The ThrottleStop Guide

From the Techinferno forums, UncleWebb is back!

Quote: "I was in the middle of adding some more 4th Gen voltage features to ThrottleStop when I decided to take a break. I don't plan to release anything new until I get the 4th Gen Haswell voltage features and the new Atom features finalized and thoroughly tested.

http://i.imgur.com/VM9dhgz.png

Now that NFL football season is finally over, I intend to resume development but it might be a week or two or a month until I have something ready to release.

There is a download link to ThrottleStop 7.00 in post #225. Hopefully he did a good job extending the expiry date. "

 

BlueSwift, thanks for reporting that.

Just a note though; the link to TS7.0 seems to be questionable, but at least unclewebb is still working on this great project of his.

 

Hi!
how I can fix frequency in my i7-4500U more then 2.4 hz whis TS 7 beta?
thanks

 

Good news.. Hopefully he keeps working on this awesome software..

 

I well agree with that. Without this awesome software I wouldn't be enjoying my computer and I'd probably be looking at something else more expensive.

Big props to unclewebb, keep up the good work!

 

Hi, In my signature there is a link how to use Throttlestop 7 b3 with 4th gen ULV CPUs.

 

Hi, thanks for the reply. However, as you said that lowering the multiplier is not a direct way to lower your CPU's TDP, could you tell me what is the more direct way or perhaps a better approach? If you are kind enough, you may aswell give us a step by step method.

I'm going to buy Gigabyte P34 v3, with a superb CPU, but not so powerful battery.

Thanks in advance.

 

For those that haven't already tried it, ThrottleStop may also work on your desktop CPU. It needs more rows in the main window and turbo ratio limits tab for all of the multipliers, but it still has about 75% functionality for the 4930K in my Clevo P570WM. Voltage adjustments are not available, but changing all multipliers simultaneously, power limits, clock modulation and chipset clock modulation are working. Disabling C1E and overclocking profiles also work, which is excellent. I used it in the Fire Strike run below. I ran Tests 1 and 2 at 4.0GHz and the Physics and Combined Tests at 4.6GHz.

 

Sounds like your BIOS is locked more than anything else. XTU simply allows changes open in the BIOS, and it's possible your BIOS doesn't support above-stock speeds, so that might be what happened. Unfortunate as it may be, this is why Lenovo machines suck for high end gaming/tasks.

 

@unclewebb - Although the tables are missing the extra cores and threads, ThrottleStop is still very useful for my 4930K hexacore desktop CPU. Hopefully, once you have time, it will not be too difficult to update the UI to accommodate CPUs with more than 4 cores, 8 threads.

Here is a video showing how I am using ThrottleStop with my Clevo P570WM. (This should be helpful for anyone that hasn't figured out how to use this amazing program.)

 

Thank you for this!! Another great video and now I know how to set up Throttlstop properly.. Was wondering why I was crashing it.. I had EIST turned off (oops).. Thank you again Mr Fox! Keep doing these videos

Also + rep but it's not allowing me.. I will give it later

 

so even with throttlestop the i7 4700mq will work maxim 3.2ghz x4 cores?
because mine is working allready all the time 3.2ghz x4cores while gaming...

 

Hello everybody!

I'm struggling with a problem for weeks now, maybe somebody could help here...

I have an Acer laptop with an i5-4210u + g840M, and the problem is that if both the CPU and the GPU is stressed, heavy throttling occurs. If I remember correctly, this kind of problem (unnecessary CPU throttling because of the GPU) was among the initial causes for Unclewebb for creating TS in the first place, a few years ago.

So a little description about the very interesting behaviour of this machine:

The 4210u performs very well alone. Under heavy stress (linpack) power consumption goes above 20W, tops at 22W but no throttling occurs at all, it can work at this turbo freq. for unlimited time. I did stability testing when I tried undervolting it, the super - accurate "log TDP" box wasn't ticked even after 10 hours, while it continuously operated at the maximum turbo freq. for 2 active cores. (XTU neither showed any throttling but it is way less sensitive for micro-throttling) Same goes for the 840M, even with ~30% memory clock OC and constant turbo GPU core clock speed, it performs flawless for hours. Nice. The temps are also unbelievable (coming from arrandale world) ~35c idle (without the fan even powering on), ~50 loaded, after 10 hours of linpack max temp was 60c. The GPU stress test generates ~30% CPU load, and the temp never even hit 60c, tops around 58. Ambient temp around 22c. Very nice.

