The ThrottleStop Guide

I can confirm the new feature works great in my Gigabyte Aero 15. BTW, if it is not too hard to implement, just remember that there are many people wishing the feature of different TPL settings per profile

 

@unclewebb Have you considered making TPL settings part of profiles? It'd be useful to be able to set different PL1/PL2/PL3/PL4 limits depending on which profile is active.
For example, my CPU supports cTDP down which allows me to throttle it to 35W and 2.0 GHz (from the regular 60W and 2.5 GHz non-turbo) - it'd be great if I could enable that every time my battery level dropped below a certain percentage.

 

Hi, after a few months of trying, playing with many EPP settings and limiting the Turbo Boost ratio all cores of my i7-9750H to 3.9 Ghz instead of 4 Ghz to avoid playing at 90° or above, I found a much more viable solution with the latest beta of ThrottleStop.
I just forced my notebook at a PROCHOT offset of 15, so the maximum temperature is now 85° C, while keep Speed Shift EPP at 64.
In this way the CPU is always seeking for maximum performance while avoiding too high temperatures. The clock is Constantly staying in the 3.5/3.8 range, with a few 4 Ghz spikes, and then temperature never goes north of 85° (with just a few 88° spikes).

Do you think it is safe for a CPU to operate near 100° for a prolonged time ?
I know, Intel is giving OEMs a limit of 100°, So it should be safe to operate at 100° all the time and the CPU will protect itself, but we don’t know how close to a “structural” limit that number is.
Because I think the point is that: many i7 and i9 from the 8th, 9th or 10th gen are operating at 90° and above when full loaded on high performance laptops.
Is that really safe on a time frame of several years ? I think only Intel knows the answer.

maybe Lenovo went a little too far with a PROCHOT of 80°, but I basically did the same (I set 85°) and to be honest performance are not impacted (in some cases like gaming are actually even higher) but I think the CPU health is preserved.

What do you think on the matter ?

 

To me thats not normal to run 100% of gaming time at 100°c. Yeah Intel is right, their chips are reliable for 100°c but I think thats too much tbh. My cpu hardly ever passes 80°c on gaming, with undervolting and some other tweaks.

 

At laptop voltages 100°C is safe (albeit undesireable) for 5+ years of use.

And again, i think 100°C will only be hit during heavy loads that stress the cpu to 100% but most of the time the cpu should run at 60-80°C so im perfectly fine with laptops having a 100°C temp limit. Thats much better than having the cpu thermal throttle prematurely.

I'd actually be very happy if nvidia sets their gpus to thermal throttle at 100°C rather than 87°C

 

You can adjust Nvidia temp limit with asus gpu tweak 2.
It was avalible on Pascal gpus so rtx should be also tweakable.
I am missing this option in afterburner. I don't know why it is not implemented in afterburnerwhile it is in gpu tweak

 

Yeah i know but limit is 87°C, you can't set it to 100°C

 

Hmmm... I remember seeing this issue on Broadwell chip. I thought it was normal. I will test out TS 8.75b once her online university exams are over. Then, I will pull a Win 10 update and drivers trick(s) to test out newer TS.

 

Hey guys, a couple of pages back I asked a question about how to limit Turbo and the solution I found was going into the TPL setting. Unfortunately this not longer seems to do anything. Previously if I capped it at 3ghz, my system would consistently perform at that speed however now the speed is all over the place.

 

if you have 6th gen CPUs, use speedshift values 130/150 to lock it above 2.x GHz. Its a trial and error method.

 

I prefer to keep that limit to 85°, even if 100° is safe.
As I said, while gaming the CPU never goes below 3.5 Ghz, being most of the time at 3.8/3.9 Ghz. Not much throttling here (many laptops are keeping below 3 Ghz ...) even with PROCHOT at 85° C.

I had two choices: play with EPP (with a setting of 110 the temperatures were around 80/85°) or set PROCHOT at 85°C. The latter allow for better performance.

