The ThrottleStop Guide

How about less fan noise?
I use TS to force a lesser multiplier so I can have it crunch movies overnight while I sleep like a baby.

 

Use rather ear plugs when you sleep

 

I'm not sure if you're serious or trolling...

Why would I want the discomfort of earplugs if I can simply turn the noise down?
You seem rather resistant to the idea that there's a use for turning down the max speed of a CPU. A very limited train of thought if I may say.

 

I just kidding. sorry. You can create up to 4 profiles inside Trottlestop for your processor that you can switch between on the fly With macro.

 

tijgert - You can use ThrottleStop to run your CPU at whatever speed you like. You can also use the ThrottleStop keyboard shortcut feature so you can setup two different ThrottleStop profiles and then easily switch between two different speeds by using the keyboard. When it is bedtime, it will be easy to switch profiles and slow your computer down for the night shift.

papusan only seems to know one speed and that is as fast as possible. He sent me his picture recently and quite frankly, I think he is from another planet.



The poor guy doesn't appear to have any ears. It makes sense that the speed of his computer is more important to him than how it sounds.

 

Unclewebb you hit the nail on the head. Most people need only one profile in Trottlestop. Only one profile set to Max multiplier or all profiles set on max.

 

Sorry for the old quote, but could you explain what you mean by Throttlestop 7 beta expires at the end of december? Does this mean if I have it installed it would stop working at that time, or would it just be unavailable for download? I've not used it a lot myself, but my friends with other haswell machines using premiere pro etc have used it to keep their CPU at their proper clocks (as it doesn't like to do that normally apparently). I'm going to start using it myself to see if it does anything about the once-every-few-hours 1/4 second stutter I get in some games, since you said it helps with Win 8 and the CPU; so I kind of don't want it to vanish from existence anytime soon XD.

 

papusan - I have always been a one profile kind of guy too. Even surfing the net, I like to get 'er done as fast as possible.

D2 Ultima - When the beta expires, ThrottleStop 7 will not run anymore. I usually try to make sure that I have a new version ready to go but sometimes I wait a few days to see if anyone notices. It is the easiest way for me to get some feedback. If lots of people complain then that usually means they were happy with the previous version so I should release it. I plan to release the next version without an expiry date. In case of emergency, you can always change your computer date back a couple of days. My expiry code is pretty simple so if I ever get run over by a bus, it should be easy enough for someone, somewhere to release a cracked version with no expiry date. No worries.

This INI option

TimePeriodAC=1

seems to have fixed the micro stuttering I was having on an older Core 2 Quad desktop. No guarantees that it will fix anything on your computer but it is worth a try. Also make sure you are using the Windows High Performance power profile.

Random stutters can be something as simple as a bad driver. The fun part is trying to find out which one and if there is an update or a previous driver available that doesn't have the stutter problem. I recently found that the Elantech trackpad driver excessively loads the first thread of the CPU. Elantech supplies track pads to many different laptop manufacturers. Simply moving my finger around on the track pad causes a load of 30% on the first thread. That's ridiculous and shows that the Elantech driver is sampling the trackpad position WAY too often. It should not take a billion computer cycles per second to figure out if someone moved the mouse or not. That is poor programming. Things like this can interfere with other apps or drivers that are also trying to use the first thread of the CPU. There are lots of issues like this happening in the background of many computers. Faster processors can cover up these problems, most of the time that is.

 

Well I don't have a ton of stuttering or anything like that. But I will indeed try out that ini mod and see.

Also, I custom set my power plans always. Battery has limits, plugged in does not, etc. But thanks for the advice. The stutters I never thought anything much about because if in SP games, I would maybe experience one every few hours. To be honest I think it's google chrome, but I wanted to see if it'd be helped by just keeping throttlestop on XD.

Hopefully you'll get the demand you want for throttlestop 7 =D. As for drivers, I have no idea. Maybe it's my GPU driver and the next time I update it it'll go away XD.

 

Hi Unclewebb,

I've noticed that you have the same laptop that I have from some of your posts (Y510p). In my opinion, the cooling system on this laptop is quite weak, thus the processor is quite gimped.

From your experience, does using ThrottleStop on this laptop, and enabling the default maximum turbo boost of 3.2Ghz across all 4 cores result in TOO MUCH heat to dissipate? For how long can the maximum turbo boost be achieved, and do you have any rough temperatures?

Thank you for your time and help.

 

unclewebb sure is coherent and articulate for a sleeptypist...

 

Okay, next question. For the core series CPUs like I'm on, should I add the
TimePeriodAC=1
line alone into the ini? Or should I add the following
TimePeriodAC=5
TimePeriodDC=16
lines? Or should I add both and set AC to 1 while leaving DC on 16? I just want to be perfectly clear XD.