The problem occurs when both components are under load. 28 seconds pass, and BOTH components frequency is throttled to around 50%.

EDIT: that is visible also in XTU as "Power limit throttling", TS as "TDP throttling" , and via decreased load on the GPU in MSI Afterburner just to be clear.

So it is obviously some power limit hitting in, but I can't figure out where does it come from, and how could I get rid of it. After a little testing, it turned out that if the GPU is disabled in the device manager, the 4210u starts to behave like how I think it should by default: There is some short time limit which is around 68 (no typo) seconds. Interesting, since the short power limit should be 28 sec, and I did not alter it. But even after trying to mangle with it, and unlocked the short power limit time window (and both the short and long power limits) in the BIOS and setting it higher in XTU (and TS, and finally in the BIOS itself) the same thing happens. Without the GPU, the CPU begins throttling after roughly 100 seconds if it consumes more than 15W.

The power brick is 65W so that should be enough too, the 840M doesn't consume more than 30W.

I've read a post from Atom Ant, in some topic here on the forums about a 4210u + AMD Radeon card, where the solution to enable both components to use @100% is setting the custom TDP level to 2, but it does not help here. Even configuring the custom TDP levels much higher (25+) don't have any effect on the 68s turbo limit if the CPU used alone, or the 32s when both components are loaded.

The first thing comes to my mind is this behaviour comes from a hard coded long power limit to the CPU itself, but that can't be the situation because it can happily consume 20+W for hours when the GPU is enabled (but switched off by Optimus)

Maybe some gurus here could help me looking after the source of this throttling, and eliminating it somehow, this is my last hope for now.

Here is the TDP configuration from the BIOS which I currently use

I'm using the default TDP level 2. The ratio is "correct", the CPU idles around 2.4 ghz which is what I want, usually the laptop is docked, and I can easily lower the idle ratio for battery use. But if I use TDP level 0, or 1, there is no way I can move the idling freq higher, than the default ratios described above. I can also limit it with TS easily to get the "idling" freq. lower.

Here is a pic of TS and the power limits window with this TDP level setting



Advanced CPU power settings in BIOS page 1


Advanced power setting in BIOS page 2


One of the custom cTDP configs I tried

The decimal value 200 translates to 25W both short, and long power limit. "long" term time window (It must be a typo instead of short) is set to 28 seconds, but it does only kick in after 108. With the "Turbo Activation Ratio I can set the idling ratio to 20.

Any advice is welcome here guys!

 

Your power adaptor is inadequate would be my first guess.

Have you tried one with double the wattage?

 

If you have your laptop set to high performance in power options and nvidia control to performance and your max temps are 60? you shouldnt have throttening...

 

I have not, but as soon as I get one I will. I tought that if it is capable to maintain power @ full load for both CPU and GPU for 30 seconds, it is OK, and if it is not then it would cause some serious crash, not just throttling. But I will try that for sure

Both settings are as described, of course.

EDIT:

I've also done some more experimenting, and it looks like something is fishy with the cuscom TDP profiles that I set up in the BIOS.

I've set up all 3, like this:


These settings should: set both short, and long term power limits to 30W, and 35W on all 3 profiles. The short term power windows are 10, 28, and 128 seconds, and 3 different Turbo Activation Ratio settings for testing. I think that setting should determine the idling ratio.

Power limits are corect in the BIOS thenafter, howewer profile 0 should have "20" as the Turbo Activation but it's 22 instead

Pic1 by lakondas posted Feb 18, 2015 at 9:52 PM

But after switching across the 3 profiles in TS, the turbo ratios of the profiles are respectively 17, 8, and 24 just like in the default cTDP profiles. Interesting, but TS7 is beta, maybe there is lack of support for activating the custom TDP levels.

 

Sorry for not being able to squeeze all this info in the first post, but here is some summarization. Everything is done with the default power levels configured by Acer's BIOS. No messing with custom TDP settings at all, you can check the BIOS settings on the first 2 pictures in my first post, but everything important is visible in XTU also.Power level 2 is used, the CPU idles @2.4Ghz.