 

Hi, I've tried using google and the forum's search function to find the answers to my ? but see it wasn't really explored here. All ?'s are in relation to improving battery life. My laptop only supports c7 or c7s according to TS. I'm looking to maximize c7 states while unplugged. I understand that TS allows even c state interrupt settings to be adjusted according to profile. The tool I've found that helps the most in identifying setting changes that improve battery life is hwinfo64's average battery life. My c7 states in battery saver mode are cycling between 70%-80%. The situation is pretty bad when I'm plugged in, I see spikes of 10-20W just by opening a browser and it cycling down from there instead of stepping down once task is complete. ?'s are as follows in order of importance:
****MY C STATE INTERRUPT SETTINGS ARE AS FOLLOWS: 78 118 148 all with a time unit of 2 1024 ns.***********

1) c_ Interrupt Response Limit and time unit, what does it mean?
2) Does increasing time limit and time unit for higher c states improve higher c state residency?
3) What are ideal values for I7-9750H?
4) regarding EPP speedshift and speedstep. I've seen it recommended by unclewebb to disable speedstep on newer processors, and give speedshift full rains. I believe my TS automatically checks off both despite my inputs, does having both checked actually make any difference?
5) Does having different TS/Windows performance preference policies cause conflict? Ie: If TS is set to epp of 51 should windows be set to 20%. I assume 99% of TS users have it set to default windows balanced power settings.
6) How can I eliminate the power spikes when opening up a browser or other program for example? I see my pkg pwr hover up to max and jump around there for a bit longer than I'm used to seeing. IME it's supposed to jump down relatively quickly.
7) C state demotions/undemotions, are any of these options relevant to be unchecked in the interest of improving battery life?
8) I'm also completely lost about the tpl window, particularly the primary plane power limits, and miscellaneous section. I know how important the three boxes at the top are, and have them dialed in well. I would like to at least make sure if I have the relevant boxes checked, like intel power balance or pp0 power limit.
9) Is there any way to stabilize my package power, currently it cycles a lot (.4W-1W at complete idle)? Windows have hundreds of processor power management settings, that affect the processors parking, idling and various other features that are well beyond my understanding, is adjusting EPP and C state settings the best option when using TS?
10) How do I manipulate my c state interrupt for battery profile vs plugged in?

Thanks to unclewebb and the helpful NR community here.

 

There is a very long and helpful thread in the Windows OS and Software Forum on configuring for maximum battery life. Definitely worth reading. My take is that the biggest dividends in battery life come from eliminating programs that needlessly run in the background. Spikes are not necessarily undesirable - when you open a browser, you would rather have the processor speed up, get the job done, and return to resting state. Others will be able to help more concerning TS settings than I can.
Keep us posted.

Joe

 

@unclewebb Could you maybe add by time a 2nd systray info text indicator for the PKG power, without the decimal, so just "5" or "15".

 

Thank you Joe, I have gone through that thread which is indeed helpful. It is where I found the driver that was causing issue. I’ve also fixed the high pkg pwr while plugged in by upgrading to nvidea’s driver and ditching the 3rd party drivers from guru3d (which people prefer due to stripped telemetry data).

The ISSUE is that the topics I mentioned are mostly to do with Throttlestop, and how it may interact with windows power management. @unclewebb your reply to my ?’s posted above would be greatly appreciated. The windows battery thread doesn’t make mention of power management settings for example windows power preference policy. You can find a list of hidden power settings and the command line to enable them all online. TS is a billion times easier to use then manually adjusting windows power settings. The fact that MS doesn’t explain these settings online should indicate there reluctance to have end users modify them. Windows allows you to adjust so many aspects of core parking, unparking, idling, and power transitions it will make you want to shut your computer off before you change a setting

 

Virtually no one in this forum during the last 1000 pages has reported any use for the IRQ settings. They will likely be removed someday. I have never had access to a 9750H to see what if anything these settings do. I currently have my 4th Gen set to 64, 96, 192 which are similar to the default settings. I think I went with those numbers because in the land of binary, they are all equally divisible by 2. Science FTW!!

