The ThrottleStop Guide

After reading those LinX numbers both Dufus and D2 Ultima have, I'm quite puzzled at how I'm getting noticeably lower numbers on my i7-4790K. For reference, here is my LinX result:


I should add that I'm running my CPU with Turbo Boost set to 4.5GHz, with 110 and 88 for Short and Long Package Power levels respectively. I do get TDP Throttle a bit after the bench starts, with the CPU locking itself at 4.0GHz.

My question is, given I've configured LinX settings (at least I believe so) to match both your settings, why am I getting drastically lower GFlops on a CPU that should theoretically have at the very least matching performance?

 

tribaljet - Open up LinX, click on Settings and try setting the Number of threads to 4 instead of 8. I think you get more GFlops out of LinX when you are not using hyper threading.

Also make sure you are using the exact same version of LinX. Different LinX 6.5 downloads will contain different versions of the Intel Linpack library which can make a big difference to GFlops.

D2 Ultima - Can't you just humor me and send me a ThrottleStop log file while doing a LinX run? I don't ask for much.
It helps me learn about Intel's CPUs and it also helps me learn about ThrottleStop.

 

Well you see... #MobileCPU4Lyf #ComeAtMeDesktopUsers #Mwahahahaha (I really have no idea).

Did you exit everything except linpack and possibly a monitoring program like HWiNFO64 before running it? And you may need to relax your power limits more; try setting your W limit to about 150W on long and short power. The desktop chips are designed to require more power to work.

Also, please note I NEVER throttle on the first run. Only on the third run (it at stock) or MAYBE at the very end of the second run for about 1 second (if I OC).

It doesn't make a difference with the threads. He correctly set it to 4 threads in the program, but he might have to set affinity in task manager, but even if I didn't set affinity and I set 4 cores I got basically the same scores. I also got the same scores when I set it to 8 threads and left affinity on 8. If he did 4 cores and 8 thread affinity, it'd just use ~50% of the CPU in general.

 

LOL where's that ThrottleStop log file, D2 Ultima?!

 

Didn't try anything yet. I'll do that in a bit XD.

 

You mean all these LinX runs you've been doing don't count?

 

I couldn't get it to do a log before, but it works now. I pastebin'd the log info for a run I just did, and another screenshot of it completed
TS log for Linpack - Pastebin.com <--- le pastebin

And as usual, the accompanying screenie:

Spoiler

 

Well, I made sure to replicate LinX settings, including setting 4 threads, but that's on the screenshot already But I believe you're spot on regarding different Linpack libraries.

Mobile CPUs are quite interesting, I'll give you that, and even my i7-2820QM did show me how mobile chips could perform admirably, outperforming many desktop chips

I did have some software idling on background, but none should impact performance as much as to cause performance numbers that lower when compared to both your mobile chips. In any case, I'm definitely inclined to accept how I just might've ran a less performing Linpack library through the same LinX version.

Regarding power limits, I was going with more conservative values due to worrying about reduced hardware lifespan. Question is, am I only going to face increased temperatures or is the increased power limit going to physically harm (even if on a long term scenario) my CPU?

And could it be possible to ask for 4/8 thread TS Bench 32/1024M results? That would be a more even playing field, unless anyone has a better suggestion for performance measurements and comparisons.

 

Thanks for the log D2 Ultima. I only had time for a quick look but you have some sort of power throttling going on. With power consumption at about 36 watts, the multiplier is at 35.00. When LinX starts working hard, power consumption goes up to about 78 watts and the multiplier drops to 33.00. When LinX starts going full bore, it is running the AVX instructions and this causes the CPU to increase the VID for improved stability. This is a known feature or issue with Haswell.

What you could try doing is use ThrottleStop to switch between Adaptive to Static voltage. Set a fixed Static voltage of about 0.9900 and try running LinX again. The desktop users are not big fans of Adaptive voltage when running LinX or Prime95. If you can lower the voltage without losing stability during this part of the test, your multiplier and GFlops might improve.