 

By increasing system timer resolution, (micro) stuttering can indeed disappear but it should be worth mentioning that timer resolution alongside HPET toggling has been reported to cause odd issues on some configurations (like X58 chipset systems). However, from personal experience with both HM65 and Z97 chipset systems (the latter automatically enabled HPET), I've only gained perceived smoothness and no issues whatsoever.

Regarding profiles, when I'm on my desktop, I do tend to use a single profile configured the way I want it, while on the laptop I actively use 3 profiles, one for maximum performance when rendering, another for gaming where I disable Turbo Boost in order to have much lower temps and a third profile that's limited to minimum multiplier when I just want the system to idle but be online to handle mails and FTP usage.

 

I totally agree. Calling the cooling system in the Y510P barely adequate is being generous. That is likely the #1 reason why Lenovo decided to disable Intel Turbo Boost. This quietly happens in the background while gaming and most users are not knowledgeable enough to understand that they are being ripped off. It is dishonest for Lenovo to sell a laptop, provide inadequate cooling and then quietly disable Turbo Boost while gaming without providing a user any indication of that. Lenovo never mentioned this "feature" when I bought my Y510P.

It really depends on what type of application you are trying to run. If you use ThrottleStop and enable Intel Turbo Boost and an application uses the newer AVX instructions, you will probably end up with temperatures going through the roof and your CPU will easily reach the thermal throttling temperature. Lenovo set this temperature to 99°C for the 4700MQ in the Y510P.

On the other hand, if you run an application that does not use the AVX instructions, it is possible that you will be able to run the CPU at full turbo boost indefinitely. Whether the CPU reaches the thermal throttling temperature or not will depend on your room temperature, how good a job Lenovo did installing your heatsink and how clean it is.

The built in TS Bench does not use the AVX instructions. While running it I was able to overclock the CPU so it uses the 34 multiplier with all 4 cores fully loaded. I adjusted the CPU core voltage a little lower and I ended up with temperatures right on the edge of throttling. It was able to run full turbo boost +2 bins of overclocking which is a feature that Intel builds into all of their 4700MQ mobile CPUs.

http://i.imgur.com/uy6Byna.png

The TS Bench only stresses the CPU. If you are gaming then there is going to be extra heat from the GPU going into your cooling system. Most games do not fully stress the CPU but it ends up being a real balancing act trying to maximize performance without the CPU hitting the thermal throttling temperature. Your results will really depend on what app or game you are trying to run and what room temperature you are at. Living in Canada does have some advantages during the winter.

D2 Ultima - If you start up ThrottleStop and it doesn't find any values in the INI file, it will use these 2 values as the default values.

TimePeriodAC=5
TimePeriodDC=16

These values change the timer resolution that Windows uses so it doesn't matter if you have a Core 2 or a Core i or 2 cores or 4 cores or hyper threading or whatever. This setting controls Windows.

I read some documentation recently by a guy that really knows his stuff when it comes to Windows and setting this to 10 seemed to be the magic number. A ThrottleStop INI setting of 16 usually results in the timer resolution being set internally to 15.6 milliseconds. That seems to be the default and I am not sure why Microsoft decided that number would be a good thing. The operating system or any user software running on your computer can change this value. If you use ThrottleStop and it sets this to 5 and then some other software running on your computer decides that it wants to set this lower to say 1 then 1 is going to win. As soon as you stop running that other program then in theory it should go back up to 5 because that is what ThrottleStop wanted. If you exit ThrottleStop then it should go back up to the original 15.6. Not all software follows the rules so it is possible that some software changes this value and then doesn't change it back after you exit.

The problem on the internet is that people try to do some testing of this without fully understanding how it works. At one time, Google Chrome used to change this value but there were complaints so I think Chrome leaves this alone now. If your computer does not have any stuttering problems that are being caused by this setting then whether you set it to 1, 2, 4, 5, 10 or 16 is not going to make any noticeable difference. When testing I noticed slightly more deep C State (C6/C7) residency time when this is set to 16 compared to when this is set to 1. Intel or Microsoft would have you believe that it makes a huge difference to C State residency time and a setting of 1 should be avoided but it doesn't make that big of a difference at all.

Try setting it to whatever you like. Numbers that divide equally into 1000 seem to be preferred. (1, 2, 4, 5, 10)

 

Here's when TS7b3 is running default.


And after it exits.


W8.1 HPET, highest resolution 0.5ms and granularity around ~0.0013ms

There is a register for forcing wake but might need an undocumented bit set so that the driver does not reset it. Let me know if you want to try it.

 

Okay, last question then: do I set BOTH lines to the number I want (being the same number)? Because I don't see either line in the throttlestop configuration ini, which means I'll probably need to add them myself, unless I'm using the wrong ini or something. (I'm just trying to make sure I understand everything, and you're definitely helping, so I hope you can bear with me XD)

Edit: I had this page up all night and replied after I woke, I think I get it more now that I see Dufus' post XD

 

Quote Originally Posted by F4rewell View Post
So if i understnd correctly my intel 4710hq would throttle itself down if temperatures are way too high regardless having throttlestop turned on?