4) When Speed Shift is enabled, it does not seem to make any difference whether the older SpeedStep control method is enabled or not. I usually leave it enabled just for tradition purposes.

5) On 8th Gen and newer CPUs, why not let Windows 10 control the Speed Shift EPP value? There is no need to check the Speed Shift EPP box on the main ThrottleStop screen. Windows and TS both writing different info to the same CPU register is not cool. Watch the FIVR window to see what Speed Shift EPP value your CPU is actually using. If you do not have access to the Windows High Performance power profile, try using this command in a command window.

powercfg /s SCHEME_MIN

That might encourage Windows to set Speed Shift EPP to 0 for maximum performance.

6) If your browser is set to launch 101 tabs when you open it then a spike in CPU usage and power consumption cannot be avoided.

7) I do not think that C state demotions and undemotions make any significant difference to anything.

8) The Intel Power Balance and TDP Level Control are mostly for the U series of CPUs. On a 9750H, ignore these and ignore the PP0 Power Limit. Some settings in TS were added just in case some OEM came up with a not so good throttling idea.

9) Your power consumption cycles at idle because the CPU is processing various background tasks. Reduce the number of useless background tasks and your CPU will cycle less. An idle CPU can have cores spending over 99% of their time in the low power core C7 state. Put your time and effort into increasing that important number.

Post a screenshot of the TPL window if you need more help.

I agree 100%. Watching core C state activity and comparing that to the Task Manager Details tab is the best way to hunt down and eliminate crap.

@maffle - I just had a look in the Options window. Perhaps there is enough room in there to add an extra icon to the Notification Area so you can look at package power without having to open up ThrottleStop. Not sure when this new feature will be finished but it is already near the top of the things to do list.

 

Greetings all. I'm doing some research on core offset vs. core cache. Specifically, over on the Alienware subreddit, another user and I are discussing Throttlestop set ups and most, if not all tutorials say that the offset, cache, and igpu slice should be set to the same value. The user reported that they had greater success with the cache and offset set to different values. I tried this out myself and noticed an immediate improvement when using the built in TS Bench.

My initial, tutorial guided settings were -.1446 on the offset/cache/igpu and yielded a TS Bench 12 thread/64M score of around 8 seconds. Using these alternate recommendations, I set the offset to -.250 and left the cache/igpu at -.1446; TS Bench came back with a 6 second score. I ran the longer bench and saw a likewise improvement from 31 seconds to around 24 with the differential settings.

Can anyone with more knowledge about these settings explain what's going on here?

 

You will get a better response if you include some basic details about your system and CPU m8. I suppose people could search your posts but that is not really in the spirit of asking for help IMHO.

Also add some screenshots of ThrottleStop

 

My bad. I'm currently at work and don't have my laptop with me. It's an Alienware M17 R1, i7 9750H. I'm running the most current version of Throttlestop.

EDIT: I didn't think that information was necessary as I was asking about the relationship when undervolting between the core offset and core cache.

 

No problem.

It seems the CPU type might matter with respect to matching undervolts but this is neither publicly documented by Intel nor certain. You just need to try to see what works on your laptop.

 

I am not sure I understand. What do you mean by "offset"? Do you mean CPU? And what did you do with iGPU and iGPU Unslice? Did you use the same values?

Thanks for clarifying.
Joe

 

Under the FIVR tab you can set the CPU core voltage offset, which I believe is the base undervolt. This is where my questions come from. What do these three settings actually do to the CPU? If the offset is the base undervolt, how does that relate to the cache setting? The tutorials that I've come across state the core/cache/igpu should all be set to the same value (with the only deviation being the igpu for those systems that may not have it enabled, this can be skipped). It's clear to me that after my experiment, this isn't the case. I set the core offset to -.250 and left the cache/igpu at -.144 and saw an improvement in performance, further reduction in temps, no effect on turbo boost, and no negative impact on stability.

