The ThrottleStop Guide

Let's talk BCLK. Monitoring programs that show the BCLK floating around all over the place are not accurate. Every time you click on the BCLK button in ThrottleStop, it recalculates the BCLK and it tries to do this as accurately as possible. Most of the time it will display a consistent number with variation of only +/- 0.0001 or +/- 0.0002 MHz. I left a tiny amount of error in the calculation just to keep people happy. Users expect the BCLK to float around a little. If they do not see this, I thought they might get suspicious.

My conclusion is that it is a waste of CPU cycles to try and calculate the BCLK continuously. Any variation being reported is monitoring error. When properly calculated, the actual BCLK MHz tends to be extremely stable.

I recently discovered how to read the BCLK directly from modern Intel CPUs. I was excited. No more need to calculate the BCLK. Then I saw the fine print. Intel's internal BCLK calculation is only accurate to +/- 0.1%. In other words, a 100.000 MHz BCLK will be displayed somewhere between 99.9 MHz and 100.1 MHz. That might be good enough for Intel and many monitoring programs that are using this data but it is not good enough for ThrottleStop. My fans would publicly roast me in the forums if ThrottleStop displayed data with that level of accuracy.

 

Also, almost all the clock sources on a laptop come from one oscillator that is feeding a multiple output PLL that then generates the needed reference clocks that the CPU in turn feeds into various PLLs to generate all the internal clocks, without a known good external reference you might even just be reading the same clock against itself from different sources with more or less skew and jitter.

Want to know the exact clock, stick at least an oscilloscope at the clock pins of the PCIe bus.

 

If I've understood correctly there's been some discussion here about Dell/Intel blocking undervolting with their latest BIOS versions.

I just updated my XPS 13 9380 to the latest BIOS (according to Dell it's 1.10 for this machine, 4/2020 version). This is the one in which Dell/Intel blocked undervolting.

I want to confirm that the "Restore BIOS factory settings" tweak re-enabled my undervolting for Throttlestop! I found this suggestion from several Reddit threads, but since this very helpful little program originated from here I wanted to share my findings to here as well.

Immediately after the BIOS update TS or HWInfo wouldn't show any undervolting, but after factory resetting the new BIOS they started to work again.

 

Hello @unclewebb
Is that suppose to work? I mean getting different voltage on cache and core?
XTU is always setting same voltage on both, as far as I was always told both voltages needs to be equal.
Would you be so kindly to clarify that case for me?

BTW if you're looking for beta testers of your unreleased or preview version I'm willing to help.

 

Oscillators and also quartz are not that simple, in fact there are more limiting factors.
There is no only jitter, but phase noise - phase noise if frequency dependent.
PPM is also important parameter - high PPM will basically mean that your clock/oscillator will suffer from higher base frequency deviation than a part with low PPM.
Some oscillators are TCXO (temperature compensated), some others are not, also these parts are aging - mean less precise after years of being used.
Internal PLL's mostly using simple quartz to generate what they need, but there is no harm if you feed them from oscillator.

 

I does work, and it works fine.
I'm using different voltages on core and cache for years now, and I started doing that on my i7-4720HQ, doesn't like to have its cache much lower than -30mV or I get random BSODs, the core can do around -50mV, not the best core, but if I put both core and cache at -50mV in 5 minutes it crashes.

 

It turns out that everyone has been brain washed by Intel. If you say the same thing over and over again, that does not mean it is true. It is possible that the Intel XTU programmers do not understand how Intel CPUs work.

There are two separate registers in the CPU that contain offset voltage information. One register for the core and one register for the cache. Newer CPUs like the 8750H and 9750H seem to really benefit from setting the core offset voltage much lower than the cache offset voltage. It is the cache voltage that is the limiting factor. Typical success stories have the cache at -125 mV and people continue to see improvements in performance or temperatures with the core at -200 mV. Some users have found that there seems to be some sort of ratio of about 2:1. I have not come across any Intel documentation that suggests this is possible.