I did some testing and my 4700MQ shows more GFlops when LinX is set to 4 threads compared to 8 threads.

http://i.imgur.com/GS3ET2J.png

http://i.imgur.com/4a95n51.png

To avoid any thermal throttling it was time to head out to the garage and put the deep freeze to good use. Even in the deep freeze, 4 threads of LinX was enough to get the peak core temperature up to 94°C but I did get a few more GFlops for my effort.

http://i.imgur.com/JZG56kQ.png

 

Probably that might be the reason. Because see, when I'm streaming or using heavy x264, my voltage skyrockets. Linpack has never drawn as much core voltage as what this has, so I figured it wasn't voltage directly. I'll try psuedo streaming dark souls 2 and copy the voltage the CPU uses when compressing with "slow" at 616p & 42fps, and force that in TS, and see if linpack works better. If it does, I'll then try the 0.9900 and see if that works okay, then if not I'll start at 1.150v and slowly go down from there. But I'll do that later tonight. If I can keep voltage high and keep temps down in general use, I'll simply set a static voltage in the BIOS and be done with it. At least the benefit of my mythlogic BIOS is coming in; I do all my OCing/etc in there because I had problems with rebooting the PC using intel XTU in the past.

 

@unclewebb

I made this account just to talk to you. You seem like the kind of guy that doesn't groan at the thought of helping others

I have a Macbook Pro Retina 15" Late-2013 model.

i7-4850HQ @ 2.3ghz
16GB of 1600mhz DDR3 RAM
Nvidia Geforce GT 750M (2GB GDDR5 Version)
512 GB SSD
Windows 8.1

I have spent months now trying to figure out why Battlefield 4 would run fine on startup and then performance would degrade rather quickly. I think it happens with other games too, but it isn't as noticeable because those games aren't pushing my machine to its' limits like BF4 does. I finally realized that my GPU was downclocking itself back and forth by a difference of about 200mhz. I have tried everything I could possibly think of to get this to stop. I set my CPU min/max through Windows as low as 80%. Even if the GPU goes back to max clock speed, it will eventually start underclocking again. I tried underclocking the GPU's memory by up to 502mhz. Same deal, even if the core clock goes back to max, eventually the downclocking starts again. This process does not seem to be tied to the GPU temp at all. Sometimes the GPU gets up to 78C and doesn't throttle, other times it will be at 73C and throttle. I also tried setting the multiplier as low as 17 in TS, by increments of one. Each time I went down by one, the throttling would go away for about 15-30 seconds and then come back.

Naturally, I assumed that PROCHOT and/or BD PROCHOT were the problem. After running Unigine Heaven for over 2 hours while watching the monitor in NvidiaInspector and Throttlestop, the check boxes for TDP Throttling, PROCHOT, and BD PROCHOT never became checked, no matter what my CPU clock speed, multiplier, or temp was. The thing about this that sucks is that before the throttling kicks in, BF4 runs great. 1600x900 @ 60FPS, low settings. I played the game on the Xbox 360 for about a year before this so this is a major graphical upgrade.

Now I'm thinking that my CPU might be hogging too much power and something in the Bootcamp drivers is overriding my settings that could lower the TDP of the processor. I'm probably completely wrong.

If there was ever a person that might be able to figure this out, it's you. I do realize that I might just be screwed because I'm using a Mac for Windows Bootcamp and have no real access to anything resembling a BIOS or EFI. If you find yourself in a helping mood, I will send you whatever logs or information you could possibly need. Thanks regardless.

EDIT: I have been running Unigine Heaven for almost an hour now with a clock multiplier of 17 (1695.08mhz) and my neither my core clock nor my voltage have dropped even once. Perhaps when I tried it earlier there was already too much heat built up in the system from it being set to a higher clock multiplier just before.

 

No need to disable C-states or mess with Windows min and max processor settings if MSR 0x64C is set for one bin below the LFM and importantly as mentioned before, MSR 0x1B0 is set for max performance.

I have seen a few complaints from desktop users that voltage increases by 0.1V but never seen this on my mobile HSW. Attached is a TS log , 3 runs at 182GFLOPS from which shows a maximum change of 0.015V. I've set a flat turbo of 34 so the 35x and 36x bin voltages don't confuse things. Note that voltage changes tend to be around 5mV, in line with VID stepping? FWIW my BIOS comes without any BCLK or voltage settings except for iGD which is incorrect anyway so have had to modify the BIOS myself for those to get that BCLK on the right side of 100MHz (99.8 just niggles me) and set both core and cache interpolation voltage at BIOS time. Don't know if this makes a difference.

If you are going to use static then I would suggest using flat bins, ie 35,35,35,35 otherwise you will need more voltage than necessary for 4 core load in order to support higher bins of 1 and 2 cores. If you wish I could modify one of my programs to narrow down what is causing your throttling. Let me know.


