The ThrottleStop Guide

I don't know when Nvidia GPU temperature monitoring through the driver became compatible with Optimus but I think it was recently. This works correctly on my system with the latest driver versions so some day when I have the time I might try going back and install a 310 series driver to see if this works or not.

All of the data shows up properly when I hover the mouse over the TS 7 icon in the notification area.



Are you running 2 different versions of ThrottleStop from the same folder or at the same time? If I can duplicate your problem I might be able to fix it so any additional info would be appreciated.

Donation software is simply not a viable option. With my RealTemp program, at the peak, I was bringing in something like 1 cent for every 100 downloads. I think ThrottleStop was worse. My wife went out one day and collected bottles and made more in a day than I was making in a month. It was great that a few people were willing to help out but it was not enough to pay for new hardware or a new version of Visual C++ so development has slowed down to a crawl. Now it is just a hobby. I still like working on ThrottleStop but it takes forever to get anything accomplished.

Mr.Fox - Are you using the Task Scheduler and how closely did you follow the method in the second post in this thread?

http://forum.notebookreview.com/har...es/531329-throttlestop-guide.html#post6865107

This method continues to work well for me in Windows 7 or 8.1 so when a person is having problems, that's my first suggestion. Is ThrottleStop in a folder on your desktop or is it in a folder of Utilities somewhere on your hard drive? Once again, if I can duplicate the problem, I might be able to come up with a fix.

Thanks guys for your help. It's the dedicated users that are keeping TS alive.

 

Ok sorry i have missed that. But something doesn't make sense : As said, before when i opened TS it greatly reduced C°, and that only by disabling turbo. When closing it, C° would go up again, which means (and that's logical) that turbo was ON again... The TS version i used before & now are same, though different acting. That's the point i can't get...

 

Well, I've had (Optimus) GPU temperature monitoring available as far back as on 280.26 (installed on Windows 7), and I'm talking about 2011 drivers.

I have a root ThrottleStop folder where I keep all TS versions on compressed files as well as, currently, two subfolders, a ThrottleStop_600 folder that I'm currently using and a ThrottleStop_700b1 folder next to it. I make sure to have a single TS instance running at any given time. So, TS 6.00 and 7.00b1 are on separate folders and said folders are differently named. I take it renaming the .exe wouldn't help, right? Also, it should be noted that, just like GPU-Z, TS isn't listed on the Program Settings tab of the 3D Settings menu on the Nvidia control panel, and I imagine it wouldn't make a difference since TS 6.00 (and all previous versions) were never listed there either.

All that popup data you have on the screenshot is displayed correctly on TS 6.00, but TS 7.00b1 simply refuses to display the regular TS icon with the number within, as well as the GPU monitoring being forcefully disabled whenever I try to tick the Show GPU checkbox.

About donations, it couldn't hurt to have something like PitchInBox available, with clear information as to what funds would go to, and I believe people would be on board with the idea. It would be a win win situation, no?

 

Hello folks. Just wanted to update you all with the i7 4930MX CPU AW 14 experimentation. I got the the i7 4930MX this morning before 10:30 via Fedex International Priority and started the CPU upgrade process right away. It took me a couple hours since I wanted to be very gentle disassembling my AW 14. So when I put it all back together, I pressed the power button but nothing was turning on. I was panicking since I was thinking maybe I had fried the motherboard. Turns out that when I disassembled it again to check, I had forgotten to securely fasten some of the ribbon power cables. Once I did that, it powered fine. Unfortunately in the A08 BIOS, there is no option to change the multipliers or voltages there. However in ThrottleStop 7.00 Beta 1, turbo overclocking max is 80 and I am able to change the multipliers fine. I have done a max of 4.1 GHZ so far. Under TPL, I have the TDP level control unlocked which was was locked on the i7 4700MQ. Also unlike the i7 4700MQ which the PP0 Power Limit was limited to a max of 58, the i7 4930MX max limit is 83.

Anyway I did a quick video rendering test in Sony Vegas with a 30 second clip with Gaussian Blur and Glow Affects. I took a before and after picture of the video rendering results done with the i7 4700MQ at 3.5 GHZ (with BCLK overclock at 102 mHz) all cores first and then the i7 4930MX at 4.1 GHZ all cores. I have to say I am quite impressed with the performance and the temperatures were pretty okay under 100% load with this short test. Just to emphasize, I do have modified laptop coolers with desktop fans, mini-portable AC fans, HWINFO fan control I am using and I am still waiting on an ETS laptop extractor to put on the back vent to further optimize cooling. Again, will have to see how this is like in the long run under longer stress test but so far these short test have been promising. I'll post pictures of the video rendering test soon when I find a decent screen capturing program. I'm not too much a benchmark guy to be honest.

If you guys want to know the rendering times of the test before I post the pictures let me know.

EDIT: Pics below of the really short video render test.

This pic below was the render time for the i7 4700MQ at 3.5 GHZ with a BCLK overclock of 102.7mHz. For this one I forgot to take a picture of ThrottleStop with it.



The pic below was the render time for the i7 4930MX at 4.1 GHZ.



Again take this test with a grain of salt. It is just a short test to verify that the i7 4930MX is functional in the Alienware 14 (Haswell 2013 model). Based on this test, the i7 4930MX CPU overclocked at 4.1 GHZ cut the render time in half from 26 seconds to 13 seconds compared to the i7 4700MQ at 3.5 GHZ. Temperatures at the highest was 73 degrees Celsius for the i7 4930MX CPU at 4.1 GHZ under load. I know this wasn't a long test to test temperature but I still do plan to optimize the cooling once I get two ETS laptop heat extractors. Those will certainly help. I also am not really into doing high benching tests as mentioned earlier but I'll see.