To test this, you need to use software that heavily uses the AVX instructions. Cinebench R20 is the perfect test. Improvements found in R20 will also show up as improvements in many modern games. Here is an interesting test sent to me by a user. First with the core and cache set equally and then with the core set lower and lower again.

Cinebench R20 - 122 mV
https://drive.google.com/file/d/1qq1BigJd_NIMRiFIRtyOBRbF0qDI0RDW/view?usp=sharing

Cinebench R20 - 160 mV
https://drive.google.com/file/d/1_W_Khc94O28XMW11s5Z5wd4AdRTg1fv3/view?usp=sharing

Cinebench R20 - 210 mV
https://drive.google.com/file/d/1DEiRoModKMmeIjBXaq3ZFQCWJqCEIFF3/view?usp=sharing

There appears to be a definite and repeatable difference when these voltages are not set equally.
I trust user testing more than I trust Intel or XTU.

Got the new font and color pickers hooked up this weekend. There is even a new option to ditch the title bar if you do not like to look at it. I am thinking that TS 9.1 will be ready early this week. Stay tuned.

 

Hi,
could anyone help me with Throttlestop settings on Lenovo X1E gen 1, i7-8750H? When I try to run Cinebench R20, the laptop powers down during the test just before test is finished. No error message, no info in log files. It worked perfectly already, so I assume that I made some bad settings there. I use Throttlestop 9.0, screenshots attached, log file here:

http://www.pkral.cz/download/log.txt

Any info appreciated.

 

When you have a throttling problem, you have to take a screenshot of ThrottleStop with Limit Reasons open while this throttling problem is happening.

Edit - Thanks for the log file. Your CPU is too hot. It is thermal throttling when fully loaded. You have to improve your cooling, replace thermal paste, etc. or you will not be able to achieve maximum performance.

Look in the right column of the log file. It says TEMP, TEMP, TEMP and it shows your CPU running at 97°C which is the thermal throttling temperature that your CPU is set to. I think that explains what is going on.

Read my last post. The one just before your post. The one that shows the advantage of setting the core and cache offset voltages to different values. You might want to try doing that. When first testing, perhaps a little less on the cache. This should allow much more offset voltage for the core.

Default turbo ratios for your CPU are 41, 41, 40, 40, 39, 39. No need to set these higher since the 8750H is a locked CPU.

 

Hi, Many thanks for your advice!. I tried to set the cache to -122 and core to -160, just like on your second example. Cinebench test has finished now, with score around 2400. I also set the speed shift EPP to 80 and reverted the turbo settings. I unchecked BDPROCHOT and removed PROCHOT offset.

The temp is still high, but it finished without shutting down now ;-)

Log file is here http://www.pkral.cz/download/log2.txt




Btw, Im not sure about my TPL settings as I copied it from some web article. Could you check it for me? Many thanks!

 

That is just one example of what worked well for one user's laptop. If your cache is stable at -140 mV or -150 mV, maybe your CPU core will be stable in Cinebench R20 at -200 mV or -225 mV. You have to test your laptop to see what works best.

If your laptop is good with PROCHOT Offset set to 3, check PROCHOT Offset and check the Lock PROCHOT Offset option. This will lock this adjustment.

Your CPU is too hot. It is thermal throttling. This reduces performance. If you want a faster laptop, you need to fix the cooling. Your power limits are OK. Maybe 65 for long and 80 for the short turbo power limit is all your heatsink can handle. If you reduce the core voltage, you might get more performance and less heat.

 

Yeah, they fooled me too, lying Intel bastards. in throttlestop any very low voltage setting resulted in crash of my laptops (8750 HP Omen and 10875 Sager) except for cpu core. I could offset cpu core to -1v and nothing would happen even if monitoring programs said there was offset applied. Changing voltage of CPU core didn't work at all. Then tried XTU and as soon as I undervolted core voltage -200mv it crashed. I was happy, finally core voltage adjustment works (funny how crashing computer can make one happy), untilll I realised core and cache voltage was set together by Intel, but in reality only cache voltage was changed and it caused crash, core voltage still didn't change at all. So yes, cache and core voltages are set separate, as unclewebb says but it could fool anybody into thinking you under volt both if you only tried XTU and never used Throttlestop. The only remaining question is how do I fix this Intel made mess and be able to change core voltage, the only one that really counts under plundervolt .