@ tribaljet, easiest way to sort out threads for AVX is to disable HT in the BIOS. With HT some registers, such as GPR's are replicated. AVX is not replicated and has to be shared between the 2 threads of each core which results in stalls and cache being polluted and so results in degradation of performance. Once you start operating the CPU outside Intel specification all bets are off. How much you push your processor past specification is up to you at your own risk. Too much voltage will kill a CPU instantly. Electromigration tends to be slower and may appear as degradation of overclock over time unless it snowballs into oblivion and kills the CPU outright. Electromigration is an effect of too much current and compounded by high temperatures.

 

Sure, go right ahead. I don't mind learning more haha. Maybe one day I'll be as good with these things as Mr. Fox haha

 

Dufus - Thanks for the info and log.

Your VID seems very consistent whether you are running AVX instructions or not. Some Haswell desktop users seem to have a larger difference and D2 Ultima's 4800MQ also seems to get a significant VID boost when LinX gets to work. My 4700MQ seems to be similar to D2 Ultima's 4800MQ but my CPU hits the power limits so quickly that it is hard to do some proper testing. Perhaps this difference in VID when running AVX instructions varies depending on the CPU or on the motherboard or who knows.

I thought a useful addition to ThrottleStop would be some sort of adjustable max VID function for the Haswell CPUs. If you had a system that kicked up the VID, ThrottleStop could switch over to a user adjustable fixed voltage on the fly to help prevent the VID from getting too high. Maybe a feature like that would win me a few fans from the desktop community.

Maybe someday D2 Ultima will be a household name just like Mr. Fox is for providing me with some useful data so I could see the problem his CPU has. :thumbsup:

ColdAtrophy - I am not too familiar with Macbooks or BootCamp but I am always interested in learning something new. The best thing you can do is post a screenshot of ThrottleStop that shows how you have it setup when you are trying to play BF4. For ThrottleStop to work correctly, make sure you are using the Windows High performance profile with the Minimum and Maximum processor states set to 100%. Turn on Nivida GPU in the Options window and of course, send me a log file so I can have a look at some numbers. I always seem to find some interesting secret hiding in there. Make sure BD PROCHOT is not checked when you are testing.

Sometimes the GPU will throttle because the CPU has already started throttling. You can also include screenshots of the TRL and TPL windows.

 

Know what'd be a nice addition? A "minimize to system tray" option. I don't mind it on my desktop, but I prefer the pre-win 7 taskbar look (where nothing is bunched together) so I'd like if I could leave TS in the system tray and just double-click it if i need it. You already have something useful there where it lets you right click the system tray icon to turn on/off and switch profiles, etc.

Also, I've done limited testing using the TimePeriodAC=1 line. I believe it's fixed my random microstutter-feel in The Binding of Isaac Rebirth, which I couldn't figure out the reason for. So that's fantastic. I've been keeping TS running 24/7 since then. I turn it on as I'm not sure if monitoring mode adjusts the windows timer resolution, but the better my PC works, the happier I am.

What'd also be a nice addition is the ability to force the high multipliers for turbo, like Dufus' executable does for me. When I restart my PC the forcing goes away, so it's not an issue, and if it were integrated into keeping throttlestop "on" and tied to a setting, then I could simply say... set it to switch profiles on battery, or something, and let the CPU properly downclock/sit in a low power state. I believe Meaker managed to use a feature of throttlestop to get about 2 hours of battery life out of his P570WM with two 980Ms and a 4930K. I'd love to squeeze a solid 3+ hours out of this machine with low power usage when I'm taking notes or typing (I just haven't figured out how to do it yet).

 

Okay, here you go, modified for use with WinRing0. It's a bit rough and lacks a lot of SW error checking but should run okay on your HSW.

Same as before, extract file LR.exe into TS directory.

When running red means active, yellow means logged. If there are some events already showing just click on each respective button to clear before running Linx. Good luck.

 

Could something like Cinebench be used, given that are less variables needed to be taken into account?

About power limits, I honestly don't feel comfortable enough to bump both to 150, perhaps something a bit more conservative yet above my 110 Short/88 Long settings, like matched values.

Overall, I believe my current full load temps are acceptable, being below 70C. Also, I'm only running 100MHz higher than stock as I'm not messing with voltages at all, I just want to push the CPU as high as it can at stock voltages, and that's 4.5GHz as 4-core 4.6GHz BSODs on me, even though 2-core 4.6GHz doesn't, but I do want to run same clocks on all cores, hence dropping to 4.5.

 

Doesn't ThrottleStop already have this option? Go into the Options and make sure Task Bar is not selected. When you minimize ThrottleStop, it should go to the system tray. When you double click on the icon in the system tray, ThrottleStop should pop open. There is also a Start Minimized option so if you check that, ThrottleStop will start up and be sitting minimized in the system tray.