CPU stressed like hell with linpack, GPU enabled but switched off by Optimus

Good temps, no throttling at all, the baby can do this for 10+ hours

GPU disabled, CPU stressed "only" with XTU

Throttling after ~45 seconds back to 15W. The CPU uses only 16-17W while stressed with XTU

GPU stressed with MSI Kombustor

Good temps, no throttling at all, but notice that the CPU backed down to 800MHz as soon as the dGPU is utilized by Optimus. Sometimes there is also a very small "CPU power limit throttling" bump (1%) visible at the exact time this happens. CPU stays @800mhz during the whole test, which is quite interesting because it idles @2.4GHz without any load thanks to Power Level 2. This throttling does not make sense because the CPU consumes only 4-5w at the time (idling @ 2.4GHz and of course 98% in deep C state), and also while running only @800mhz. This could be kind of a clue maybe...

GPU burning finished, and is deactivated by Optimus

Notice the small "Power limit throttling" bump again, and the CPU goes back to the normal idling multiplier.

CPU & GPU stressed simultaneously

The temps aren't even hitting 70c, and 13 seconds after the CPU usage hits 100% the throttling hell is unleashed. CPU turbo is gone, and GPU usage fluctuates between 50 - 70% . This is the problem. Of course this happens with real gaming usage too, not only when loaded like this. Also, this is happening while using internal screen, external with VGA, and HDMI too.

Sorry for the big pictures.

 

I think is because of a crapy bios, this are the settings that they put...

 

So I have done some deeper investigation with MSR Tool, and the results are quite interesting.

The MSR register 0x690 contains the "Performance Limiting Reasons" which is just what I'm trying to determine so I re-done all the testing, while observing this register.

With disabled dGPU, and the CPU idling @2.4Ghz, the package consumes ~5W, and no throttling is visible on the clock speeds, but the register shows evidence of "Package-Level PL2 Power Limiting" happening very often, in every 2-3 seconds.

With disabled dGPU, and the CPU stressed to consume more than 15W, after ~48 seconds the unit starts throttling because "Package-Level Power Limiting PL1"

With enabled dGPU (but not utilized so switched off by Optimus), and the CPU idling @2.4Ghz The package consumes ~5W, no throttling occurs at all, no traces of any limiting in MSR0x690

With enabled dGPU (but not utilized so switched off by Optimus), and the CPU stressed to consume more than 15W there is no throttling at all, no traces of any limiting in MSR0x690

When the dGPU gets stressed, one short "Package-Level PL2 Power Limiting" happens visible in the registers log field), the CPU clocks reduce to 800Mhz (but no TDP level switching to 1 happens, it remains in 2 so the idling freq. should stay at 2.4Ghz), and no more CPU limiting is visible any more during the stress test.

When both components are stressed, there is "Package-Level Power Limiting PL1" happening after a couple of seconds and both units are throttled like f*ck

MSR0x640 "MSR PP1 POWER LIMIT" is all zeros, all the time, so I assume PP1 and PP2 has nothing to do with the PL1 and PL2 limiting that is happening.

But I've found out that MSR0x610 "MSR_PKG_POWER_LIMIT" althrough it's not documented under the supported MSR registers by Haswell, it has both PL1 and PL2 settings in it which are the following:

Package Power Limit #1 = 25W
Time Window for Power Limit #1 = 3.6 million seconds, so roughly 42 days
Enable Power Limit #1 YES
Package Clamping Limitation #1 YES

Package Power Limit #2 = 25W
Time Window for Power Limit #2 = 2440 microseconds (0.00244 seconds)
Enable Power Limit #2 YES
Package Clamping Limitation #2 NO

And the register is not locked according to the lock bit.

So the settings in MSR0x610 "MSR_PKG_POWER_LIMIT" suit the BIOS settings regarding the power limits, but not the throttling behaviour that is actually happening.

While the dGPU is disabled, the long power limit looks like 15W, this value is identifiable easily so later if I will have some time I will start investigating undocumented MSR registers trying to find ones thats structure resembles "MSR_PKG_POWER_LIMIT" or "MSR PP1 POWER LIMIT", and the power limit value is 15W. But since the reason for the throttling is indicated in MSR0x690 "Performance Limiting Reasons" as PL1 and PL2 power limiting, and the power limits are set to 25W for them in MSR0x610 "MSR_PKG_POWER_LIMIT" and of course there are 3rd party power mangling software installed like Acer ePower, the results of this investigation just made me more confused.