 

@pete962 - When adjusting voltages, the core offset can be set to a bigger number than the cache offset. When making adjustments, always start with them synced together until you get a general understanding of how stable your CPU is. After that try adjusting the core offset further and leave the cache alone.

In Cinebench R20,

cache -125 mV and core -125 mV is OK
cache -125 mV and core -150 mV or higher is probably better

The opposite is not true. If you try this,

cache -125 mV and core -100 mV

the CPU will probably ignore the extra cache voltage and set them both internally to -100 mV.

If your BIOS is using a microcode update that has killed voltage control due to a Plundervolt fix, you will need to try to go back in time by installing the previous BIOS version. There are a few other tricks depending on the laptop model to help fully restore voltage control.

 

Wow, that is actually unique, never seen another benchmark do that now that you mentioned it.

 

Hello again,wanted to ask something.Yesterday I installed Fallen Order, I randomly checked Throttlestop and then I see Max Power:100.4W. Never seen it drawing so much power.Any other enthusiast with 9750h passed this value?

 

i've only gone as high as 86W in blender. my VRMs will hit temperature limits shortly after.

curious that a game would hit 100W on CPU. doing a quick google there's various posts on it though, top few results suggests origin / steam overlay is doing something weird in background.

 

Probably just a spike, nothing to worry about. Try running linpack extreme and see how high it goes, it is one of the most stressful stress tests you can run.

 

Running CB R20 my 9750H stays in the 80W range, with a -120mV undervolt.

 

So far the random bench seems to be working took a bit of time but i didn't crash maybe i have to do more different tests on this laptop.

 

That is a good thing. The 8950HK CPUs are likely better binned compared to the similar 8750H and 9750H. Your undervolt settings are not as aggressive as some people are using.

When PROCHOT 97°C lights up in red, this indicates thermal throttling has occurred. Just like before, you can click on this to reset this information that is stored in the CPU. I removed the little check box but this feature works the same as in previous versions.

If you do some Cinebench R20 testing, try setting the CPU core voltage lower than the cache. These two offset voltages do not need to be synced. This can help improve scores or reduce heat.

Thanks for posting some pics. I think the next TS version is ready to be released.

 

I'll do cinebench tests tomorrow with different settings like you said, see if any crashes occur, also how far apart should the core and cache be?

 

I just realized something that I don't understand: core cpu voltage on 2 of my laptops (intel 8750h and 10875) can be adjusted but in a limited way. No matter how low do I set voltage offset, it won't go lower even in idle than about 532 mV. Originally that made me thinking the voltage adjustment is not working at all, since I could not set it low enough to crash the computer. But now I can see that I can offset it to higher voltage, no problem (for example I can set idle at 890mV) and max undervolt offset I ever got was about -100mV regardless of settings. Not ideal, but better than nothing. So my question is how it's done, since on my older laptop I reloaded 2018 BIOS, reset it multiple times and also deleted mcupdate from system32 to no avail. I know that laptop can do -170mV, all I need is to defeat that stupid lock. Any ideas?

 

Are you adjusting both the core and cache offset voltages together? That will crash your computer if you go too low.

Some users have said that the core offset voltage has to be set between 1:1 with the cache offset voltage and 2:1 compared to the cache offset voltage. Going outside of these limits will be ignored by the CPU. If -100 mV is OK for the cache, you can adjust the CPU core offset voltage between -100 mV and -200 mV. If you go outside this range, the excess will be ignored.

When going as high as 2:1, this only seems to work on some software that heavily uses the AVX instructions. On some software, there seems to be no difference between 1:1 voltages set equal or 2:1. When testing with Cinebench R20, there seems to be a repeatable difference between setting the voltages equally and setting a much larger offset voltage for the cache.