In the ThrottleStop 7 download is a link to this website.

How to Clean the Notification Area Icon Cache in Windows 7 & Windows 8

If you are having any weird problems with the system tray icons, run the Notification_Area_Cleaner.bat program that is available from that site. When it comes to system tray icons, I had no idea what I was doing when I wrote ThrottleStop 6. Windows seems to get confused if you switch back and forth between ThrottleStop 6 and the new and improved ThrottleStop 7. Cleaning out the icon cache seems to fix this issue. After you do that, the system tray icon should work pretty much the same in XP, Vista, Windows 7, 8, 8.1 or 10.

The TimePeriodAC=1 INI option has totally fixed the micro stuttering I was having on my older desktop computer. That problem drove me nuts. I always thought the problem was that the video card was not switching to the 3D clocks when playing a simple game on the desktop. I was very surprised that adjusting the timer resolution has completely solved the stuttering issue. I will definitely add this to the ThrottleStop - Options window in the next release so it is easy for users to find and change. This feature presently works even if ThrottleStop is only in Monitoring mode.

Some of the magic that Dufus does is beyond my level. ThrottleStop has a feature in the Options window called Run Program Before or After Profile Change. Maybe you could add his magic program to ThrottleStop using this feature. Where it says Before - After, select After. Click on the Profile 1 button at the bottom of the Options window and tell ThrottleStop where on your hard drive you have located the magic LR.exe program. His program will start when ThrottleStop starts using this method. The Before option will let ThrottleStop start up without running a program. If you set your computer up this way then all you would need to do is open up ThrottleStop and click on the 1 at the top left of ThrottleStop where you switch profiles. You could also create a keyboard shortcut to switch to Profile 1. Even if you are already in Profile 1, switching to Profile 1 from Profile 1 will cause LR.exe to run. If you swap profiles back and forth then maybe you could ask Dufus to create a quiet version of LR.exe where nothing pops up.

Edit - I could also create an option which lets a user run any .exe once when ThrottleStop starts up. This would avoid the problem of the same program being run every time you changed to Profile 1.

I know there are a few people out there that think pulling 75 watts out of a 47 watt CPU in a laptop is a good idea but that kind of fun can scare people away from using ThrottleStop. I prefer if enthusiasts have to work a little to accomplish their goals.

Dufus - Can you explain what LR.exe does? Maybe send me a PM if it is a secret. No need to get too complicated. Are you just fixing up the memory mapped power locations or is there some other magic going on too?

 

http://i.imgur.com/uEeElP1.jpg

This is the most successful setup I have found thus far for playing BF4. I read many pages of this thread before posting, so I did already have Power Options set to High Performance. I also checked Nvidia GPU and unchecked BD PROCHOT. If I leave High Performance on without Throttestop, the GPU throttles within about 5 minutes, which I believe is a result of the CPU being locked into its highest frequency - 3.4ghz.

I can try different settings to show you what happens when the GPU throttling occurs. It might seem that this setting for Throttlestop fixes the problem, but I noticed while playing with that multiplier of 17 that my frames dip more often and I cannot maintain V-sync. My GPU usage never goes over 98% and it dips much more often than usual. I'm thinking that I'm having to throttle the CPU too much to keep the GPU fed.

Today: http://i.imgur.com/MznjmsF.jpg

After running Unigine for about 30 minutes, the throttling did not kick in, but PROCHOT 100 did about 5 or 6 times. Each time I unchecked the box and no throttling occured. I then started to play Battlefield.
Fifteen minutes in or so, I started to get some hardware lag spikes. I checked Throttlestop and PROCHOT 100 was checked. I unchecked it, but this time it made no difference. PROCHOT 100 did not recheck itself, but the throttling had begun and once it kicks in, there doesn't seem to be much I can do to stop it.

You will see that I am running a GPU clock offset of +135mhz, bumping my GPU core up from 926mhz to 1061mhz. The throttling will kick in whether I have leave it at stock or not. There doesn't even seem to be a time difference in how long it takes for the throttling to kick in. I also tried over and underclocking my VRAM by as much as +250mhz and as low as -502mhz. I really thought underclocking it would help to reduce temps and voltage, but that doesn't seem to be the case.

It should be noted that my laptop sat idle all night and was quite cool when I started this. It took FAR longer to get the throttling to kick in today (15 min last night, 1 hour now) but I am home alone and the ambient temperature in the house is lower than it was last night while testing (74F last night, 71F now). I have seen videos on Youtube of people who had idle temps around 30-40C, particularly after repasting the thermal compund on their CPUs and GPUs. I have never seen my temps go below 45C, even with my fans at max speed. Perhaps I need a repaste and laptop cooler?