I've compared a few MSR dumps, and there is no change triggered by enabling/disabling the GPU in any of them.

Some info about how other machines with the same GPU + haswell behave would be nice, maybe somebody around?

 

I am still alive! Just taking a break from the forums this year. I do plan to release some more ThrottleStop versions but progress will probably be slow this year. As soon as something is ready to be tested, a link will be posted here. Thanks for all the positive feedback.

lakondas - The problem you are running into is that there is a duplicate set of registers that control the power limits and the TDP Level limits. These secondary registers are programmed via memory mapped I/O. The version of the WinRing0 library that gives me access to the CPU registers used to have a function to access the Memory Mapped I/O but this function was removed from WinRing0 for security reasons. WinRing0 is no longer in development and older versions are impossible to find on the internet.

That is the first wall that I hit. ThrottleStop 7 can not access this secondary set of power control registers. The secondary set of registers take precedence over the similar registers in the CPU that you can access with the MSR Tool.

The MSR Tool was something I quickly through together for testing purposes. It has a few bugs so you would be better off trying to use RW Everything. The problem I had with RW Everything is that it works fine on my laptop that has a 4700MQ but it crashes when I try to run it on my daughter's laptop that has a 3rd Gen low power U CPU. With no hardware or software for testing purposes, and no way for ThrottleStop to access the Memory Mapped I/O, there is nothing I can do to move this part of the project forward. If you can get RW Everything working, it will let you access the Memory Mapped I/O.

Dufus understands the Memory Mapped I/O stuff far better than I do but he is a busy guy and also doesn't have a 4th Gen U CPU to play with. If a manufacturer left all registers fully unlocked in the bios, it is possible that the performance of the U CPUs could be significantly improved by permanently locking the CPU into Level 2. I think there are some hard power limits set by Intel in these CPUs. You might have an unlocked bios and it might let you input some sky high power or time limit numbers but these values will be ignored by the low power U CPUs if you go beyond the hard coded limits.

At the moment, the majority of U CPUs in the majority of devices are not fit for any serious gaming. It is impossible to compare different U CPUs in different devices because manufacturers do not publish how low they have set the various power limits. The U CPUs can throttle down to a crawl and there is no software available that can fully fix this problem.

I wrote another tool called MSR Finder that you might be interested in.

https://www.sendspace.com/file/luzete

It lets you take a Snapshot of all of the important registers in the CPU. After something changes, you can click on the Compare button and it will take another Snapshot of the registers and compare them to see if anything has changed. It should write a new file to the log directory if it finds anything that has changed. It has helped me find a few interesting registers in the CPU but the main problem is still the Memory Mapped I/O registers. The pictures you are posting are huge and I am sure it is just a matter of time before the mods complain but thanks for the additional info. Without full access to all of the Intel documentation, this project became too big, too frustrating and too time consuming.

 

Welcome back! I'm really happy for You to return!

I have sources for v1.3.1b, if You'd like a copy PM me( After a quick peek into the headers it has "Memory (Special API)" functions)

RW Everything runs perfectly, althrough I don't know yet how, and what should I RW with it

I hope He shows up, althrough after todays researching, I have some basic understanding of the package RAPL,now I'm very confused about this whole thing and maybe He can help. Also I think the CPU is lockable into TDP Level 2 (as far as locking it via the "standard" MSR registers) I can switch between the 0-1-2, but alongside the "ratio" which is the idling ratio for the cores changes respectively (17-8-24), the power limits configured in them are not working I think, because throttling occurs (if it does) at a TDP level which I can not find in them.

In fact, I am quite sure that these cpu-s power limits are fully configurable. Just take a look at the performance when the dGPU is enabled, it does behave like a 25W TDP CPU. No throttling at all. And for the factory clocks, it is far enough to operate at max freq. Since it is not overclockable, at least the throttling could be eliminated. Also, I'm not really interested in using it with disabled dGPU, but since I've noticed that it causes different throttling behaviour, I'm using it for testing.