EDIT 2: For all you AW 14 owners who are curious about upgrading to the i7 4930MX, I would say it is only worth it if you do any intensive video rendering or any other CPU heavy tasks. Also I would only do it if you have the necessary things to keep it cool like strong laptop coolers, heat extractors, AC fans and good thermal paste. In theory, an i7 4900MQ CPU would have been enough to hit 4.0 to 4.2 GHZ but because of the locked TDP at 47 and the lower maxed out PP Power Limit at 58 compared to 83 on the i7 4930MX, I went with the i7 4930MX.

 

Hope there will be a way to change the voltage in the next ThrottleStop when I plan to OC further in future.

 

Congrats on your new CPU senzazn12, for the record however the 47W limit of 4700MQ can be got around as explained earlier in the thread.


This was an old run of AVX2 at over 70W on the 4700MQ but being run on an entry level laptop temperatures are a problem at the higher powers.

 

@ Dufus Oh I didn't know that. Thanks for letting me know. The i7 4930MX CPU is pretty beast! I really see it being a benefit for rendering times on lengthy videos. I will just keep adding and experimenting with cooling mods since the default cooling is limited.

One thing I like about HWINFO fan control on the AW 14 is that since there is only one fan I'm not experiencing the bugs of the AW 18 users who use HWINFO fan control with their GPU fans not working when HWINFO fan control is being used manually since there aren't any GPU fans in the AW 14.

Also since the AW 14 is just using one GTX 765 and not SLI GT780 cards, the heat isn't so bad for practical use and short benching. Not too sure about long term benching but like I said I'm not really a bencher so I'm not going to be close to hitting 100% load all the time.

All in all, I'm very happy I took the risks in getting the i7 4930MX. At least I can confirm it has been running great so far using ThrottleStop 7.00 Beta 1.

So far it has crashed on 4.4 GHZ because I had dropped the core current voltage from 1.250 to 1.200. There could be other settings that could have caused this as well like Dynamic Offset Voltage and my TPL and TDP settings. I'll see as I do more tests.

 

tribaljet - Thanks for the feedback. The 2 separate folders with 2 different versions of ThrottleStop is probably not a good idea. Rename the ThrottleStop6.exe to TS_back_up or something like that and then try copying TS7 into the TS6 folder. There are values in the INI file that keep track of the identification number of each system tray icon. I think Windows gets confused when one instance of ThrottleStop.exe has a totally different ID number than a different instance but the program is still called ThrottleStop.exe. All this stuff is new to me so I will do some testing this weekend to try to sort this out. I finally have access to a Windows 8.1 laptop for testing purposes. Thanks for your help. The donation / begging for money days are over. I will continue to do what I can with the resources I have available.

senzazn12 - :thumbsup: on getting your 4930MX up and running. Always good to see someone risk their wallet in the name of science. Adjustable voltage for the 4th Gen CPUs will definitely be added to ThrottleStop in the near future. Have you tried running Intel XTU on your 4930MX? Does XTU allow you to access all of your multis and voltages? It always makes me feel good when I see a success story like your swap.

Dufus - Welcome back! The 4700MQ isn't a bad CPU, especially if you are lucky enough to have a decent heatsink and a bios that lets you adjust the turbo power limits to full speed.

 

@ unclewebb thank you so much! It was a very tedious project to do but very rewarding so far. I can confirm now that I am able to use Intel XTU to change the parameters. Upon reboot the changes that I made to voltages, multipliers and core current limit all worked. The reason I wanted voltages to be added in ThrottleStop is so that I wouldn't have to reboot each time that I made a change. Thanks again unclewebb. I will keep you all posted of any updates, findings and future screenshots.

 

+1 for donations. You deserve it.

 

What's the max core voltage you have on your 4700MQ? I remember some time ago that you undervolt the CPU.

 

Hi unclewebb:I get same ring as yours lenovo y510p with gt750m.I would like to use throttlestop with it.I am not really an expert can you advice the best setting for our laptop/ I use it for gaming mainly so any performance increas is welcome. Should i use the last version?
Thank for any help you may give

 

ciccio64 - I am not a gamer. The settings you use in ThrottleStop will depend on what games you play and how hot your laptop gets. The laptops that come with 2 GPUs and use SLI put out more heat than the single GPU Y510P models. In this case, depending on your room temperature, you might not be able to run your CPU at its full rated speed. My room temperature is only 20C and I have a single GPU so I go for maximum CPU speed.

The Y510P has a couple of CPU throttling issues to prevent overheating. In my opinion, these safety features are a little too conservative. For maximum CPU speed, I use version 7 so I can set the turbo multipliers to 36, 35, 34, 34. On the main screen check Set Multiplier and set that to 36T for maximum turbo performance. Also uncheck BD PROCHOT. Click on Turn On and away you go. What I like about the Y510P is that it's throttling issues are easy to solve. You do not need to check off either type of clock modulation since the Y510P does not seem to use these.

To reduce the CPU temperature you could lower the CPU voltage but if you do not know what you are doing you should leave this adjustment alone. Most 4700MQ CPUs will be OK when you set the Offset Voltage to -50 mV. Some of the better CPUs might be OK with less voltage but you need to do some thorough testing when playing with voltages.

senzazn12 - I was told years ago that donation is like a 4 letter word on NBR. I am not starving so no worries. I just wish I had more time for this project.