-125 mV for the cache and -200 mV for the core are popular numbers for Cinebench R20. These numbers are not written in stone. Every CPU is unique. Many users have had success with voltage numbers very close to these. If you go wild and crazy and set the offset voltage to -300 mV or -400 mV or -500 mV; this is way beyond a 2:1 ratio so this excess will be ignored. The voltage sliders in ThrottleStop are only for requesting a voltage. The CPU makes the final decision and decides, based on your request, how much voltage to actually use.

 

Waiting for 9.1
BTW: I tested Cinebench R20 on my PC with core lowered voltage up to -250mV (cache at -125mV) and I barelly get little better result.
So I don't know is it working or not at the moment.

-125/130mV is my maximum stable undervolting point for cache, going lower will make my system little unstable, reaching -150mV will make it highly unstable.

I've got 9th gen i7-9750h on my laptop.

 

Hello @unclewebb, I noticed bit strange behavior on cpu.
Recently I was playing with disc cloning, so I put the friend hard drive to my machine. Win started, but system hanged on VGA driver and running out of Hyper-threading because of friends i5 2th gen laptop but me is i7 3th gen, so I have been booting just on 4 cores.

The TS was installed so I run bench for fun and surprisingly got 3800 Mhz on all 4 cores, so I took pics and forgot about it.

Now remembered it today so was turning out of Hyper-threading just for testing, run TS bench and do you know what?
I got 3500 Mhz on 4 cores exactly how i7-3770s max specs are supposed to be.
The thing is PlatformInfo register is hardly locked (bit 28 I haven't figured out yet) so I did not get reason for this behavior.
Could you pls shared your opinion on it?
Thanks a lot and sorry for poor pics and my bad English.



https://imgur.com/b5cqEHT

https://imgur.com/TRQ9SdW

 

During this benchmark, is your CPU throttling? Power limit throttling or thermal throttling? Open up Limit Reasons while the benchmark is running and watch. It would be great if you could run a couple of log files while testing with Cinebench R20. How about one log file with the two voltages set equally at -125 mV and a second log file while the cache is at -125 mV and the core is at -250 mV. If you do not see a change in performance, you might see a change in your maximum temperature.

When testing on a laptop, it can be difficult to get consistent results if your CPU is throttling. If you quickly do back to back Cinebench R20 tests, performance can drop significantly even when you do not make any changes to your setup. The amount of turbo boost can be reduced during back to back tests so it helps to watch the multiplier in a log file.

@golovkin - Run CPU-Z while testing. Both of your tests are using 4 threads but I think one test is using 4 cores - 4 threads while the other test is using 2 cores - 4 threads. The first test disabled hyper threading. The second test has hyper threading enabled but your BIOS disabled 2 of your 4 cores. CPU-Z will confirm cores and threads. The maximum multiplier when 2 cores are active is 38 and the maximum when 4 cores are active is 35 so I am pretty sure that is what is going on.

I think the 3770S supports limited overclocking. I am not sure if that option is available in your BIOS. Post a screenshot of the turbo ratio limit window. How high can these multipliers be set? Limited overclocking allows you to increase your turbo ratios +4 bins compared to normal. If this feature is not locked by your BIOS, I might know a trick to do some factory overclocking. Try adding this line to the ThrottleStop.INI configuration file and restart ThrottleStop.

LockPowerLimits=1

 

Thank you for answering Uncleweb. I was not aware of any correlation between cache voltage and core voltage, as a matter of fact when I got my Omen in 2018 I don't remember even bothering with cache voltage, I just adjusted core voltage and after testing I found between -150mV to -170mV to be stable and bellow -210mV it would crash within min. and -250mV causing immediate crash. Now, no matter how low the offset, it won't crash ever and it seems the best it can do is about -80mV (my cache is set at -70mV and the laptop will crash if I set it too low). Is this part of the plundervolt fix? It would make sense, instead of blocking undervolting altogether, just put the hard floor how low can you go so CPU won't crash, problem solved.