I'm not sure that you want me to copy + paste my Throttlestop log file here as that would take up quite a bit of room. Please let me know how to get it to you as I don't see an option to send you a direct message. Should I use a file hosting site?

If in your readings or research you come across the tip that resetting the SMC (System Management Controller) and PRAM (Parameter RAM) on a Mac could help the GPU throttling issue, know that I tried this already. It does nothing.

It should also be noted that whenever I am trying to play a game, my laptop is plugged in and I use MacsFanControl for Windows to set both cooling fans to max speed.

TRL: http://i.imgur.com/YDMkP9p.jpg
TPL: http://i.imgur.com/OHvp6Th.jpg

Your program is the ONLY thing that has helped thus far in even the slightest way. I am convinced that the answer to this problem lies within the software you have created, but I just don't know enough to use it correctly. I am very grateful for your help. Thank you.

 

Hello everybody,

Firstly regardfull thanks everybody especially unclewebb with this software helped and saved my laptop from weird gpu throttle. I have Asus N550JV laptop and i have. I have stopped Cpu throttle with ThrottleStop but could not stop Gpu Throttle although my temperature does not go above 6070 celsius. Here is the link of the topic i have started :

http://forum.notebookreview.com/gam...67124-nvidia-gt-750m-throttle-80-celsius.html

And here is the last monitoring of my laptop in an average heavy game:



And here is the log file of ThrottleStop:

TinyUpload.com - best file hosting solution, with no limits, totaly free

At the start of the recording log file i have ticked BD Prothocol and after few mimutes i have unticked it. After the unticking BD prothocol my Cpu Throttle has stopped. I have selected different multipiers for cpu but whatever i do and whatever the temperature is my Gpu always throttle.

I hope your help and many many thanks.

 

Oh that system tray is wonderfulllll <3. I'll do some more testing on the LR program later tonight with Linpack. Right now I'm gonna be doing some testing with Open Broadcaster Software using SLI compatibility game capture.

 

ifmyn - Your CPU is still throttling. A 4700HQ should not be spending most of its time using the 28 multiplier, the 24 multiplier or the 8 multiplier. A 4700HQ should be running faster than that.

I need to see a picture of how you have ThrottleStop setup. If BD PROCHOT is checked and that causes your CPU to throttle and use the 8 multiplier then you know that is a problem so fix it. Clear the BD PROCHOT box and leave it cleared when testing.

Take a screenshot that shows how you have ThrottleStop setup, check the Log File option and then do not change any settings in ThrottleStop after that. When you are finished gaming, exit ThrottleStop and send me the log file information and the screenshot.

ColdAtrophy

The PROCHOT 100 box can not be used to control thermal throttling, it simply records when your CPU is thermal throttling. Your statement does not make any sense. If the PROCHOT 100 box keeps showing a check mark then that shows you that your CPU is thermal throttling. If you use ThrottleStop to clear that box and a minute later that box is checked again, that confirms that your CPU did some more thermal throttling sometime during the last minute.

Was that the Unigine - Heaven benchmark that you were using? That benchmark puts a very light load on the CPU. If your CPU is getting up to 100°C during that benchmark then that means there is a problem with your cooling solution. If I grabbed 5 brand new Macbooks off the assembly line and they all reached 100°C during this test then I would conclude that the heatsink and fan are completely inadequate. You could try pulling your laptop apart and replacing the thermal paste and using a cooling pad and any other tricks you can think of but if the problem is bad design, this will probably not solve the problem.

If you are having over heating and throttling problems then first of all, do not overclock your GPU. Leave it at its default clocks. Overclocking consumes more power and creates more heat so you need to avoid doing that. It is possible that your power adapter is not powerful enough to run your CPU and GPU at their full rated speeds at the same time. Do you know how many watts your power adapter is rated at? Google shows some 60 and 85 watt adapters for the Macbooks. For comparison, a Lenovo Y510P with a single GPU comes with a 120 watt adapter. These are just the theoretical power ratings. Some companies like Dell have been known to throttle their laptops before the power adapter reaches its full rated output so as to avoid the power adapter from tripping itself. After this happens, the power adapter needs to be unplugged and then plugged back in to reset itself so manufacturers like to avoid that happening to consumers. Maybe one of the reasons your GPU is throttling is because it is starving for some power.