In theory the VID voltage can be adjusted up to 2.0 volts but it is hard to say how much voltage the CPU would actually end up with at that setting. If it got the full 2 volts while running Prime 95, the phrase "nuclear meltdown" comes to mind. I put a limit on the TS offset voltage feature to keep users from going too crazy. I think I have seen desktop users pushing over 1.50 V through their Haswell CPUs so I guess I need to open that feature up some more. The new static and adaptive voltage feature will allow access to the full 2 volts. That should be fun.

 

I was asking Dufus what his max undervolted voltage is. I want to have a better idea of Haswell's ability to undervolt, as well as see if 4700MQ undervolts as well as the 4930MX.

Looking forward to see the new features. There'll be more tweaking to do. So far for below 4GHz, I don't have to hit 1.1V.

 

Tks Unclewebb I shall download version 7 and give a try.Regarding the undervolting i would like to try (how exactely do that in throttlestop?).Would be nice to reduce the heat a bit.I came from an AMD computer were i was using rmclock to do that and actually was helping a lot.Again Tks for help and keep doing the great job

 

Another week or so before I'll be home.

kh90123, I use a -100mV offset to give 1.000V vcore at 34x multi. With the default 1.1V that would mean an increase of 20% in power and heat generation. Unless running a bclk strap (125, 166...) then IMO a negative offset should be doable by most to give better performance and reduced power consumption. I did post a rough guide but basically use negative offset to adjust the lowest stable voltage for the highest standard multi, (34x for 4700MQ) and then use core voltage to give additional voltage for unlocked multi's in adaptive mode. Some of the desktop boys reported some excess voltage with adaptive mode but I've not seen more than 0.020V difference in vcore from light load to heavy load.

A re-paste made the 20000 problem size doable in Linx but with just a 120W brick and single heat pipe I don't think this laptop is up to running the heavy loads at higher clocks than that.

 

Click on the TRL button and adjust the Haswell Offset Voltage slider to the left from 0 to -50 mV. Click on OK and you are done. This is about all I could do while using the 36 multiplier. As Dufus pointed out, if you set the max multi to 34 for 1, 2,3 or 4 cores active then you might be able to get away with a -100 mV offset. What you can run reliably depends on what apps you are running and how much heat your laptop is putting out. The trouble I was having was not during full load Prime 95 testing. My laptop would lose stability when lightly loaded at the desktop if I tried to push the offset voltage lower than -50 mV.

kh90123 - Max under volted voltage makes more sense than my interpretation of your question. I thought you were getting ready to strap on some N2 to see what a 4930MX can really do.

 

Hi guys,

Glad to see both Dufus & unclewebb are now sporting a 4700MQ!

I have a clevo W370ST laptop with a 4700MQ and 765m GPU. I've been playing with XTU a bit and am still scratching my head a bit. When i first got the laptop I undervolted by -50mV which was stable and cut the heat down well. I only recently saw the 'benchmark' button, and have been trying that. with core offset and cache offset at -50mV I get a 'score' of around 550. After comparing with the HWbot online tests, I downloaded a profile which was -50mV core, -2mV cache and 32,30,28,28 turbos on the cores which scores around 660 - quite a bit more.

The odd thing, though, is using 3Dmark to benchmark the GPU and CPU combined, the CPU settings that were giving good XTU scores (660) give poor 3Dmark scores and vice versa. Even stranger, when i reset the XTU settings to stock, and re-ran the benchmark, I was getting something like 630?! with excess voltage?!

I assume this is because the turbo bins were capped so low with the downloaded profile, but often, under steady state loading of 4 cores, with max allowed turbo ratios of say, 34,34,34,34 I hit TDP and the multiplier drops to around 27x

I'm interested in trying out the adaptive voltage mode and seeing if i can get a single core to run stable with the two unlocked bins, but I assume you guys are also TDP throttled by the 47W limit with 3 or 4 cores active?

 

Can someone help me to find a list of typical stable voltages of the actual 22nm Haswell CPUs?

This is my old graph for 130nm, 90nm, 65nm, 45nm.


Off cause I know that individual CPUs and Notebooks will need
a little bit different voltages. But a table like this might help to find
the right voltages faster.

Link: Gentoo Wiki Archives - HOWTO_Undervolt_a_Pentium_M_CPU

 

carbide - The XTU Benchmark is an interesting benchmark. With the multipliers maxed out to 36, 35, 34, 34 and with an offset voltage of -60 mV, I can run the entire test and stay right at the 47W limit. No significant throttling so the multi was staying at 33.97 to 34.00 when loaded for the entire test and it would shoot up towards 35 when the load was reduced at regular intervals during the test.



That's a pretty pathetic looking score. At the moment I am only using a single stick of memory. I am hoping that when I upgrade to 2 sticks and run my memory in dual channel that some benchmarks like this one will show an improvement. If anyone has some time to kill they could do an XTU dual vs single channel memory comparison for me.

All laptops run Adaptive voltage mode by default. Most users like to combine this with some offset voltage. The next version of ThrottleStop will let you run a fixed (static) voltage regardless of CPU load but hardly anyone with a laptop is going to be interested in doing that. I plan to give fixed voltage a try on my laptop. My theory is that sure, you will have more voltage when lightly loaded but I think with the CPU using the C6/C7 C States, this won't matter too much when plugged in. The advantage will be improved light load stability with less peak voltage needed for full load stability. It should be interesting and it will be nice to be able to use ThrottleStop to easily and quickly switch profiles without needing to reboot.

The new feature will look something like this. I just need to hook it all up and do some testing.

 

@unclewebb So the PP0 Power Limit could also be used to increase max TDP? For example if rated TDP on my i7 4930MX is 57W I can increase it to something like 70W correct? The max I can go for PP0 Power Limit on my i7 4930MX is 83W. Anything higher than 83W just defaults back to 83W.