Unclewebb:

Correct. ThrottleStop cannot be used to disable the thermal throttling that Intel builds into their CPUs. ThrottleStop only lets you control the throttling schemes that laptop manufacturers like Lenovo continue to use. No matter how you have ThrottleStop setup, your CPU will throttle if any core reaches the thermal throttling temperature. For the 4710HQ, Intel has set the throttling temperature to 100°C.

ARK | Intel® Core™ i7-4710HQ Processor (6M Cache, up to 3.50 GHz)

Intel also allows manufacturers to use an offset value so they can choose to lower the thermal throttling temperature by up to 15°C. My Y510P only uses an offset of 1°C so it starts to thermal throttle at 99°C instead of 100°C which is reasonable. Randomly disabling Intel Turbo Boost is not so reasonable.

On the main ThrottleStop screen, to the lower right where it says PROCHOT (processor hot), there will be a number that indicates what temperature your laptop is designed to thermal throttle at.

If you are seeing throttling at a core temperature of only 65°C, that is usually being triggered by power consumption.

If you have any more ThrottleStop related questions, head over to the ThrottleStop Guide thread. The link is in my signature. The NBR mods have told me that they prefer this so individual threads like this one do not get off topic.

-------------------------------

Next to prochot there is no number for me. "If you are seeing throttling at a core temperature of only 65°C, that is usually being triggered by power consumption." Well is it safe then to use throttlestop and reach temperatures of 85 or is the power consumption too high then, or what do you exactly mean by that? Thank you for answering my questions =)

Edit: With throttlestop 7 beta 3 there is a number. =) If you may answer my question regarding power consumption ill be happy =)

 

Hi Unclewebb,

Thanks for the reply.

So yesterday, I finally undervolted my CPU by -100mV. Ran Prime95 small FFT (with maximum stress using AVX instructions), and the results were good:

* Under the default Lenovo throttling profile (no-turbo boost, 2.4Ghz maximum), I dropped around 7W under full load from 39W to 32W. Initial load temperatures also decreased from 74C to 68C.

* After implementing the stock Intel turbo-boost profile (max 3.2Ghz across 4 cores), I re-ran Prime95, and was able to reach 3Ghz across all 4 cores for a short period, where the power consumption reached 54W, and temperatures hit 88C. Within a few seconds, the CPU throttled down to 2.4Ghz. Can't remember exactly why.

* Next I tried TS bench, and was able to hit 3.2Ghz on all 4 cores indefinitely, with relatively low temps. (Around 70-75C from memory).

Finally, I later tried it on my X240, but the gains were considerably less, and anything less than -60mV UV caused BSOD. But on the bright side, Prime95 (Small FFT) showed that at 15W TDP (long-term), temps were 3C less, and I was able to achieve 0.1Ghz more on both cores. LOL.

 

You need to increase the power limits in the modded BIOS if you want to hold full Turbo Boost in Prime95. That is, of course, assuming the cooling can handle it.

 

F4rewell - Different manufacturers use different methods to throttle their computers. When a computer starts to throttle at a very low temperature like 65°C, this throttling method might be based on power consumption of the CPU. A CPU that is rated to be able to run reliably at up to 47 watts might be told by a manufacturer to start slowing down way sooner than this when the power consumption is only at 20 watts.

Some of these throttling schemes that I have seen over the years do not seem to make any sense at all. Without being one of the engineers in the little white coats, it is impossible to say what they were thinking. Big companies are full of people that like to micro manage other departments. If someone, somewhere decides that 65°C is an appropriate temperature for their morning coffee, they might also conclude that 65°C is an appropriate temperature for a CPU. It is not. Intel designs their CPUs to run 100% reliably at up to 100°C for most models. Does ThrottleStop 7 show PROCHOT 100? That is the Intel specified maximum safe operating temperature for your CPU. Intel designed your CPU so when it reaches this temperature, it will automatically slow down as much as necessary to prevent the CPU from going beyond this temperature so it remains on the safe side of 100°C.

Anyway, I was helping a Y50 owner out today and he provided me with some log file data while he was using ThrottleStop while gaming. The temperatures of both his CPU and GPU were great and well within specification. With the help of ThrottleStop, his CPU was finally running at its full Intel rated speed and using full Turbo Boost while gaming like it is supposed to be doing. It is your laptop so you need to decide what you want to do. You can use ThrottleStop and try to force your CPU to run at its full Intel rated speed or you can use your laptop as is. I got fed up with my Lenovo Y510P throttling the CPU so I use ThrottleStop and run it as fast as possible. Problem solved.