 

@pete962 - CPU voltage requirements change over time. The changes you have noticed are probably not Plundervolt related. The Plundervolt fix disables CPU voltage control. Your laptop still allows voltage control so you are not Plundervolt compliant.

Are your core temps up over 90°C when testing? Temperatures also play a part in stability. As a CPU wears in, perhaps its ability to run reliably at high temps decreases.

If you were originally not aware that there is a relationship between core and cache voltages then maybe some of the conclusions you came to were not valid. The relationship between core and cache is not fully documented or fully understood. Just something to keep in mind.

 

Here is an interesting example.



The individual cores are being reported at temperatures between 37°C and 39°C.

Why does ThrottleStop show 42°C below that? Must be a bug right? Nope, everything is OK.

This is showing that the hottest spot on the CPU Package is 42°C. It might be the Intel GPU or it might be some other temperature sensor somewhere on the CPU Package that is reporting this higher temperature. The individual cores are mostly idle (C0% = 0.2) so none of them are the hottest temperature within the CPU package.

As soon as you put a big load on the CPU, then you should see the highest core temperature and the peak package temperature reporting the exact same thing. The hottest spot within the CPU package should definitely be coming from one of the cores when fully loaded.

 

Just curious - is there exist a way to report iGPU temp?
When TS 9.1 will be available?

 

I am pretty sure that GPU-Z reports the iGPU temp correctly when on the GPU-Z Sensors tab and it is set at the bottom left to monitor the iGPU.

Hopefully tomorrow. TS 9.1 is done but I still need to test it on as many different computers as possible. No problems so far.

Here is a list of changes if you are interested.

https://pastebin.com/9GEQNvdj

 

If you need someone to test basically anything I'm willing to help you. Me and MSI laptop are ready.
Considering that I'm using your app for something about 3,5 years it would be nice to contribute even a little.

 

I untderstand what you're saying Uncleweb, but I bought this Omen laptop in 2018 and back then, when I set it up, it would crash right away when core voltage offset was bellow -250mV. I had it running for few weeks at -170mV offset, but it would crash once or twice a week, so finally I ended up with -150mV and that was good, no more crashes and I left it like that. Fast forward to spring 2020, the same computer, no matter what core voltage I set in your program, it won't go lower than 533mV and it would never crash, no matter how crazy the offset is. This only applies to core voltage, all other voltages act the same as before, you set offset too low and computer will crash. Now, if I was Intel engineer, this is exactly how I would fix plundervolt in all gaming laptops: limit the lowest core voltage you can set to the level that won't crash cpu but leave rest of voltage adjustments intact and problem solved. So here is my speculation: at some point Intel changed the way core voltage is adjusted and put limit how low it can go, either as direct response to plundervolt or for some other reason and because of this, they don't need to block voltage adjustment on the laptops like mine and my laptop is fully updated to latest BIOS (April 2020) and Windows. I don't believe Intel would just leave my computer vulnerable. Other computers that have total voltage adjustments disabled, either can not be set to have minimum voltage set, or Intel expects them to be used in more secure environment than gaming laptops and set higher security standard. The way my computer works now is I can't set core voltage low enough to make it crash and flip the bits (this is the way plundervolt works I think) so I speculate I'm protected from plundervolt.
BTW love your program, keep up good work.

 

I have Intel Cometlake Sager, coffelake Omen and haswell MSI, if you need help with testing, let me know.

 

@unclewebb I just wanted to thank you for this amazing tool, and the effort you keep putting into it. I've been reading this thread for a while, and it's so cool to see you continually developing this, and giving people support for 10 years now. Where's your donation info? I'd happily throw you a couple bucks for your work.

I've been using Throttlestop for a year now on my Dell Precision 5540 (Basically an XPS 7590), with an i9 9980HK, and Quadro T2000. The i9 is arguably not a great match for the XPS form factor. Recently I've started looking into ways to squeeze a little more performance out of the CPU. I was hoping you could maybe give me some advice.

This is a log of a Cinebench R20 benchmark (scored 3531). My settings were as follows. Any thoughts on it?