I was wondering also what toes TDP Level Control do? It is unlocked on my i7 4930MX which wasn't on my i7 4700MQ. It only has the option of setting it to 1 or 2? Not sure what this does.

 

Looking forward to this.

 

Cool & Quiet: Undervolting results

These are the stable voltages of the
4800mq in my Dell M4800 Notebook (Haswell, 22nm)

MHz / MilliVolt
800 / 496
900 / 514
1000 / 533
1100 / 551
1200 / 570
1300 / 588
1400 / 607
1500 / 625
1600 / 643
1700 / 662
1800 / 680
1900 / 699
2000 / 716
2100 / 734
2200 / 753
2300 / 771
2400 / 790
2500 / 808
2600 / 827
2700 / 848
2800 / 878
2900 / 907
3000 / 932
3100 / 961
3200 / 985
3300 / 1015
3400 / 1044
3500 / 1073
3600 / 1103
3700 / 1127

The Frequency is on the left side / and the VCore is on the right (MilliVolt).
Tested with XTU Stresstest (ca. 15 MilliVolt above negative Stresstest).



800MHz: -170mV Offset
1200MHz: -150mV Offset
1600MHz: -130mV Offset
2700MHz: -70mV Offset
3500MHz: -70mV Offset

800MHz idle: 3,2 Watt (CPU+Mainboard+RAM+Harddisk)

 

PP0 is the power limit for the CPU cores and PP1 is the power limit for the Intel GPU. There is also a separate power limit for the entire CPU package which includes both PP0 and PP1. If the CPU Package is limited to say 45 Watts then it would not make any sense setting PP0 higher than this. The CPU cores can not consume more power than the entire CPU package. You can set PP0 less than the CPU Package value but trying to set it higher does not make any sense. Setting PP0 higher than the CPU Package will be ignored by the CPU.

The only possible use for this feature is if you want to create a CPU with a lower TDP. If you bought a 4700MQ with a 47W TDP but it burns through your lap when gaming, you could use this feature to easily turn your 4700MQ into a 37W 4702MQ. If your lap is still on fire at 37W, you could use the Clamp feature and convert your 4700MQ into something similar to a 15W 4500U low power CPU. The PP0 setting is an easy way to lower the power consumption and heat output of your laptop. The old school way to reduce the temperature of the CPU was to lower the multiplier or turn off Turbo Boost. Adjusting PP0 lower allows the CPU to still use Turbo Boost and run at full speed when lightly loaded and it will automatically throttle back under heavier loads.

Intel has provided manufacturers several ways to limit the performance of their CPUs if heat or power consumption is a problem. It is possible that by default, the PP0 register could be set to a very low value by the bios which could severely limit performance. If this throttling method is being used and the PP0 register is left unlocked, a user might be able to adjust their CPU a little higher as long as they did not go over the Intel rated TDP of the CPU.

TDP Level Control is an easy way for a manufacturer to change the TDP of a CPU while a user is in Windows without having to reboot. When you switch to battery power, your CPU could automatically drop down to the lower TDP. In the low power U CPUs, the TDP Level control can also limit the maximum multiplier, often times to only 8. Level 0 is the base level and the other 2 levels allow the CPU to go either higher or lower than the base TDP. On a 4930MX your multiplier and power limits are fully unlocked so I don't see that this Level Control would be of any use. It is not used in the 4700MQ/4800MQ/4900MQ CPUs. A manufacturer might want to include a 4930MX option but they are not required to leave the CPU fully unlocked. They could decide to lock the multiplier and power limits in the bios. The TDP Level Control could then be used to at least provide a user with some benefit for spending all that extra money on an Extreme CPU without heat or power consumption getting totally out of hand like some of the Alienware 18 owners like to do.

heinz2005 - Thanks for posting your table of voltages. That kind of data is always useful.

 

That's interesting, looking at the XTU inbuilt graphing, I see the core utilisation bounce up and down from 1-4 cores through the benchmark, maybe 5-6 times. That does change though depending on whether i'm running stock, -50mV, -60mV etc. whenever I'm drawing max TDP for more than a few seconds i suddenly see the multiplier start falling.

are you running throttlestop alongside XTU and giving the CPU the lions share of the TDP limit? I haven't tried adjusting this yet, may be worthwhile

do you offset the cache voltage to the same value as CPU? I'm unsure whether it causes any low load instability, same goes for its multiplier - it always seems to want to try to match the 1 active core multiplier speed, so I've let it, but there seems to be very little information about regarding its operation - whether its imperative it run at max active core speed etc.

I finally bit the bullet and flashed my clevo's bios to a Prema mod one, now I have a fully unlocked bios, XTU and throttlestop, good times!

What's the variance of your scores? I assume that one was with a -60mV offset? do you get higher as I did (to the tune of around 100 points!) with a completely stock default setup? I will see if I get a chance and pull a couple of sticks - I currently run 3x 4gb modules (triple channel?!) maybe that's why my scores are higher

I see. I was reading Dufus's post: http://forum.notebookreview.com/hardware-components-aftermarket-upgrades/742522-guide-haswell-voltage-adjustments.html and seeing as my XTU was so locked down due to the bios, I didn't notice the greyed out toggle which is already selected as adaptive! I've always left core voltage as default and applied an offset, though after looking at the graph i'm keen to try adaptive offset with a static voltage.
Your theory sounds quite sound. I would have assumed that with a fixed voltage you're just allowing a higher potential - as the core isn't drawing a high load current, it shouldn't be making excess heat, and neither should the VRM until it's put under load? i don't know, i guess the overclocking aficinados may have a pretty good idea, there must be a downside for intel to implement a floating voltage - I guess just the inherent resistance through the chip's circuitry is enough to raise temps and sap battery. I always remember back in the old days of core 2 overclocking how it was pretty much deemed step one.
It's funny, i was just sat here writing this on battery, nearing 20% with a -50mV offset and the laptop died and rebooted! i think perhaps my chip is not as good as yours for tolerating reduced voltages.