D2 Ultima - Adjusting this INI option in ThrottleStop will not release a nuclear war head and your computer will not catch on fire.

In some situations, adjusting the timer resolution might help fix some obscure problem so I included a hidden option in ThrottleStop to control this. I actually added 2 hidden options so you can control the timer resolution separately when plugged in (AC) or when on battery power (DC).

When ThrottleStop starts up, it checks the INI file. If ThrottleStop does not find these options, it concludes that the user does not want to change these values. After doing some testing I found that leaving the timer at a resolution of 15.6 milliseconds when on battery power would help maximize C State residency time which in theory should minimize power consumption. That is a good thing when running on battery power.

I also concluded that if a user is plugged in, setting the timer resolution to a slightly lower number might improve performance or for some users, it might help reduce micro stuttering. I reached in my hat and pulled out lucky number 5. If ThrottleStop starts up and does not find a value in your INI file, it tries to do you a favor and automatically sets the timer resolution to 5 ms. No one has complained yet and one user recently said that my tools seem to get rid of his micro stuttering issues so it looks like doing this was a good idea.

If you want to play with this setting, open up the ThrottleStop INI file and if you do not see this option, you can edit the INI file and add this option to it.

TimePeriodAC=5

You can set it to whatever value you like. At the moment the lowest you can set this to is 1 ms. If anyone can convince me that a setting of 0.5 ms is necessary, I will consider editing ThrottleStop to make this possible. A lower limit of 1 ms seems OK to me.

The Windows default value of 15.625 ms is exactly one sixty fourth of 1000 ms. ( 1000 ms / 64 = 15.625 ms ) and 64 decimal is exactly 1000000 in binary.
I couldn't figure out where Microsoft came up with the 15.6 ms number from but that makes sense.

 

So does that mean that the current power limit is set at approx 39W if it never goes above during long-term boost?

On the plus side, the UV has made a huge impact on even idle temperatures, my fan almost never goes on any more. It also seems to use less power than my i5-4200U CPU during idle, makes Haswell ULV processors seem useless.

 

Should be 47W.

 

My thoughts exactly! I like creating my own ULV processors. This way if you ever need to go fast, you still can. ULV processors can be severely limited depending on what mood the manufacturer is in and how much stuff they decide to lock in the bios. Much easier to limit a fast CPU than it is to overclock a slow one.

It depends what app you are running. If you run the most recent version of Prime95, it is going to try and consume a lot more than 39W so it will probably settle at a different value. You can also be limited by current and there are multiple ways to throttle a CPU based on power. The 47 Watt limit can be set in multiple locations.

Over on TechInferno was a very interesting post by MrAngry. He showed what a 4700MQ is really capable of in the right hands after all throttling methods are disabled. Pulling 76 Watts out of a CPU that has a 47 Watt TDP rating is impressive. The cooling system in my Y510P cannot cope with that and would be doing some severe thermal throttling if I tried to run it at 76 Watts.

Help! Dell E6440 previously TDP throttled i7-4700MQ at ~47W, now at < 40W. Why?

Maybe if you ask MrAngry, he might share his secrets. No one on Tech|Inferno was interested in this magic.

 

=D! Now it makes perfect sense. Thanks!

 

@unclewebb, thank you so much for answering! That little program made my day!

 

Not as impressive as Mr. Fox's 140W, nice bit of hardware there. 76W stretches this entry level laptop to it's limits and can only run 76W for a couple of minutes with decent TIM and room temps of 24C or less before thermal throttling kicks in.

Not my secrets but Intels and the manufacturers secrets. No point in asking on that forum, I've given up due to too many posts going for "approval" and never seeing the light of day. The final straw was posting some help with W7 booting in UEFI with no legacy option ROMs but that never made it either. Strangely a couple of days later a post was made with info that could have been directly cut and pasted from my own missing post, perhaps it was taken from a much earlier posting I made elsewhere on a different forum.

I ended up resorting to editing posts for info to be shown so might be why it did not receive much interest but then I also made a post on here TDP and Power Limiting (Haswell) and seems to have received the same amount of interest. That's fine, gives me more time to do other things.

I am grateful to the original post on the other forum though as it got me interested enough to look further into the problem. Before I use to just trick the CPU into believing it was using less power as posted a while ago in this thread, but it's a bit risky especially battery wise. Now I can run only limited by thermals and no power limiting except when on battery which is limited to 38W and I like that much better. The trick method runs across the board so the battery could easily be overloaded in that case. Not good. Tricking the CPU can also affect the Thermal Margin register which is based on power vs temperature so that could possibly lead to problems if it gets too negative.

An example of tricking the CPU into thinking it's running Linpack at 166GFLOPS using less power than core-m. IOW setting power limits to say 5W would result in no power limiting though the temperatures are a bit of a giveaway.