I have a couple specific questions also, that I hope you can help me out with. Sorry, if it's a bit much:

1. I see PL1 limits the TDP to 56W. When searching online I came across this blogpost explaining how to disable the power limit altogether. This doesn't seem to do anything on my system, however. Is there something else I can try, or do you reckon there is some low level limit at work?
2. I've repasted the system with Kryonaut, but that doesn't seem to make much of a difference. When I look at the Max temperature during the benchmark, there's about 5 degrees difference between the cores; some running at 100C while others are at 95C. Do you think this is an acceptable difference in core temperatures, or did I inadvertently make things worse than what it was before? (Unfortunately I don't have a record of what the differences were before repasting.)
3. I'm considering repasting with liquid metal to help with the thermals, but I'm trying to figure out if it's worth the risk/trouble. Do you think it would help with performance, or does it not matter much if the system is going to be power limit throttling shortly afterwards anyways?
4. Is it advisable to increase the PROCHOT offset for my system? I see the CPU is rated for 100C max on Intel's site, so I'm guessing it would probably not be a great idea.
5. Does Throttlestop have a way to show when the system throttles, because the VRMs are running too hot?

BTW looking forward to the 9.1 update with the TS Bench random frequency option!

 

@golovkin - Run CPU-Z while testing. Both of your tests are using 4 threads but I think one test is using 4 cores - 4 threads while the other test is using 2 cores - 4 threads. The first test disabled hyper threading. The second test has hyper threading enabled but your BIOS disabled 2 of your 4 cores. CPU-Z will confirm cores and threads. The maximum multiplier when 2 cores are active is 38 and the maximum when 4 cores are active is 35 so I am pretty sure that is what is going on.

I think the 3770S supports limited overclocking. I am not sure if that option is available in your BIOS. Post a screenshot of the turbo ratio limit window. How high can these multipliers be set? Limited overclocking allows you to increase your turbo ratios +4 bins compared to normal. If this feature is not locked by your BIOS, I might know a trick to do some factory overclocking. Try adding this line to the ThrottleStop.INI configuration file and restart ThrottleStop.

LockPowerLimits=1[/QUOTE]

Hello again @unclewebb
thanks for the explanation, you was absolutely right, so I confused two things. At first example the machine was booting wit Hyper - off and secondly booting just on 2C/2T due to way how Windows was configured for i5 cpu.
Secondly thanks lot for your tip, so I added your line to TS INI file + added Rwe driver, get started TS + monitoring Mmio range 15900 but to no avail.
Multis and Mmio was looking the same way as before test. I got some clue LockPowerLimits=1 is supposed to do 15990 reg editing but it is already at FF (max).
Also strange is, since I added that line, the TS complained about -error creating service- so maybe it is not accessing memory correctly due to some bad configuration from my side.
In any case, I will be happy if you would give me some info, perhaps in the message in regard what address to manually tune and which way.
In regard of limited overclocking, yes the cpu i7-3770s is capable and the TS is showing + 400 Mhz but no Bios access on OEM H61 nor unlock bios menu, so I dirty patched flex ratio reg manually but as you can see in TPL table the Multis are locked, not able to write into.
Your idea you shared regarding the lock is on Bios side is very helpful to me, because it also supporting my similar findings. Here is link to list from which bit 16 or 17 could do the task.
https://lkml.org/lkml/2019/9/2/579
I added pics of TPL table
https://imgur.com/W96CjTd
The true is I don't definitely need that limited overclocking for my daily tasks, but on the other hand, I don't want to deal with the fact that the processor is not working properly as expected.
Thanks for your work and sorry for bad English.

 

Even after 10 years, I am usually good for a few suggestions.

1) You have the long turbo power limit (PL1) set to 56W in ThrottleStop. Your log file shows your CPU PL1 throttling at 56W so your CPU is doing what you are telling it to do. When you set the long turbo power limit higher in ThrottleStop, can you get beyond 56W long term? Some Dell laptops will set a hard power limit. If this is the case, you will not be able to use ThrottleStop to get beyond a hard limit that Dell has set.

