to tweak it.
But this setting has actually changed by itself. No jk. Could this be the issue of TVB throttling?
And how does this setting even work?
I have quoted a screenshot from before, which shows that first column has 9 and second has 13. Now first has 16 and other has 7.

(CURRENT SCREENSHOT)

View attachment 298094
(SCREENSHOT FROM EARLIER)

 

[unclewebb]

Try not to forget that your 8665U CPU has a 15W TDP rating. The log file shows that it reached 48.7W or more than three times its rated TDP.

Intel® Core™ i7-8665U Processor (8M Cache, up to 4.80 GHz) - Product Specifications | Intel

Intel® Core™ i7-8665U Processor (8M Cache, up to 4.80 GHz) quick reference with specifications, features, and technologies.

ark.intel.com

Be happy with the extra performance you have been able to get out of this CPU. A few minor things lighting up in Limit Reasons is not that important when you are getting way more performance out of a CPU than it was designed to deliver.

Lots of Intel CPUs can trigger random yellow boxes in Limit Reasons at random times. At the moment my Limit Reasons screen shows PL1 in yellow under the CORE column. I have both power limits set to 300W and I have not run any hard core software to trigger this. This probably happened during the last sleep resume cycle before the power limits were properly setup. It is nothing that causes me to lose any sleep.

I have not done any Windows 11 testing. Perhaps a new driver for the Intel GPU is changing the Power Balance settings. If this is not changing your overall performance, it is not important. If you want the Power Balance CPU register to be set to 9 and 13 then check the Power Balance box and enter 9 and 13. Problem solved. This is designed to balance the power going to the CPU cores vs the power going to the Intel GPU. Your power limits are set sky high so there is no need to be concerned about power balancing the CPU and Intel GPU. They both have access to lots of power. If you want to raise the power limit for the Intel GPU, edit the ThrottleStop.INI file and add this line.

IGPU=50

That line will raise the Intel GPU power limit to 50W.

TS 9.6 added support for newer 13th Gen CPUs and it fixed some very minor bugs. There is no fundamental difference TS 9.6 and TS 9.5 so use whatever version you like. I recommend using TS 9.6.

Some of the random issues you are having are just that, random issues. No Intel CPU is 100% bug free. They all have a long list of known issues including lots of issues that are either not known or not documented and ignored. Your CPU is performing well beyond expectations so stop worrying about minor issues that cannot be fixed. Go run Userbenchmark. Your computer is probably going to score at the 100th percentile or very close to it.

Edit - Your log file shows that your cooler can only handle more than 45W for a few seconds. Leaving the turbo time limit at the default 28 seconds makes no sense. Your cooler cannot handle dissipating 45W to 60W for 28 seconds. Lots of thermal throttling in the log file. It only takes a couple of seconds running at full power like this to trigger thermal throttling. That is why PROCHOT is easy to light up red on the main screen. Setting the turbo power limits sky high overwhelms your available cooling.

 

[hkgute99]

Can you tell that what 9 represents and what 13 represents? Basically which is the cpu and which is the GPU in the power balance column.

 

[unclewebb]

My 10850K also defaults to 9 and 13. The number 13 is closest to the iGPU label so that must be for the Intel GPU.

These numbers are like a teeter totter. The higher number directs more power to the CPU or GPU. As far as I know, the only time this setting is important is when the CPU Package is forced to power limit throttle. These values determine whether the CPU cores or the Intel GPU should throttle first. Your power limits are set sky high so I do not think this setting is going to make any difference.

No one ever seems to use this setting. Let me know if you learn anything new.

 

[WhenMusicAttacks]

No idea. Just remember to undervolt both the Intel GPU and the iGPU Unslice equally.

Hi, do you mean same value? As i undervolt maybe 10 machines every week - even though most using voltageshift on macos, but the same applied in windows with throttlestop - the usual pattern for intel skylake based machines (so up to 10th nn G) seem to be CPU and iGPU highest value around 150, cache and igpu unslice about 90 and then system agent around 70.. so igpu unslice is much lower than actual igpu itself. In voltageshift i have it named as "Analogy I/O" though