 

I won't lie. What you've just said is over my head XD.

That being said, I'm finding it amazing that your 4700MQ was able to draw 76W indeed. My 4800MQ at 3.8GHz, stock voltage, current limit set to 100A and max power draw set to 100W each doesn't cross 62W (package power) under any load I've tried putting it through (mainly streaming, rendering x264 videos in handbrake under quick presets for fast FPS, and general gaming). What we found out about the MX CPUs though is that they draw more power and require more voltage to run than the MQ CPUs (at least on the haswell line). Tossing a 4900MQ to 4.2GHz can apparently be done somewhere around 0.1v LESS than on a 4930MX, and not even getting into the power draw.

That being said I want some nice, cool 980Ms (the power sippers compared to my power devourers; 780Ms) and a 4930MX so I can balance the load evenly hehehehe.

 

@D2 Ultima That's interesting. Is that 3.8GHz on 4 cores? Should be able to run 4 cores at 3.9GHz IIRC.

If you are using default voltage then max voltage will flaten off at 3.7GHz so you'll be using more or less the same voltage at 3.7GHz for any unlocked bins above that unless using adaptive voltage in interpolation mode. My 4700MQ needs about 1.000V at 3.4GHz, what does your 4800MQ use at 3.8GHz?

Please try this. Set to 4 threads at 15000 problem size and set each thread via task manager to seperate cores. Check to see if there is any throttling and maximum power used. It contains a slightly improved Linpack from the one I posted so you should be hitting the magic 200GFLOPS at 3.8GHz. If temperature is a problem then reduce the problem size, this will give shorter runs but also lower GFLOPS.

Please make sure you have everything saved before running, just in case.

 

Well, I don't have it at 3.8GHz right now as I ran into thermal throttling (95 deg C+) when doing Handbrake renders, but it worked just fine otherwise. And yes I know I could set it to 3.9, I simply had chosen 3.8 at the time for "I want to see" sake. I currently have it -50mV with max current 100A and max wattage 100W, so I will run this there and check. Later I can run this at stock voltage and 3.8GHz and tell you if it passes. I'll edit this post with the 3.5GHz results or if it crashed.

Edit: Welp, I left it for "3 runs" and it started thermal throttling like crazy after a minute or two, so I just ran it for one minute. Still managed to thermal throttle and pull a ton of power, so here you go:

Spoiler

I really need some CLU for this thing haha. See above post about CLU + P370EM heatsinks + 980Ms (double edit: just realized I didn't mention the last-gen model's heatsinks or CLU for cooling)

 

Dufus - Thanks for showing us what a Core i7-4700MQ is really capable of. With these sort of tricks, maybe you could smarten up Intel's ULV series too.

I checked out the MCHBAR locations you mentioned and my Y510P does not seem to be using any of those power registers to limit my CPU. My power limits are controlled by the MSR registers and those are locked. I would need to use the modified bios that svl7 created to try and push my CPU further.

The problem is cooling. My Y510P will start to thermal throttle when running at the 58 watt short term limit and it will continue to be on the edge of thermal throttling even after it throttles back to the long term 47 watt limit. Without significantly improved cooling, I don't have any need to go beyond the present limits.

Your screenshot showing 166 GFlops at 2.6 watts looks like something out of the future. I wonder how many years before it will be possible to get that kind of performance out of a mobile 2.6 watt device for real.

No worries D2 Ultima, me too. I need to avoid reading posts like that before bedtime. Let's just say that my daughter's ULV laptop experienced a couple of BSODs at 1:00 AM or so last night. It wasn't me.

 

Forget the useless ULV CPU's. What we need is a 4700MQ in an ultra-book form factor and weight (weight not exceeding 1.6KG), superb cooling, and a 6 cell battery. The performance of a workstation, combined with amazing battery life.

One can only dream on, lol. Anyway, no more off-topic from me, don't want to clutter your thread sir.

 

@D2 Ultima Thanks for running Linx. Your screen shot shows no data though perhaps because the "All" memory is selected which will take a very long time to complete. Running with negative offset is okay, I originally thought you were running with default voltage. No need to run default as every bit of power saved counts towards higher achievable performance. Try to keep it at the edge of thermal throttling or below, thermal throttling is just going to down clock the system. Reduce clocks if needed. Best IMO if you also make sure CPU performance BIAS is set to maximum, MSR 0x1B0 set to 0 on all cores. That is if you are looking for maximum performance.

Linpack is very useful towards stability testing and the GFLOPS results give a good indication if things are working as expected. For 3.8GHz 200GFLOPS should be obtainable if there is no throttling. You can also take solace in that you'll probably never run such a demanding load with real world applications. Something else to watch out for is that Haswell seems to dynamically adjust voltage, except when using fixed voltage, with temperature. Not up but down and by a fair amount. This means with an undervolt and coming off load with a hot CPU there may no longer be enough voltage for the lower bins resulting in a crash/BSOD.