I disagree with Brad's Hacks suggestion to disable the short power limit check box. This might work on his CPU but this trick might not work on all CPUs. I prefer to leave both power limits checked with the power limit values set appropriately. I know this works. His method might work too. I have not tested a wide enough range of CPUs and motherboards to make sure that his method always works.

2) A 5°C difference between fully loaded cores is normal. These temperature sensors are not 100% accurate temperature monitoring devices. Even if your paste job was perfect, there is still some slight differences from core to core because of temperature sensor accuracy. These are not space shuttle calibre sensors. They do not have to be. Intel only used to rate their sensors to +/- 5°C so you are well within that spec.

3) Using liquid metal might help you achieve a tiny bit more performance but the 56W PL1 power limit is still going to kick in when fully loaded. Hardly worth bothering.

4) Your CPU already has the thermal throttling temperature set to 100°C which Intel lists as the maximum safe temperature. You cannot use ThrottleStop to go beyond 100°C and I would not recommend trying to do this even if you could. Some desktop boards have BIOS options for some Intel CPUs so you can get beyond the 100°C barrier. Not sure if any laptops have this option.

5) If you open up Limit Reasons, it should show you when throttling is voltage regulator related. Your log file only shows CPU thermal throttling and then PL1 power limit throttling kicks in soon after that.

@golovkin - Setting LockPowerLimits to 1 just makes sure that FED15990 is set to 0xFF so this register does not limit the maximum multiplier. If your computer is already set to 0xFF then there is no reason to use this ThrottleStop INI option.

Your CPU supports overclocking but you need both an overclockable CPU combined with a chipset that supports overclocking. I do not believe the H61 chipset supports overclocking. ThrottleStop shows that your CPU supports +4 bins of overclocking but the turbo ratio limit adjusters in ThrottleStop are all locked to their default values. They are probably locked because the chipset does not support overclocking.

Error Creating Service error you get in ThrottleStop happens when you are trying to run ThrottleStop and RWEverything at the same time. There is a conflict because they are both trying to access the same RwDrv driver. ThrottleStop 9.0 uses a new driver so you should be able to use it and RWEverything at the same time without any problems.

@pete962 - Perhaps a BIOS update changed how your CPU handles voltage requests. It is impossible to say what happened. The original Plundervolt fix disabled CPU voltage control while in Windows. Maybe Intel got enough negative feedback that they found a way to be Plundervolt compliant and still allow users to adjust their CPU voltage while in Windows. There was lots of doom and gloom last December when the original Plundervolt fix was released. Voltage control is still working OK for most users, including on Intel's 10th Gen mobile CPUs.

Almost forgot. Try setting the CPU offset voltage to a much bigger number compared to the cache offset voltage. These do not have to be set equal. You might gain some Cinebench points for your troubles. The 8750H and 9750H are often times stable even when the core voltage is up around -200 mV.

 

@unclewebb Huge thanks for the detailed response!

None of the power limit settings inside of Throttlestop seem to do anything. Even though the PL1 is set higher than 56W, after 28 seconds the 56W limit is being enforced. I even went through the trouble of trying to set the levels with Intel XTU, even though that probably works the same way as TS. Alas, no dice. The only way a user has any control over the power limits is through Dell's own Power Manager, it seems. The Ultra Performance profile in Dell Power Manager sets the PL2 and 1 at 107W, and 56W respectively, while the Cool, and Quiet profiles set PL2 and 1 at 37W, and 17W respectively. I guess this is a dead end then.

I also looked at Throttling while both the CPU and GPU were under load. This was while playing a game for a couple of minutes. Even though the PL1 is never reached, it still occasionally hits the thermal limit. I'm guessing that's because the cooling system has the extra heat of the GPU to deal with on top of the CPU heat.

A part of me still wants to try out the liquid metal, even though the risk probably far outweighs the reward . For 100% CPU load it may just help thermals a bit for 28 seconds before PL1 throttling kicks in, but I'm thinking it might improve the thermal performance when both the CPU and GPU are under load. I'll do some further testing to see when I actually encounter throttling during real world use, which is video editing, and animation.