That TRL window selection looks great, am looking forward to the release!

 

See if these charts from Intel help explain things any better.



Chart 1 is default voltage, this is how my laptop came.

Chart 2 is using just offset. You can see the unlocked bins recieve no more voltage than the highest standard turbo multi.

Chart 3 is using adaptive. This can be used with offset voltage also. It provides extra voltage for the unlocked bins without having to run the highest standard multi at the same voltage as the highest unlocked multi. For my 4700MQ that means I run with a -100mV offset and use adaptive voltage to provide an extra 30mV and 60mV for 35x and 36x multi's respectively. The adaptive voltage will only kick in when the combination of core voltage and offset voltage is set higher than the VID for the highest standard multi. IOW for me to achieve 1.060V for the 36x multi with -100mV offset I need to set core voltage at 1.160mV. If I were using offset only then I would only manage -40mV instead of -100mV.

Chart 4 shows fixed voltage for all multi's.

 

Thanks Dufus. That chart and your guide explains a lot. I just had a quick look at a few points on the mid to lower end of the voltage curve and didn't notice much difference between default and adaptive so I thought they were exactly the same but higher up they are obviously not. Looks like a guy could spend a day or three testing the various voltage curves. I kind of like the keep it simple voltage curve in figure 4. Not ideal for a laptop but I can test that curve with some Prime95 and half hour later after a few BSODs, I can usually come up with a fixed voltage and know that my laptop is going to be reasonably stable. Them were the good old days of overclocking and voltage tweaking.

I originally thought that adding offset voltage to ThrottleStop was good enough but this explains why giving users some more control over their voltage curve is a good idea. Being able to quickly swap voltage curves with a keyboard shortcut is going to be an appreciated feature, especially if desktop users ever notice this. I was trying to enlighten them at overclock.net about ThrottleStop but unfortunately, the mods shut me down and made me remove the links to my free software from my sig. Very odd. My work on these new features must be getting the attention of somebody important.

carbide - XTU does reduce the number of active cores a handful of times during the benchmark test. When the ThrottleStop reported multi goes higher than 34, that's an obvious sign that the higher 1 core 36 multi and 2 core 35 multi are being used.

When testing, I had PP0 disabled and my CPU is locked to 47W / 58W for the long and short term power limits. Maybe this could be a good test for a 4700MQ. At default settings you could see what multi ThrottleStop reports and then see how much or how little voltage you need to maintain the full 34.00 multiplier. Perhaps there is a golden low volt 4700MQ out there that could run the XTU benchmark with the 34 multi without needing any voltage offset.

During XTU I assumed that the Intel GPU was using a fairly consistent percentage of the power budget so I haven't tried adjusting the Power Balance feature. I don't think this would make a difference.

With the bios mod, can you now run your 4700MQ long term beyond the 47W TDP limit?

Edit - I did another XTU run at the default voltage and default multis, 34, 33, 32, 32 and guess what?



Exact same score as overclocked. On this run the multi spent most of its time at 30 compared to 34 during the previous test. A benchmark score that doesn't change based on CPU speed is not a very useful CPU benchmark. I will stick to wPrime. It is multi threaded, consistent and tracks changes in CPU performance very well.

Edit 2 - And finally with Turbo Boost disabled and the CPU running like a slug I get this. The XTU benchmark is definitely not too CPU speed dependent.

 

The XTU bench is very dependent on RAM so not sure how useful it is. With dual channel 1600MTS RAM you should hit the high 700's on the 4700MQ and with faster RAM of 2000MTS and higher probably in the 800's if not throttling.

A lot of third party software wasn't so good at reporting Haswell offsets, perhaps Intel were not passing on enough secret info to them. This can make things harder to understand as to what is going on. HWinfo was fixed, not so sure about HWiMonitor. Last time I looked it would show erroneous settings of over 3V for core and cache voltage when using negative offsets.

The adaptive mode is IMO more for those using unlocked multi's and then more important when using stepped multi's rather than flat multi's. Offset on it's own should be fine when not using unlocked bins.

The coarse bclk strap is something else I need to look at when I have time. Earlier Intel info says the 4700MQ should support this so if that means being able to run 125MHz the 3.4GHz becomes over 4.2GHz. Not really useful for this laptop considering an extra 800MHz might require and extra 0.24V of core voltage assuming ~30mV per 100MHz increase. This would result in the max 70W core power at 3.4GHz becoming 70 * 1.25 * 3.4 * (1.24/1.00)^2 = 134W plus a few more additional Watts for package power at ~4.2GHz.

Of course how well the CPU scales and how much extra voltage it will need will be dependent on each chip with the 4930MX and 4940MX being king for Haswell mobile. Unfortunately there do not seem to be many people overclocking laptops so getting some real idea of what these chips can do can be difficult although the heat issue is apparent. I think a lot of people get put off when asking about overclocking and get a plethora of answers saying they are going to "fry" their laptop. I had my old C2D laptop CPU more than 30% HW overclocked for nearly 5 years and never had a problem with the CPU or the overclocked RAM. IMO as long as current and temperature is kept in check and the additional power draw is within the capabilities of the laptop hardware then overclocking works fine. Of course not using silly voltages is a given.