My hope is someone can explain better what these three functions do and how they work together. If this is behavior specific to 8, 9, 10th gen Intel, that could be useful information too.

 

@Gumwars - I think the big lie that the core and cache offset voltages must always be set equal was started by Intel XTU users. XTU syncs those two voltages so everyone was fooled into believing that these two voltages must always be in sync.

While working on ThrottleStop, I discovered that the core and cache offset voltages could be adjusted independently within the CPU. There are two separate registers within the CPU that contains offset voltage information for the core and cache. At one point during TS development, I re-programmed ThrottleStop to sync these two voltages so ThrottleStop could be just like XTU. That did not go over well.

Users immediately complained and they were right. There is no reason for software to sync the core and cache offset voltages. There are some Intel CPUs that show no benefit when you adjust these voltages independently but for the 8750H and 9750H, that is not the case. Being able to adjust the CPU core offset more than the CPU cache offset has shown benefits in many benchmarks and games and has shown a reduction in CPU temperatures. It is a win win situation compared to Intel XTU's decision to lock these two voltages together.

This topic has been brought up before in this ThrottleStop thread but not a lot of user testing was posted. I do not own either of these CPUs so I had to sit on the sidelines and could not contribute any new information to this subject.

A user recently contacted me after noticing the same thing you noticed. I suggested that he run some tests using Cinebench R20. I thought it would be a good idea to verify his offset voltage tests by running an independent benchmark. Everyone knows and trusts Cinebench results. The TS Bench is not in the same league for popularity.

For a baseline test, he set both the core and cache offsets to -122.1 mV. Cinebench reported a score of 2547 points.

For the next two tests, he kept the cache offset consistent at -122.1 mV but dropped the core offset first to -160.2 mV and then to -210.0 mV. His Cinebench results improved to 2621 points and then to 2725 points.

https://i.imgur.com/q8FkqcG.png
https://i.imgur.com/ABFawZk.png
https://i.imgur.com/MR940CV.png

His laptop has a hard 45W TDP limit so his scores are not on par with HWBot scores for this CPU. Even so, there seems to be a definite advantage to setting the core and cache offset voltages independently.

I am not sure if 8th and 9th Gen desktop boards allow you to run independent voltages. Some desktop motherboard manufacturers were probably forced to drink the same Intel Kool-Aid and decided to sync these two voltages together. My 7th Gen Asus desktop board has these voltages synced. There appears to be no reason to do that.

Thank you Marc S. for sharing your results.

 

@unclewebb Thank you! This was exactly the information I was looking for!

 

Thank you for bringing this subject up again. Hopefully you can post your own Cinebench R20 results. I brought this subject up on Reddit recently and was ignored.

There is a reason why many users with top scores on HWBot are using ThrottleStop. It is a big advantage on newer CPUs. The 10th Gen are probably the same but will have to wait and see.

 

I'm at work presently but I'll be back with some screen caps.

 

OP of that thread here. The reason I didn't bother listing the options is because I didn't find changing those settings to have a significant impact on my system. I did state that users should try experimenting with the options themselves. That thread was written almost two years ago, and since then, absolutely zero has changed, although I'm now realising that the CPU is in fact the biggest controllable element in battery life. I wrote the guide so that users would get the lowest system idle power. Increased power consumption when doing tasks is unavoidable, and so I ignored that section.

 

The top two scorers on hwbot for the i7 9750h are Mr. Fox and Prema and they are getting those scores mainly due to bclk overclocking and memory overclocking.

I can place 3rd on the leaderboard but i haven't bothered to make an account. Even then my scores are quite far away from theirs.

 

Have a look at the 3259 point Cinebench R20 benchmark score that @Mr. Fox posted.
There was no BCLK overclocking involved.

https://hwbot.org/submission/4257469_mr._fox_cinebench___r20_core_i7_9750h_3259_marks/

 

Indeed. ThrottleStop has been a key ingredient in achieving maximum CPU performance for me, especially with laptops that have been artificially crippled by the imbeciles that build them.