There is really only so much you can do with software like ThrottleStop if the real problem is at the hardware level. I would look at investing $10 or $15 bucks in a Kill-a-Watt meter or similar so you can keep an eye on power consumption while your are using your laptop. You might see that your GPU throttling is directly related to power consumption or that your power adapter is fully maxed out when gaming and is tripping some safety feature.

If the cooling solution in your laptop is not adequate to cool down a 47 watt CPU then you might want to try using ThrottleStop to reduce your CPU to 25 watts to try and control power consumption and heat. The goal is to avoid whatever is causing the GPU to start throttling. Go into the TPL window and in the Primary Plane Power Limits section, check off PP0 Power Limit and set that to 25. On that same line check off the Clamp option. At the bottom of that screen, try checking the Intel Power Balance option and set that to CPU 31, GPU 0. Push OK, go back to the main screen and maximize the Set Multiplier value and make sure that is checked. This will allow your CPU to throttle based on power consumption instead of throttling based on speed. You might also have to combine this with reducing your GPU speed below its factory rated value.

You have an incredibly powerful laptop but it is being limited by bad design. You are going to have to lower the CPU and GPU speeds to lower power consumption and avoid over heating. That's sad. Kind of like telling the Kentucky Derby champ to go pull an apple cart. No pun intended.

 

Dear unclewebb,

Here you can find the screenshots of my ThrottleStop settings and my log file. I have created them like what you said. Gpu throttled at this process like the SS above my post. Many many thanks.

Log file: TinyUpload.com - best file hosting solution, with no limits, totaly free

 

ifmyn - Why are you throttling your CPU? Why is the Set Multiplier value set to 28 T? That should be set as high as possible which is 34 T for your CPU if you are not overclocking it. Your CPU is not too hot so there is no reason to throttle it.

Your GPU temperature gradually goes up to 79C. If the GPU starts throttling, you will usually see an immediate drop in the GPU core temperature.Can you try the GPU-Z program and run a log file with that while you are logging with ThrottleStop. Play a game for a good 15 minutes or longer with ThrottleStop set to 34 T so I can see how hot your CPU and GPU get and if there is any throttling recorded. Do not ALT + TAB out to your desktop while you are playing. Just play the game and if it throttles, let it throttle and just keep playing. If it throttles and never stops throttling, let it throttle. I just want to see as much data as possible. Sometimes a laptop will throttle and then it will go again and run normal and then later it will go back to throttling over and over again.

Try downloading ThrottleStop 7. It has a few more features for the 4th Gen CPUs like voltage control which might be help improve your laptop's performance.

 

Dear unclewebb,

Many thanks, i will do what you need as soon as possible. I was using Cpu multipier 28 T because higher multipier like 34 T throttles my Gpu much frequently. If i use notebook fan control software and set it to %100 fan speed; my cpu + gpu temperature never goes above 70-75 celsius.But if i dont use notebook fan control software; default BIOS fan control let my gpu + cpu to 80 celsius but not more. But the most important is Gpu throttle whatever the temperature is, gpu throttle is not temperature dependent, i think its "heavy load" dependent. So may you kindly tell me should i test your needs with or without notebook fan control software set to %100 fan speed?
View attachment 118864

 

Nothing special, it just monitors the registers for limiting reasons, well the one I posted has been modified to use MSR's instead for driver compatibility with WinRing0. Nothing secret here, it uses MSR's 0x690, 0x6B0 and 0x6B1 which can be found in the Intel SDM starting from page 3271. It breaks down throttling to show if it was caused by PL1, PL2, PP0, PP1, PL3/current, Thermal, BD PROCHOT, VR Thermal... Thought it might help D2 Ultima with identifying his throttling problem.

 

Okay, Dufus, I did a check. When at the 3.3GHz portion of the test, "EDP" was the limiting factor (I assume; it lit up in red) for both "core" and "ring".

 

That's either PL3 or current then. Easy enough to check the current by increasing from 100A to 110A and see if your FLOPS go up.

 

okay. I'll set the current limit to 120A or so... you said that works for you, right?

 

110A will be enough to see if it makes a difference

Once you're happy you should try to work on reducing your voltage as much as possible to use the least amount of power for maximum performance and stability.

 

I had found that using -80mV had it downclock a bit, but -50mV was more stable, so I'll bump it to 110A and keep -50mV for stock. When I get CLU I'll gun for 120A/3.9GHz/as little voltage as I can use.

I tried -50mV on this at 3.9GHz before and it instantly BSOD'd haha. Barely loaded into windows before that happened. Might also relax power limits to 120W just in case anything gets draw-happy (as 3.8GHz with that current limit pulled a solid 94W on the package). If I can handle a solid 3.9GHz at even -20mV I'd be beyond happy.