I've done the comparison via AIDA MSR dumps, but if I'd know about this tool my work could have been easier, anyway I will give it a go again with this tool too. Thanks for another great piece of software

EDIT: Did the comparison with MSR Finder, and it came up with 2 differences. MSR0x34 increments each time the dGPU is enabled / disabled in the device manager. Does not react to enabling and disabling wlan card. MSR0x3FC also changes, according to documentation, it is most likely some core C-state residency (MSR_CORE_C3_RESIDENCY in SB architecture).

EDIT2: So turns out Dufus had a forum topic for this problem ( MSR configs does not affect true power limits) here http://forum.notebookreview.com/threads/tdp-and-power-limiting-haswell.766743/ and with a solution. Too bad he left the forum in January. I'm quite stuck with this MCHBAR mangling right now :/

 

Here is a datasheet for the 4th Gen low power CPUs.

http://www.intel.com/content/dam/ww...obile-u-y-processor-lines-vol-2-datasheet.pdf

The individual TDP Level power limits are locked. The only way to get max power out of a low power U CPU is if you can lock the CPU into the highest possible TDP Level. This register is at offset 0x5F50 and is publicly documented by Intel.

http://i.imgur.com/swxN2Zf.png

Using RW Everything, you can click on the Memory Dump icon, third icon from the left. On many computers, the memory base location is at 0xFED1000. You need to add the above offset to that so the memory location you are interested in is at 0xFED15F50.

http://i.imgur.com/pzhai18.png

The screenshot is from my 4700MQ which doesn't use TDP Level Control. The leading 8 in that register is the lock bit. With the above documentation and RW Everything, you should be able to poke around and figure out some interesting stuff.

My theory is that if you have a U CPU and this lock bit is not set, you should be able to set this to TDP Level 2 and then also use the lock bit to make sure it doesn't change. You would then need to do the same with the MSR register in the CPU using ThrottleStop or the RW Everything MSR editing feature.

The WinRing0.dll I am using is version 1.3.1.19. I think that is the one in the WinRing download that you have. I think you need to go back to version 1.0 to get the function that is needed to fully control these CPUs.

Without a 4th Gen U CPU to play with and without access to an early version of WinRing0, I decided to work on other things instead. Hopefully you learn something interesting with your U CPU.

 

Thank You very much for the detailed help.

According that memory location, it is locked already into Level 2. The throttling behaviour also reflects this with the dGPU enabled, exactly until it gets utilized. I've also locked the MSR to level 2, but as soon as the dGPU gets utilized, the ratio behaviour reflects Level 1. No change in MSR nor 0xFED15F50, but the idling freq. drops to 800mhz. And after 48s stressing of both components, CPU is throttled below 15W TDP , also dGPU too somewhat. Tonight I will dig myself more deeper, hopefully to find something in the memory, but after checking out Dufus's researching, I'm afraid that the EC is responsible for this throttling, by setting the power levels via PECI...

 

That's the problem with the U CPUs. There are a variety of ways for a manufacturer to throttle them and software might not be able to do anything about that. I think the PECI settings take precedence over the MCHBAR and MSR values. When the CPU gets forced into TDP Level 1, the multiplier drops to 8 and you usually end up with a CPU running below its rated TDP power level. Very frustrating for consumers when they see that their core temperatures are fine and power consumption is fine but the CPU is stuck running at one third or less of its rated speed.

 

can I make i7 4710mq stik to 3.5ghjz 4cores? without puting my bios in no post mode...???
now it holds perfect 4 ores 3.3ghz with max temp 72c....24/7

 

I'm glad to see you're back @unclewebb! As per usual, anything I can test for you and Dufus, you simply need to ask. Send a PM if you feel like it! I'd gladly dedicate time to helping throttlestop or any project you're considering doing that I could help with.

 

I don't get where do you find the new ThrottleStop 7.00?

Also, is there any API to the TS? I want to create some algorithm that automatically takes into account throttling issues and react to them. Shouldn't be too hard, at least for many laptops

 

Indeed where it the link for the latest version? Also @unclewebb can you please release the beta etc with the timer updated?

Thanks,
Tom

 

Hey guys just getting into this post. I have a I7 4810MQ and no matter what i do the thing constantly throttles. it even throttles with the factory settings. Not sure where to start.

 

You can start with naming your notebook model, your O/S, BIOS and other pertinent info...

 

And also post the temperatures its reaching.. if it's always throttling, its probably overheating or a PSU issue..