Here's an example of Linx running 4 threads, one per core on a quad i7-4700MQ at ~3.44GHz.

It's possible but also it's a bit of a dirty trick in that it can push a machine way past it's power capabilities so should be used with extreme caution. Also appears to require setting before POST so not sure that a software solution would work.

I don't think the MCHBAR memory mapped registers are used that much for now but easily changed if required unlike the Embedded Controller method. EC PECI is a much tougher nut to crack which will likely require data sheets for the device and time spent working out what needs changing without changing something that will leave the machine bricked. I had a quick look at the Y510p 3.05 BIOS and unlocking PL1/2 MSR should be possible without a BIOS mod by changing the relevant NVRAM variable. This could be done through Windows if using UEFI/GPT or through the UEFI shell.

Yep, know what you mean. Haswell has a lot of promise but is kind of mostly destroyed with power consumption and heat generation. I would like to push further but I am also limited with power and temps to about 3.4GHz for the heavier loads with the hardware I have.

 

Thanks for clearing up how to set it right! Got a proper bench for you. Interestingly enough, the CPU sat at 3.3GHz for most of the test, only jumping to 3.5GHz in some parts. Contrast to earlier today (I have not restarted the PC since earlier) where the setup I had it sat at a full 3.5GHz constant until thermal throttling began. The 4th and 5th tests had thermal throttling going, probably the CPU not having enough time to rest inbetween runs. Amazing the Gflops you managed with a lower clock though. And I don't get why my CPU sat exactly at 3.3GHz for most of the tests. The temps were in the 70s for this, and this was when the temps hit highest. Near to the end of the runs is when they bumped to 3.5GHz. It was a weird experience. The power draw though, was even higher this time as you'll see below:

Spoiler

 

Set for 3 runs if it's throttling after that. Do you have anything running in the background? Some things you could try are

Run without any monitoring programs.
Set a higher priority
Set current limit to 120A with BD PROCHOT enabled

If using TS please clear the TDP log event. These can be triggered just by changes to the power limiting registers even though the power limit has not been exceeded. I vaguely remember there was a way to inform the CPU of a power change policy before making the change but didn't pay it much attention at the time and can not remember where I saw it.

Linpack does not run full 100% load from the start, it usually takes a little time before it comes on then some breathing space before the next run so if there's some throttling that might be why you see full multi sometimes but not always.

 

I'll try it for three runs and see. I'll also close everything but HWMonitor this time. My PC has never had problems with needing throttlestop to run; it's always kept its boost clocks, I've just been testing something by keeping Throttlestop running 24/7. I'll make a new post with the results within the hour.

Closed everything except my mouse software, GPU drivers, audio drivers, screenshot taking program and HWiNFO64 (not counting other background services of course) and even unplugged my second monitor. Basically idled at <1% util like that. Still thermal throttled in the run, but it kept close to the gigaflops in the first two runs.

Also, funnily enough, it STILL drew more power this time. And it followed the same 3.3GHz trend during the beginning of the bench, but got to 3.5GHz during the meat of the bench. Anyway, here you go Dufus:

Spoiler

 

Thanks for trying that. Your GFLOPS are comparable to what I use to get with an older 11 version of Linpack at problem size 20000, not sure why the newer version gives me higher GFLOPS but not yourself. As you can see by the power draw it is a demanding bench using up to 80J vs real world of around 62J. It seems the Clevo's are not as restrictive with power limits as other ODM's as was also shown by n=1.

 

It's possible your machine has better RAM than mine, or some other factor that affects these things. Also, if you managed to get your machine to sit at 3.44GHz constantly, that could be another reason. The problem when I set up linpack like you set it up is that it simply for some weird reason locks to 3.3GHz in most of the bench. It just... does. Throttlestop or not. In fact, I cannot get throttlestop to keep my machine at the turbo clocks 24/7, though they will happily stay so for hours while gaming or streaming. They drop under 100% load after a while when rendering though, usually to 3.4GHz or so. That's the other weird thing. I may end up having to use Intel XTU in the future, if only to set my bclk to 100 instead of 99.76 as it currently is. I really would like to see a flat number hehe.

Either way, when I get CLU I will be trying to keep 3.8 or 3.9GHz as my daily clocks on this machine. If I can max out at 90 degrees or so while rendering/streaming, then I have no need for anything else, as no other games make my machine as hot as streaming/handbrake rendering. Linpack however will probably always cause a solid overheat haha. At least unless I get the EM heatsinks. With some lapped EM heatsinks and CLU, I expect my CPU to never under any circumstances cross 85 degrees with max fans.

 

Can you run a ThrottleStop log file while this is happening and send it my way? That sounds like you are hitting either a power or current limit of some sort.