I've tried increasing the CPU offset voltage in a 2:1 fashion, and that does help a bit!

Again, thanks for the help.

 

Could you try with dell power manager uninstalled?

 

ThrottleStop 9.1
https://drive.google.com/file/d/1ve03GyMmRv45od3pjr4K0btqboCovp5z/view?usp=sharing

New Features
- updated TS Bench test with random MHz option.
- fixed BCLK MHz reporting when using Core Isolation.
- added support for devices that use connected standby.
- added reporting of suspend / resume times to the log file.
- removed auto BCLK updates when resuming.
- fixed GDI handle leak.
- new color and font options.
- new black notification area icon option.
- new option to remove the title bar.
- removed PROCHOT indicator box.

 

Good call! Unfortunately uninstalling didn't help, though . Appreciate the advice!

 

@WigsterR, did you also reboot afterward?

 

Got a startup failed thingy with v9.1:

Spoiler: ThrottleStop v9.1 - user32.dll error; RegisterSuspendResumeNotification function missing

Version on my installation (Windows 7):

Spoiler: user32.dll v6.1.7601.23594

Microsoft's help page:
RegisterSuspendResumeNotification function

So ... guess it needs Windows 8+ for this particular option?

 

I knew the Register Suspend Resume Notification function was not Window 7 compatible. I found some code on the internet so TS 9.1 was supposed to check for compatibility before trying to run that function. I will have to come up with a plan B for Windows 7 users. I will send you a beta when it is ready if you would like to do some Windows 7 testing for me.

Once I get the above bug fixed, the suspend / resume times should show up in the log file when using Windows 7.

Thanks for the bug report.

 

Thanks UncleWebb, for all your good work! TS has consistently helped me get more use out of laptops that had inadequate cooling - from my first experience with Alienware m15x (big time throttling!) to my current Gigabyte, to my wife's Lenovo Extreme V2. And your patient responses to all of us have set an example to all. BTW, 9.1 fixed the font issue I was having - since it occurred on machines with 4k screens with scaling set, I think it was a scaling problem.
Have a great weekend!

 

The simple MFC programming tools that ThrottleStop uses are rooted in the Windows 95 era. Windows scaling on these old school apps is horrible. Glad to hear that the new font options are making things a little less blurry for you.

 

I found this bug on my daughter's laptop.



The C0% used to be reported correctly. After doing some testing, I discovered that a timer within the CPU that used to work correctly is now broken. It is not really a ThrottleStop bug. More of a bug within the CPU.

The timer that ThrottleStop uses to calculate C0% has been working exactly the same in every Core i CPU produced during the last 12 years. Never had a problem getting some extremely accurate C0% data by using this timer. My best guess is that there must have been a microcode update that suddenly changed how this timer operates.

I have seen screenshots of this bug but no one has bothered to report it to me. If you have this bug and want to do some testing, let me know.

I found a different timer that seems to be working well. It would be nice to run the updated code on some hardware that has this bug. Maybe this new bug only happens on early 8th Gen U CPUs.

Edit - Just saw a screenshot of a 9750H that also showed this bug. Glad I was able to come up with a quick fix.

0.5% in the C0 state when idle looks a little more believable.

 

Is this what I think it is? In ver 9, when my computer went to sleep (close lid or time out) ThrottleStop would unload and all voltages revert to normal. will this reload TS voltages on it's own? If so that would be cool.

 

@pete962 - Older versions of ThrottleStop never knew when a computer was going into and coming out of connected standby mode. If the BIOS reset the voltages, it would take a while before ThrottleStop would notice this.

ThrottleStop 9.1 is connected standby aware. This new version should update the voltages immediately when a computer resumes. If you enable the Log File option and do a suspend resume cycle, the log file should report these events so you can confirm that it is working OK.

When testing, remember to exit ThrottleStop after you resume so it has a chance to write this data to the log file.

 

Thanks Unclewebb, I can also see voltages in HWmonitor and can tell right away if they're set or not and I think new version is working properly.