Here's an interesting 4930MX overclock result from Fugger

 

I agree. Overclocking a laptop is not as bad some think AS LONG as you keep an eye on temperatures and have strong cooling mods and coolers. I'm running the Alienware 14 fine with the i7 4930MX at 4.3 GHZ overclocked under normal to slightly heavy load. I got some cooling gear that just came in today and waiting for four more cooling gears to arrive in the next week or so. After that I can't wait to further test the i7 4930MX in combination with my new cooling mods. BTW, that is a crazy overclock by Mr. Fugger!

 

Hi there,

How can I tell what is throttling my CPU?
I have Lenovo T440s with i7-4600U. Here se part of the log:
2014-04-11 12:27:34 32.47 99.6 100.0 100.0 0 66 1.0525 24.8
2014-04-11 12:27:35 32.59 99.6 100.0 100.0 0 68 1.0525 24.8
2014-04-11 12:27:36 32.57 99.6 100.0 100.0 0 69 1.0525 24.9
2014-04-11 12:27:37 32.58 99.6 100.0 100.0 0 70 1.0286 24.8
2014-04-11 12:27:38 32.52 99.5 100.0 100.0 0 71 1.0286 24.9
2014-04-11 12:27:39 32.55 99.5 100.0 100.0 0 71 1.0525 24.8
2014-04-11 12:27:40 32.50 99.6 100.0 100.0 0 71 1.0286 24.9
2014-04-11 12:27:41 32.52 99.6 100.0 100.0 0 72 1.0286 24.9
2014-04-11 12:27:42 32.45 99.6 100.0 100.0 0 74 1.0525 24.9
2014-04-11 12:27:43 32.49 99.6 100.0 100.0 0 74 1.0023 24.8
2014-04-11 12:27:44 32.47 99.6 100.0 100.0 0 74 1.0286 24.9
2014-04-11 12:27:45 32.46 99.6 100.0 100.0 0 75 1.0525 24.9
2014-04-11 12:27:46 25.65 99.6 100.0 100.0 0 67 0.8615 15.7
2014-04-11 12:27:47 25.30 99.6 100.0 100.0 0 67 0.8615 14.9
2014-04-11 12:27:48 25.35 99.7 100.0 100.0 0 66 0.8997 15.0
2014-04-11 12:27:49 25.22 99.6 100.0 100.0 0 66 0.8348 14.9
2014-04-11 12:27:50 25.37 99.7 100.0 100.0 0 67 0.8615 14.9
2014-04-11 12:27:51 25.33 99.7 100.0 100.0 0 72 1.0286 14.9

I have changed only the ratio for 2 cores active to 33. I don't seem to understand how to make it stay longer than that at max clock. If it is throttling because of high temperature how can I tell? The settings for TPL are the defaults PPL - 25W, PPS - 25W TTL - 28 sec.

Thanks.

 

ARK | Intel® Core

DeXa - Intel designed the Core i7-4600U to have a long term TDP power limit of 15 Watts. For short bursts it can go higher than that but long term, that's the limit. If you look at the far right column in the log file that is exactly what it is showing. The CPU has slowed down just enough to keep just a hair under the built in 15W TDP limit.

Sadly, your CPU is working exactly as Intel intended. That's why I would not recommend a U CPU to my worst enemy. They are badly crippled and are not built for performance. I am not sure what is possible. I would never spend my own money on one of these CPUs and I don't have access to Intel's full documentation. I have not tried to figure out what can be done with Intel's 4th Gen low power U CPUs but I am assuming, probably not much.

 

That graph explains things well, think your thread: http://forum.notebookreview.com/hardware-components-aftermarket-upgrades/742522-guide-haswell-voltage-adjustments.html definitely explained it the best for me.

so, I've been doing some more testing, well, 'shooting from the hip' is probably a closer approximation, I'm not an overclocker as such by a long way, just have a little interest. Anyways, I pulled some ram:



The first test is with only one stick of 4gb 12800. It's samsung 11-11-11-28 so pretty basic stuff

The second one was in dual channel with 2x 4gb sticks, and the third with 3x 4gb. Re-testing with 3x4gb has seen it fluctuate up to about 638, never higher than the 2x though, so I wonder if i'd be better off sticking with 2 modules of better ram.

So, it would appear that the XTU is, as Dufus said, heavily ram biased. Think i may get a copy of wprime.

Indeed! :-D my BIOS as it looks now (apologies for poor photo taking)





before, i could adjust boot order, rapid storage, uefi etc. but that was about it, now it seems every supported MB option is unlocked, which is awesome...

Unfortunately, trying to adjust the memory timings/freq either is not saving, or these samsung value basics are already at their limits and refuse to accept manual timings (or i need to learn more about ram than the simple 4 latency adjustments!)

However, I have current & TDP control, pretty much everything in XTU and throttlestop is now unlocked.

The below is with TDP limited to 75W and processor current limit raised 10A to 69A. The test peaked at about 56W TDP, 1 active core momentarily reached the 31x multi, otherwise 4 cores never managed higher than 29x sustained, 30x peak.

.

This is where i shoot from the hip - at stock settings, XTU never got higher than about 27x with 4 active cores. With my new found power, I loaded up the CPU with OCCT's linpack test on 8 threads. Using XTU as a monitor, I kept pushing the max current limit in the hope of maxing out 4 active cores. I got to about 90A, about 90c on the hottest core, and a 32x multi.
It started bouncing at this point, and despite the fact i'm not even giving it the full ref voltage, I started getting a bit nervous about the MB's VRM's, I assume there's only a max current that can be drawn at a specific voltage, and that I'll trip the board before i do any damage but I don't know enough about overclocking and voltages to start pushing voltages and give the CPU free reign of current it can draw...

with one thread on OCCT test (i assume prime based), with a 75W TDP, I left 1 active core with both bins unlocked to 36x yet it never exceeded 35x and 33W. Weird. Perhaps the voltage could do with being raised, or the offset reduced.