 

Where would I find the log file? My CPU won't go above its base clocks unless under load even if throttlestop is set to 37T (as you saw in previous screenshots, where the multiplier was hovering around 27, as I took the screenshots after the bench).

I just ticked "log" in the program but I don't know how long to run it for.

The CPU *DOES* clock up properly when in demand of some kind. The only two instances where it won't sit at 3.5GHz constantly under load is if I'm doing a Handbrake render (it'll go to 3.5 then downclock to ~3.4GHz and sit there constantly). But it won't use nearly as much wattage or whatever as his Linpack did. In your throttlestop bench program, the multiplier correctly sits at 3.5GHz constantly.

 

If you want, try putting the attached app in your TS folder and run it. It sets a couple of HSW registers so the CPU should run full turbo. Works for me under Windows balanced power plan with no changes to defaults. If you run TS in monitoring mode your multi's should be rock steady on full load. If they are dropping then that is an indication of throttling.

 

if i use thorottlestop for long periods of time (4-5 hours) would this cause my laptop to overheat? or will it reach 100c and throttle by its own ?

 

If you unzipped ThrottleStop into a folder then log files should show up in a sub folder called Logs. Exit ThrottleStop so it can write all of the info to the log before opening it up to see what it says.

The TS Bench does not use the AVX instructions so it consumes less power and produces less heat than Prime95 or LinX. A full multiplier in the TS Bench that drops when running LinX is a sign of some sort of throttling, power, thermal or both..

If you want to see ThrottleStop report a high multiplier when your CPU is lightly loaded or idle, you will need to disable all of your C States. I don't recommend doing that. It will make your laptop run hotter and on non-Extreme CPUs, disabling the C States will also limit the maximum amount of turbo boost.

undertrip - Using ThrottleStop in a laptop with poor cooling might cause it to go up to 100C. If this happens, your CPU will thermal throttle and slow down to protect itself. Intel doesn't like warranty claims so they built this protection into their hardware. Software like ThrottleStop cannot interfere with the CPU protecting itself from damage.

 

Well, unfortunately I can't upload any pictures right now because my ISP is purely retarded and needs to give me free internet for the next year to make up for their crap.

Hmm... Well, I can say that when it drops to 3.3GHz, the C0% in throttlestop reads 100% straight (set it to 8 threads to check this). When it's at 3.5GHz, it seems to top out at around 95% (average) C0%. When it's on 4 threads though and running the bench, the C0% sits at 100% flat for all cores regardless of the speed. Also, my power draw limit is set to 100W, which makes this odd. It only pulls 80W at max, but it's still somehow being limited? It's not temperature for certain; when it's stuck at 3.3GHz, the temp is quite low. It's when it goes to 3.5GHz (and power draw is a bit slacker) that the temp starts skyrocketing. I wonder if I should set my current limit to 120A or something? This is the part I don't get. Also, Handbrake which drops the speed slightly (to 3.4 or 3.3GHz) when I do a fairly long render doesn't draw NEARLY that much power (in fact, I've never gotten my system to draw over 52W on the package) before he asked me to run Linpack, which tops out at 80W. The only other thing is probably voltage? But the max voltage is usually attained when it's at the higher speed too.

It's like my system just doesn't feel like being 3.5GHz for that whole bench >_<

 

what do u think of the lenovo y50 cooling?

 

Here, I did a bench at 3.8GHz in TS, a couple 32M and then a 1024M one. Never dropped below 3.8GHz once.
It be here: Throttlestop Benches log - Pastebin.com

Also, it sat at 3.4GHz this time during the linpack (where it'd usually be 3.3), when I used 3.8GHz on the CPU, except for when it actually started getting hot (and was drawing less power). I know that the WAY something uses a CPU = the heat, rather than how much of it is used, but... to draw half power and heat it up an extra 20+ degrees? It's weird. Power draw this time topped out at 94W on package, though during that part of the bench it again was NOT at the max speed. Max speed (and most heat) sat around 50-60W to be honest.

Only thing I could think of is that I'd need to overvolt it for it to keep the max speed during that part of the bench. Blah. I wonder if I could make this thing draw like 140W... have half a mind to set current limit to 120A, power limit to 200W and add 100mV and see if it works. I bet I wouldn't be able to finish a single run of linpack without thermal throttle XD.

Also, I have no idea if that's even safe for my 4800MQ. I should probably find out what's the max current limit I should use for it.

Edit: Finally got that picture to upload. Finally. Here it be:

Spoiler

 

If you were to increase voltage to 1.4V at 3.8GHz and have no throttling then 90W would become close to 150W. However it is pointless doing this as it is just wasted energy, if anything you should try to lower voltage. Your residuals are good so stability should be good and you can use these as an indicator. If you start getting bad or different numbers between runs then something is causing error due to instability.