 

@unclewebb I have a quick question about the capability, I was originally going to Pm you but, it says I can't so anyways. You can probably see it as my wanna be server in my signature. The optiplex 380 board sports a G41 chipset. This chipset supports the QX9650. This is an extreme edition processor. While the actual chipset supports this processor, the bios doesn't nessisarly allow the overclocking features, as it was never shipped with it originally from dell. I've had luck in the past putting processors that intel claims support into computers that state they don't. Now since the bios won't allow access to the voltage and multiplier or even the FSB. Will throttlestop single handedly be able to change the multipliers and the voltage if it recognizes it as an extreme cpu, regardless of the bios being capable of doing it or not?

 

Touche
It is an Ibuypower cz-27 which is an MSI whitebook MS-1763. I7 4810mq, 32gb ram, GTX 880M 8gb running windows 8.1 and as for my bios it has been flashed over to an unlocked version of the latest MSI bios for the 1763 mobo's. so essentially its an unlocked GT70 Dominator PC2.

As for temps and throttling, im running into either current limit or powerlimit throttling. Temps are 77-82 during xtu cpu stress test. Laptop cooler and PK3 TIM will be here Friday.

 

Trickster29 - For the Core 2 Duo and Quads, ThrottleStop writes multiplier and VID information directly to the CPU so in theory, it shouldn't matter what chipset or motherboard you have your QX9650 installed in.

Motherboards that allow you to change the CPU core voltage in the bios will usually ignore the VID setting. My QX9650 would be getting rather toasty if it was actually getting this much voltage but luckily the bios is ignoring the VID signal on this board.



Thanks to Dufus for introducing me to Core 2 Extreme overclocking.

 

Can somebody unser my question 2?
I74710mq in alienware 17r5 a14bios
Can worked 3.5ghz x 4cores 24/7 without breaking the bios?

 

The only reason your getting those failures is due to undervolting.. Your trying to maintain 3.5GHz and then Undervolting too much.. My approach would be to set clocks at 3.5GHz with stock voltage and see how well it holds up by running the XTU stress test.. If it's holding 3.5, then uv a bit.. Like -10mV steps and then -5mV steps etc till you find a stable point..

 

There is usually a too low voltage on processor that allows you get NO POST and bios will not start .. Some processors can not handle too low volts. Although the processor does not get too hot, it does not mean the processor overclocking very good.

 

You didnt understand me
I didnt try to see if it holds, I asked if anybody else tried allready? I will never try again to see on my procesor...I tried with the old one and everytime I got no post...so no way

 

Possible there are a few In Clevo thread with the same processor as you that can give you an answer. It is not crucial whether the processor is in an Alienware or Clevo. An Alienware 17R5 manages to cool down most of processors regardless Overclock on processor. Bios version in AW17 does not matter.

 

no topic on sager/clevo for this settings?
Nobody put in throttlestop 35x for all cores insted of 33x?

 

So? no guide?

 

You didn't ask anything... you need to be specific in what you want to do.

 

Look one post upper....

 

Look at this Trottlestop guide setup from Mr Fox. Only difference is that you must insert your own values for your processor. Many think Trottlestop is more stable to use than Xtu utilities.

 

You didn't say what your objective was, nor what CPU it was, nor the laptop it was, nor what problem you were trying to circumvent.

There are currently about six clevo lines in production: P6xxSx, PxxxSM-A, P7xxZM, WxxxSSx, W6xxSx and P570WM.

If you tell us exactly what you're trying to figure out if throttlestop can do, then we might be able to help.

 

So, bios oclvl1 end up with no post
Xtu oc end up with no post
I asked if I could get 4cores at 3.5ghz with throttlestop?

 

See xtu settings for your processor on different scores. Lots of i7-4710mq with 4x35 or higher. Use same settings from xtu in Trottlestop. http://hwbot.org/benchmark/xtu/rankings?hardwareTypeId=processor_3611&cores=4#start=0#interval=20

 

He asked for it for a clevo, which means the M17x R5 in his signature isn't the machine he's using it on.

 

Been away a bit and i still cannot get this thing to overclock at all for more than about 30 seconds before it TDP throttles.

 

So even Throttlestop adjusting TDP limits doesn't help? I replied in your original thread earlier about undervolting; that might be your next best step.