Without it I can barely reach 100W with the crappy locked 9750H. Using ThrottleStop I can squeeze 135W from that crippled critter. I have to launch it quickly, before the cancerous gremlins in the EC are awakened by the demonic forces lurking in the Intel ME shadows.

 

My mistake, that was just Prema but memory overclocking is what really pushes the cb scores above just unlimiting the cpu power draw. And yes, throttlestop is elemental in that process.

@Mr.Fox weren't you were running a bclk overclock on prema's bios as well?

 

Correction on the 135W.

 

No way, that's not possible. Even in AVX workloads the maximum I get is 85-90w at 4ghz all core. 75-80w after undervolting.

Thats an erroneous reading without question.

 

You're mistaken.

 

You're welcome to back your claims.

This is clearly not a correct reading. Check if your imon slope and imon offset are correct. Imon slope should be 100 and imon offset should be 0.

A regular 9750h at 4ghz all core will not go over 90w much less 135w+

 

You are assuming I am running a stock BIOS. If I were, your assumptions would be correct because it is artificially capped at the value you observe with your system due to cancer firmware. That's according to Intel spec, which laptop manufacturers tend to go with, so what you see is normal. It works as intended. What I do is abnormal. Normal is inadequate. And, what I do is not possible without ThrottleStop because the EC has to be blocked to keep it from overriding the BIOS power limits. ThrottleStop does that if I launch it immediately upon reaching my desktop. If I wait about 2 seconds the EC cancer kicks in and snips the nuts off the CPU and limits it to about 90W.

It is not an incorrect reading. I have my power limits set to 2000W, which effectively allows the CPU to pull as many watts as it is capable of pulling. The readings are verified by XTU, ThrottleStop, HWiNFO64 and the Kill-A-Watt meter the AC adapter is connected to. My system is only crippled by a BGA garbage CPU with locked multipliers. All of the cancer firmware castration crap has been mitigated (although not entirely eliminated, unfortunately). The best hardware in the world is worthless when firmware has been coded by idiotic scumbags.




You'll see a minor variance between software monitoring tools because they poll sensors at different intervals and none are 100% precise when software is involved.

But, that's still pretty mickey-mouse compared to the 1200+ W my 7980XE sucks from the wall outlet running CBR20 at 5.2GHz.

 

Assuming your readings are correct, you would have the worst binned 9750H in existence. Higher power draw thats not backed by higher performance is meaningless.

But logically speaking i dont think thats the case, the readings are just plain wrong. I'm not sure whats causing it on your system, just that the readings are off.

 

Feel free to believe whatever you wish. That's fine. And, I'll enjoy being at the top of the leaderboard for the crippled 9750H BGA turd CPU. Can't make a silk purse from a sow's ear, but the numbers don't lie. Bin quality theories don't matter all that much when you're winning. If you're not, then poor silicon bin quality might play a part in the losing streak.

More watts, more voltage, more performance. Less watts, less voltage, less performance. Always works out that way for me... at least for the last decade or so. Bin quality only seems to matter when it is poor enough to limit your maximum clock speed. Otherwise, it's just a matter of goosing the voltage a little more to make it scream.

 

I posted this at least once before, maybe more, in this thread. But, it's a massive thread and you could have easily missed it.

If I don't launch ThrottleStop immediately to intercept whatever nonsense occurs with the EC and Intel ME demons, look what happens to the power limits when the digital cancer metastasizes. Once that happens, you're barely going to manage to eek 90W out of that wimpy 9750H turd CPU. BIOS power limits are overridden and the tail starts wagging the dog. But, I wouldn't be able to do this at all without an unlocked @Prema BIOS and ThrottleStop working together to overcome the artificially imposed performance limits. I'd have a run-of-the-mill crippled turdbook instead of a somewhat less-crippled turdbook.