I see TDP log is still checked, did you uncheck this before the runs? This will show if there is limiting due to PL1/PL2/PP0/PP1. Limiting due to current or PL3 is indicated by EDP which is not shown by TS, so have to go by multiplier state. So if I understand correctly, you have increased clocks from 3.5GHz to 3.8GHz but are only seeing a small increase in GFLOPS not proportional to the clock increase. Certainly seems like throttling is going on if that's the case. Did you try running a TS log file while running Linx? If there is no throttling for your real world apps then IMO no real need to worry about it.

Forgot to answer one of your earlier questions, I use standard 2x 8GB 1600MTS DDR3L CL11. Nothing special.

 

I don't know. I tried ticking to make a log file and then running TS, but it didn't seem to create one. Was weird. Also, I'm back to normal -50mV and 3.5GHz now. I didn't try un-checking TDP log. I can do that a little later for sure. But yes, only when Handbrake is doing a long render or in your Linpack have I seen throttling. When I was overclocked, I even opened up OBS and started playing Dark Souls 2 while psuedo-streaming using "slow" compression and it basically sat at 3.8GHz all the time and never crossed 92 degrees (hovered between 78 and 88 usually) except for one time it apparently went down to 3.7GHz on 3 cores and 3.6GHz on one of the cores, but when I noticed, I was unable to replicate it later on (so I'm assuming I hit 100% on all cores at some point when I sat at a bonfire or somesuch) and it hiccupped or something.

Anyway, your little program thing helped too. My CPU sat at max turbo after using it. It was quite nice. I think I might use that more often; it even helped keep my 3.8 clocks when I OC'd.

Also, I REALLY need a tube of Liquid Ultra.

 

Exit ThrottleStop, go to the folder you have ThrottleStop in, run the program, check the Log File option, wait 10 seconds, exit ThrottleStop and in the same folder where ThrottleStop.exe is sitting, you should see a new folder called Logs. Inside that folder should be your log file data.

Your laptop has the thermal throttling temperature set to 96°C. Both HWiNFO and ThrottleStop show that your CPU was thermal throttling. When this happens, the multiplier will immediately start to drop. When the MHz drop, the GFlops also drop. That is probably why the GFlops go from 177.7 to 166.6 during your LinX run above. When thermal throttling kicks in, the CPU will rapidly adjust the multiplier and voltage just enough to keep the temps under control while maximizing performance. ThrottleStop does a good job of showing you how much multi you are actually achieving while your CPU is throttling. That's why I like seeing the log file. It tells me a lot more than just a screenshot does.

undertrip - I don't own a Y50 so I don't have any first hand information about the Y50 cooling system. ATG on the Y50 thread sent me a couple of log files while gaming at full speed and his CPU and GPU temperatures looked great.

When trying to get the most out of any laptop with a 4th Gen Quad Core Haswell processor, the cooling systems on all laptops are under designed. Some are just worse than others. Gaming oriented laptops are always going to be a compromise. No one is interested in lugging around a big thick laptop with a huge heatsink and a fan that sounds like a 747 is getting ready for take off. Powerful CPUs and GPUs shoved into smaller and smaller cases equals higher temperatures. The processing power of the 4th Gen AVX instructions is incredible but unfortunately, so is the heat. If you are mostly interested in gaming then the cooling system in the Y50 should be OK. You can always use ThrottleStop to try and get the right balance between performance and heat. I disagree with Lenovo disabling Turbo Boost while gaming in the Y510P and Y50 but at least the throttling method they used is easy to overcome by using ThrottleStop.

 

very informative, thanks

 

Well, I understand when it thermal throttles. What I'm saying is that when its speed is set fully to 3.3GHz, it's sitting at 70 degrees (not throttling). When its clocks hit 3.5GHz (and ironically it draws less power at this time) is when it starts to throttle due to high heat. What I could not figure out is why the machine would simply pick 3.3GHz to drop to during the non-hot part of the bench. But it doesn't happen anywhere else as far as I could see, so I'm just leaving it be. His high multiplier executable helped greatly, I keep my max turbo basically 24/7 now. Also, it's not a voltage limitation why it runs at 3.3GHz during the test, because after I did the 3.8GHz linpack test, in my screenshot you can see that I hit 1.109 max (except core 3 which hit 1.132 for some reason at one point) while running linpack. Well I loaded up Dark Souls 2 and started psuedo streaming using x264 and "slow" compression preset. I hit 100% on my CPU now and then (sometimes DS2 spikes its CPU usage when resting at a bonfire etc) and I didn't drop below 3.8GHz, and I was also using 1.193v on all cores almost constantly. No thermal throttling either; highest value was 92 degrees on core 1, 90 on core 2 and 87 on cores 3 and 4. So I guess something on my PC directly just doesn't like that particular part of linpack? Either that or it really is the current limit. No idea.