Without being able to remove the TDP restriction, probably so. Otherwise a modded BIOS could do that, depends if there's any out there, would definitely be a buying choice for me in the future...

 

with 1600 i'm getting ~690 and that's with the raised TDP. I assume you're talking with low latency, quality ram?!

OCCT, I've just realised, hasn't been updated since march last year. Its voltage and freq reporting is all over the place, what do you use for realtime graphing?

I have tried some adjusting with XTU. I can get to about 102 which results in a mem speed of 1620 which promptly freezes the system. I can't adjust the strap multiplier in the bios, only max allowed speed

 

Thanks. The thing is ... I really do not need more performance. I need battery life, so in my case it is doing its job perfectly. I wanted to see what is possible ... just from curiosity. I did some changes and when nothing happened I thought I am doing something wrong. Either way thanks a lot for the info.

 

I think Unclewebb is looking more from an out and out performance perspective - the CPU is seen as crippled as it has the capability of much more performance, just that it's artificially held back by the low TDP limit. probably, as is often the same in the automotive industry, because it's easier to come up with a design, then impose a few limits and create a model line, than it is to start at the bottom, and end up having to completely redesign the product by the time you get to the high performance model because the architecture is too limiting.

I took a look at your laptop, an ultrabook with integrated graphics. Whilst to desktop owners 15W TDP may seem small, as you say, you need something small, light and portable, and there's no way of bending physics to have both the most protable, longest lasting battery, and highest TDP. It looks like a good balance of power and portability, i'd much rather that than two piece or a tablet!

 

The whole U series of low watt processors is mostly for marketing purposes. It gives consumers the illusion that the U series are magical processors that can operate with very low power consumption. The thing is, these processors are all coming down the same assembly line. To create a U, Intel disables some features, sets a register to limit the TDP and puts it in a low TDP box. What I am saying is that the U series is pointless. At idle when a CPU is spending 99% of its time in the low power C7 C State, there is not a significant difference in power consumption between any of them. The Quad cores consume a tiny amount more than the Dual Core CPUs in this state but that's about it. At full load, I can use ThrottleStop and adjust the PP0 Power Limit down to 15W and then my 4700MQ will consume exactly the same amount of power as a low power 15W U series CPU does.

There is even a way to disable 2 cores so with the appropriate software, you could convert your Quad core CPU into a Dual core by simply checking a box and rebooting. It would be great to see Intel include more options like this in XTU.

What I really do not like about the U series is that a manufacturer can use the TDP Level Down feature which locks the multiplier to 8 so it is running at only 800 MHz. In this mode, your CPU might only get up to 10 Watts of its 15 Watt TDP limit so it will not even be running at its rated TDP. I think most consumers that buy a CPU like the 4600U which is rated up to 3.3 GHz are going to be disappointed with its performance if it gets stuck at 0.8 GHz. That can be the reality of the U and as a consumer, it is impossible trying to find out how a manufacturer has decided to implement this feature. Intel has provided manufacturers with unlimited flexibility so a 4600U in one device can perform significantly different than the exact same CPU in a different laptop model.

If you buy a low power U CPU, the performance can be severely limited by a manufacturer and there is no easy way around those limits. If you buy a 47W CPU, you can run it at full speed or you can choose to limit it however you like. I like keeping my options open but the U series doesn't let me do that.

 

@unclewebb

I'm not sure why, but my CPU is being throttled automatically for some reason.
Even throttlestop cannot prevent it. I tried v6 and v7b1, both.
When my CPU is idle, turning off C state seems to lock the multiplier at 12x, but the more load my CPU is under, the lower the multiplier goes. Please help, this is very frustrating. I will run any test you tell me to, with any tools, and provide as many screenshots as you need. I just need my CPU to stop throttling. I cannot even open a new chrome tab without having to wait ~10secs for the tab to load.

 

Intel says that the thermal throttling temperature for a T5750 is only 85C so your CPU is probably over heating and slowing down.

ARK | Intel® Core™2 Duo Processor T5750 (2M Cache, 2.00 GHz, 667 MHz FSB)

With very little load on your CPU it is already running at 80C. That is way too hot, especially for a Core 2 Duo. ThrottleStop is not designed so you can overheat your CPU. Intel protects their CPUs from overheating and there is no software that can turn off that safety feature. That's a good thing. Fix your over heating problem and your CPU will probably be able to run at its full rated speed. Time to open it up and clean out the heatsink and fan.

Edit - You can try reducing the VID voltage. This will reduce heat output of your CPU but you still need to clean it out first.

 

Thanks. I can't believe I overlooked this. I didn't think TJmax could be lower than 90C. I will clean the insides of my laptop and report back.

 

I have MSI Ge60 with i7 4700mq, when I apply UV I can't see any effect on my processor max power dissipation and voltage, but when I apply more ten -150mV then it freezy, so it something react. Any sugestion?

 

I cleaned it out and it's running at a relatively cool 52C now! Thanks!
But I'm not sure what happened in the process. The left side of my keyboard has stopped working. I'm copy-pasting characters that I can't type -.-

 

txtsd - It sounds like you did not fully plug your keyboard back into the motherboard. The connectors are usually tiny so you have to be careful when putting it back together. Time to take it apart again and have a look.

diablomichal - When under volting the 4700MQ, you might not see a significant difference in your core temperatures. When your CPU is idle at the desktop, it will mostly be in the low power C7 state getting virtually no voltage. The offset voltage adjustment will not show any benefit when idle.