It's unfortunate that the core multipliers are not fully unlocked. It would be interesting to see what it could do if they were not. Might even be able to push it hard enough to pop some capacitors or blow out some MOSFETs on the motherboard before tripping the circuit breaker in the AC adapter.

 

Could you please run same 1024M TS benchmark? The last drop performance from BGA junk will need its power to keep clocks at max. Maybe Fox then can re-run for validating the scores and power consumption from your chips.

 

I'm using an imon slope tweak and offset to bypass the cpu power limit so it will show a very low wattage (like 12w at all core 4ghz) but that's the only way I can prevent power limit throttling as Throttlestop can't change the power limits on MSI laptops.

At 4ghz constant (imon tweak) my TS bench score is 86 seconds.

If I reset my imon, undervolt and run TS bench i'll hit the 65w pl2 limit so the cpu won't be running at 4ghz throughout the benchmark.

When pl1 is active the cpu draws 65-70w at 4ghz constant.

 

@Mr. Fox what do you mean by imediate TS launch. Did you change something in task scheduler to run in faster or some regedit tweak?
Could you guide?

 

@Mr. Fox

Share your tweaks then - my 9750H isn't hitting anywhere near that heavily modded.

 

With my stock 2666Mhz and just a bit of undervolt I scored 31xx on CB R20 with a 9750H,not even touched imon slope yet..

And my cpu was using around 82W if I'm not mistaken. No way I'm going to spend days tweaking ram timings to achieve 0.5% better score..

 

How can you reach 4ghz all core on a 9750H with no IMON slope tweak? The thing is limited at 45w ...

 

What you can do to get 4ghz all core depends on your laptop. On most MSI laptops the TDP controls are locked down and the only way to remove the power limit is to do the imon slope tweak.

On other brands you are able to change the TDP limits using throttlestop. Msi laptops also seem to have different PL1 and PL2 limits depending on the sku.

For example in this video the GE 65 has a 55w PL2 while the GL 65 has a 42w PL2 limit
8:48 onwards

 

I keep thinking I'm still posting on the Razer Blade forums ... woops. We're locked down over here, only way is heavy BIOS mods - I thought he was on a RB15/17 with a BCLK overclock, as far as I know that's impossible on a Razer product.

 

Currently i have a c940 (i7-1065G7) and it is working alright at -85mv and mostly 128 speedshift. Thanks for your great work and dedication.

I just wondered how the built in windows10 power plan options and the power slider on the battery icon is affecting throttlestop. What exactly does the slider? Does it change the speed shift settings or is it basically useless if throttelstop is already running?

 

You are welcome.

Clear the Speed Shift EPP box on the main ThrottleStop screen. Open up the FIVR window and watch the Speed Shift EPP value in the monitoring table. That is the value that your CPU is currently using. Now take the Windows power slider and switch it from one side to the other and watch what happens to the Speed Shift EPP value. Does it change? For newer CPUs, it is often times easier to just let Windows control the EPP variable.

If you want ThrottleStop to control EPP, you might have to set the Windows slider all the way to the right. Enable the ThrottleStop Speed Shift EPP option, enter different EPP values on the main screen and watch the FIVR monitoring table to see who is in charge of EPP. If EPP in the monitoring table is not showing a consistent value to what you are requesting, that means Windows and ThrottleStop are fighting over control of this variable which is not good. To avoid a fight, let Windows control EPP.

Edit - On some CPUs, setting EPP to 80 or 84 is better for performance compared to 128.

 

i did a littl ebit of testing about independent core and cache voltage and I do find interesting result.
My current laptop is 2019 Nitro 5 i5 9300H model. I did 3 test for each setup. This laptop had hard 56W power limit.

1st test is -160mv both core and cache. Max value it can go without Error in TS benchmark
https://imgur.com/gallery/LS3Z94k

2nd test is -155mv on cache and -245mv on core. Max value it can go without Error in TS benchmarkh
ttps://imgur.com/gallery/q9eH9ql

C20 score do increase but temperature looks the same to me.