 

Okay, I set the current limit to 120A and it still throttled with "EDP" being the reason. I think it might at this point have to do with the voltage for the type of load it's pulling. When I left it at stock voltage (which is probably WHY intel left that as stock voltage) it throttled less, somewhere around 3.4 or 3.5GHz, so that may be it. Oh well. When I get the CLU and begin OCing I'll hunt more into that. Next time I restart my PC though I'll make sure and set it back to normal and see if it throttles.

 

Just make sure you have set windows power plane to High Performance - maybe its "blind" shot but if it's set to Balanced or Power Saver then EDP is flashing RED here.

 

It's custom set, I have no problems except in this linpack program. Here's a screenshot of what it looks like mid-run (minus the program that's showing me EDP is the cause)

 

That's strange, I wonder if it's just coincidence that the multi you drop to is the top 4 core default without unlocked bins. I have noticed that when changing power limits or voltages on the fly it can sometimes cause a glitch like a reset to multiplier operation. Be interesting to know what is causing it. BTW, you can move LR around if you need to with left click, hold and drag.

 

I have, that's just an old screenshot. I didn't take any while it was running is all, and I didn't feel like closing everything to take a run to take a single screenshot XD.

Also, I think I'm gonna try stock voltage next time I restart the PC, just to see if it'll let me keep the 3.5. If it does, then linpack is somehow flawed in itself. Because the voltage it draws is much less than the voltage I can get my CPU to use while streaming, which means the CPU isn't voltage-starved, or shouldn't be. Either that or Haswell is just bloody retarded.

 

ifmyn - Your ThrottleStop log shows that with BD PROCHOT disabled, your CPU is running fine. ThrottleStop can only help with CPU throttling. It cannot control or fix your GPU problems.

You can use ThrottleStop to reduce the CPU multiplier like you were doing. This will reduce CPU power consumption as well as heat. You will probably have to do the same thing with your GPU. Reduce the GPU speed and voltage if possible to reduce power consumption. That is about all you can do. You will have to run your laptop slower to try and avoid the GPU throttling.

 

Thank you very much for your kind and helpfull effords to fix my problem. It seems i can do nearly nothing with this problem is about Gpu.. If there wasnt such a wonderfull software Throttle Stop, i couldnt play any game on that engineering fault laptop. I know this Asus N550 JV is not a gaming laptop but it cant give the performance it has to give with its price and hardware. If someone read this post and thinking to buy Asus N550JV please never give any money to this engineering shame.

Best regards.

 

This is my favourite quote of this entire forum

Hey Dufus, I tried to get my friend to use the hi multis program you send me, but it returns this screenshot error: Gyazo - 878b6e55f7ae68b0727474cb3522adc6.png

He's using Sandy Bridge-E (the i7-3820 to be specific). Is there a version you could do for him?

 

I think the HiMultis program is only for the 4th Gen CPUs.

ifmyn - Have you tried using Nvidia Inspector? On the right side you can overclock the main P0 performance level but you might also be able to change some of the other performance levels. When your GPU throttles, maybe it will throttle down to a performance level that you can control with Nvidia Inspector.

 

@D2 Ultima, I only provided that as a means of testing your HSW. To make it more universal would require much more error checking, for instance checking CPU support. IIRC DT IVB has these registers but may not work the same way as HSW or need some extra settings.

IMO better you request Unclewebb for extended features, if enough people ask it might persuade him. As it is Unclewebb has spent a huge amount of time on TS and offered support when asked but sadly there only seems to be a very small percentage of users such as yourself, who are willing to give up a few minutes to offer suggestions or give some detailed feedback. Ironically I think if it were not free he would probably get a lot more response.

I understand your frustration with HSW but I think I have already shown that it can operate without throttling providing power delivery and cooling are sufficient AND the laptop manufacturers are not imposing limits. Maybe you can ask Prema if there are any hard set throttling mechanisms with the Clevo you have.

I did try upping the bclk to run at 3.5GHz with the 4700MQ and although I had to increase voltage for stability it did run at 187GFLOPs with just a little thermal TCC towards the end of the third run. No EDP or other throttling, running in Windows default balanced power plan. Be careful not to degrade your CPU running at twice the SKU power, Linpack is a hard test of a CPU and when operating the CPU overclocked / out of spec, should IMO be used sparingly.