If you do some full load testing with a program like Prime95 - Small FFTs, lowering the voltage will let your CPU run a little faster but it will still be reaching the same 47W TDP limit. Same power consumption so you should not expect any difference in your CPU core temperature. The only difference is a small increase in MHz. When gaming, this small increase will not be noticed.

If lowering the voltage does not provide you with any benefit then do not adjust it lower. Undervolting the original 65nm Core 2 mobile CPU was a lot more noticeable.

Edit - I decided to do a quick test. I recommend Prime95 - Small FFTs because it can put a very consistent load on a CPU. With that test running I started out at default voltage and then I tried -50 mV, -75 mV and -100 mV offsets. For each picture, temperature and TDP are similar. The difference is that by lowering the voltage, the CPU can run faster. During this test my CPU can run about 250 MHz faster when undervolted 100 mV. 9% faster for free seems like a good deal to me. Interesting that in the 3 undervolting screen shots the VID voltage is exactly the same. Lowering the voltage allows the CPU to run faster so that seems to automatically increase the voltage back to the exact same level. If you were running a less stressful program that was not reaching the TDP limit then you should see the voltage decrease.

Default Voltage = 27.50 multiplier
http://i.imgur.com/VxX3rRn.png

-50 mV Offset = 28.75 multiplier
http://i.imgur.com/FSGuQcU.png

-75 mV Offset = 29.35 multiplier
http://i.imgur.com/GJUfNUc.png

-100 mV Offset = 30.04 multiplier
http://i.imgur.com/7aViWB6.png

Nice having a precise monitoring tool like ThrottleStop so you can see exactly what your CPU is doing.

 

Need help , i can`t get set higher multiplier. Pasteboard — Uploaded Image Acer Aspire 5750ZG

 

I did the same test and in my case I can't see any difference in results. In all test multiplier is the same 27.5 at few first minutes.

 

I plugged and unplugged it several times, but a few keys on the left side of the keyboard still don't work

 

Just standard 1600MTS CL 11. I removed XTU quite some time ago, it was useful at the start but now unnecessary. I might have run the XTU bench with HT disabled, cannot remember off hand, and have a couple of other registers set to get performance more to what I want as well as running 34,34,35,36. Everyone has their own needs. XTU score was somewhere around 775 if I remember correctly. My own preference is to run Linpack as this includes the new registers and provides an extreme load which is unlikely in real world operation but provides an ultimate test with figures that can be used to measure performance increase providing there are enough results to compare with other systems.

I would think the low power chips could be run at higher power without throttling providing cooling and hardware is sufficient. But perhaps people buying those are looking more for lower power usage than running them faster than intended.

I just use my own rough software to get results and this is where I found power throttling problems with heavy load and Haswell earlier however TS has an excellent indicator of TDP throttling and I would recommend resetting that before running any tests.

The small deviation from standard bclk typically doesn't gain much percentage wise and can hit stability problems quickly. The bclk strap should offer increases of 25% and perhaps more but have not seen any results with this being used on mobile chips.

 

XTU is one ridiculous benchmark, not only does it generate excessive heat, it's so heavily ram dependent it's not even funny. This is the first time I've seen 8GB vs 16GB of ram make a major difference in scores. At bone stock settings with my 4900MQ, running 16GB 2133 gets me around 890 +/- 5, but 8GB 2133 only gets 820 +/-5.

And if you're gunning for the absolute highest XTU score, forget about timings, XTU likes speed and speed only. (or more precisely, bandwidth) 2133 even at CL13 blows 1866 CL9 out of the water.

Oh and Dufus, I can confirm I've seen my 4900MQ draw 120+W when benching XTU at 4.2 GHz. It's absolutely insane.

 

You guys are funny (re: XTU being RAM biased).

Work(done)=CPU+RAM


Everything is as it should be in the universe.

 

The argument I believe is that it is too biased towards RAM. IIRC XTU uses a spin off of prime95 for that test and I do not remember RAM influencing the original prime95 benchmark as much. Even other benchmarks that use large amounts of RAM are not biased as much as XTU is.

Have you run the XTU bench Till?

XTU does have some improvements that could be made if they haven't been done already. When adjusting voltages with 3rd party SW XTU would not update with the new values. Running some tests on the iGD with different multi's and voltages would sometimes result in artifacts when not stable but it would pass the XTU stress test with flying colors (pun intended).

@rhutor, the power draw of Haswell is not so surprising but being able to run that on your laptop says good things about your hardware. There were a lot of discussions prior to Haswell release that it was going to be a power efficient CPU but as already seen in the posts on this thread that pushing performance can not be done without high power and keeping temperatures under control. During each shrink the heat dissipation problem is going to get worse unless some new technique in fabrication is found so while it might look like lower power chips are a result of the die shrink efficiency it also has to do with having to reduce power to keep thermals under control.

 

Looking throught the XTU install last night trying to find my presets files, I stumbled upon the test programs, I think there was prime, a couple of linpack .exe's and something else that slips my mind. I always thought XTU is fairly low on the load front - only see about 4 temp peaks throughout the test, and they produce less heat than OCCT, although OCCT is a steady state load, so i guess after a few mins there is the heat soak aspect.

@rhutor: I noticed that the difference from halving my ram also caused quite a big difference in the scores, I only tried this out after Unclewebb mentioned it for testing, I hope the issue is just swapping between single and dual channel, my results were 449 to 644, about a 30% loss in performance going down to single channel. I have a couple 4gb 1866 modules in the post that i'm excited to try out, am hoping they'll run a little cooler too...