 

Yes master, i have tried various Gpu utilities including MSI afterburner and Nvidia Inspector but unfortunately both overclocking and downclocking dont stop Gpu throttling. It seems its a locked BIOS feature behaviour that anyone cant answer why exist when temperatures are 70 celsius. The only thing i havent tried is to downgrade my System BIOS version to previous ones. Gpu BIOS is locked and cant even save to edit. The overall problem seems an unknown "TDP limit".

Best Regards.

 

You might be interested in this. I just ran this test to show someone that not all HSW increase VID by 100mV when running AVX2.

Surprisingly the VID doesn't increase at all from cool idle but actually drops a bit when load is reduced, possibly due to that temperature phenomena mentioned earlier.

 

I am not sure why VID boost seems to happen to some Haswell CPUs but not others. I thought this might be more exaggerated on the desktop CPUs since the original Haswell desktop K CPUs used a much higher turbo multi when overclocked, compared to the default multi. My 4700MQ seems to add about 35 mV when it starts running AVX2 instructions but it is impossible to properly test this with my locked bios and poor laptop cooling.

Perhaps this could become a method to find golden Haswell CPUs. Perhaps the golden chips do not need extra voltage to remain stable while running AVX2 instructions while Intel programs the crappy chips to add a little extra VID to insure stability. Buy me a box of Haswell CPUs and I will start binning them.

Thanks Dufus for all of your testing and helpful apps.

 

Well all I can say is that I see a definite lack of my CPU wanting to hit the max voltage if left on adaptive for linpack, but it will do it in other instances. I have two examples:

First, my normal VID for linpack sits at 1.031v for linpack (spike to 1.036v sometimes), and this is with the lower speed. However when I try streaming CoD: Black Ops 2 (1080p, maxed, 120fps) using x264 at 540p, 1600 bitrate and "slow" compression (what my internet allows), it will top out at 1.041v easily when OBS (the streaming software) starts to use large amounts of CPU power (sometimes it tops out around 56% of my CPU with those settings). It feels like my CPU has more voltage to give but linpack doesn't want it.

Also, the same thing happened when I'd OC'd it to 3.8GHz (using stock voltage). In the screenshot I have of my OC'd chip it drew 1.109v (with core 3 spiking to 1.132v for some reason; may have happened post-benchmark) but when I tried the same streaming setup with Dark Souls 2 this time (streaming at 616p, 42fps, slow compression, high encoding profile, etc) my CPU would max at 1.192v on all 4 cores during the stream.

THE WHOLE VOLTAGE THING MAKES NO SENSE. *slays intel*

 

I have found that with a too low voltage (then increases volts more than the standard during xtu bench). By increasing voltage up slightlyas (standard) and running the same test bench then becomes volts present as the standard. That not increases Volts on processor at all(chips do not need extra voltage).

 

What if you run at a lower multi?

Kind of reminds me of an older problem on SNB

D2 Ultima if you use fixed voltage does your VID drop? If not using fixed voltage I would recommend trying to run your 4800 with flat bins at 3.7GHz using MSR 0x1AD to find your lowest stable voltage then add 30mV per multi above that to interpolation voltage. You can then try lowering the interpolation voltage.

 

I seriously have no idea what you just said. No joke. I don't mind trying fixed voltages someday, but I've no idea what MSR 0x1AD means or what the interpolation voltage is. XD. I have a lot to learn about this. I might have to start to use XTU for more options here. I hope it doesn't mess with my restarts like it did before.

 

Is it possible to use a "startup profile"? When I launch throttlestop it'll use the default profile instead of my battery saving one or performance (it's something I can test to forget).

Perhaps there's command line functionality to switch profiles?

 

Wouldn't be surprised if it did


Sorry for the confusion. You could basically set all bins to 37 in TS then adjust a negative offset as much as possible while testing stability. You might want to give yourself a margin of 20-30mV once you've found your maximum stable offset.

As an example say you end up with -60mV (0.060V) as your offset and at 37x multi VID reads 1.050V then to run 39,39,40,41 which is up to 4 bins higher add 4 x 30mV to (VID - Offset). So you would get 120mV + 1.050V - -0.060V = 1.230V.

You would then use the 1.230V for your Core Voltage and this will give a VID of 1.170V for 41x bin, 1.080V for 38x. Slowly reduce the 1.230V Core voltage as much as you can while testing stability. Again, you might want to add a margin of 20-30mV to that voltage.

You might also want to adjust cache voltage at the same time, if not then keep the cache ratio limited to 37x or below.

Hope that's a bit easier to understand. Remember to check lower multi's with a hot core as VID's can drop with temperature increase, i.e. run a heavy load to heat the cores then switch to say 8x bin and test.