Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

qnxor

macrumors member
Original poster
May 2, 2014
78
0
I do agree with you on the GHz illusion when talking about Apple. In Luxmark (again off top sorry) no macbook pro to date can beat my G46VW with a 3840QM/HD4000/GTX660m due to power and temp throttling. I own a 2013 Macbook pro that scores 536 while using all OpenCL CPUs and GPUs where as my G46VW gets above 1000 using the sala scence. 1041 to be exact.

As far as again with the Intel Power balance again. I can't find any papers on the TDP budget but the creator of ThrottleStop states "The Intel Power Balance feature lets you decide whether the Intel CPU or Intel GPU should get the biggest share of the TDP budget. You can't get full CPU turbo and full GPU clocks while both are being heavily utilized no matter how good your cooling system is/what the cTDP you have your CPU set to.

I like to make sure even a non real world scenario can be sustained as I run BOINC for 24/7 on this laptop. I read your post so I shouldn't have brought it up. Just wanted to share some facts with that poster. :)

You raise a valid point, though the above discussion was regarding only the CPU. On that side, max load does not only mean max GHz. You can achieve max Ghz on all cores with a trivial multi-thread infinte loop, and that would hardly consume much power. True max load is when all CPU blocks (ALU, etc), i.e. most transistors, are switched on (current passes through them, not the leakage kind), which generates the maximum heat and maximum power draw. It's hard, or close to impossible, to achieve true max load, as some blocks are not used depending on the operation. Prime95 comes close to what is maximum achieveable as max power draw from a CPU (again, leaving GPU aside, which is actually a different chip but on the same die) but is also very far from a representative -- as in usual -- high load computing task, especially on laptops.

x264 is stressful enough, yet still far from what Prime95 can stress, as you've discovered too. I for one regard x264 as more representative of usual high load on laptops, which is why I recommend it for this MacOH stress tool.

I only included Prime95 in MacOH to reveal if the machine copes because I agree with you that all machines should be able to cope with any type of load. Note, however, that throttling can also triggered by the motherboard.

On the TDP budget, as you call it: The power draw limits and allocation scheme between the CPU and GPU should be in the Intel specs (the full ones). I haven't checked but I'd be surprised if they weren't since motherboard manufacturers need that, particularly for dual GPU laptops (or could it be non-public information that Intel releases only to mobo manufacturers?).

The reason why you can't get full turbo or full max load on both CPU and GPU is beacuse they are actually two separate chips on the same die, yet the TDP is for the whole die. It's a physical power draw limitation limitation which has nothing to do with the cooling performance. If you were to push it beyond that limit (*), your chip can suffer irremdiable lattice changes, e.g. electromigration.

(*) the TDP is not the actual limit. Most chips can be pushed beyond the TDP (if the mobo allows) and be just fine. The actual limit is usually higher, sometimes much higher, but there always is one.

Cheers
 
Last edited:

ha1o2surfer

macrumors 6502
Sep 24, 2013
425
46
You raise a valid point, though the above discussion was regarding only the CPU. On that side, max load does not only mean max GHz. You can achieve max Ghz on all cores with a trivial multi-thread infinte loop, and that would hardly consume much power. True max load is when all CPU blocks (ALU, etc), i.e. most transistors, are switched on (current passes through them, not the leakage kind), which generates the maximum heat and maximum power draw. It's hard, or close to impossible, to achieve true max load, as some blocks are not used depending on the operation. Prime95 comes close to what is maximum achieveable as max power draw from a CPU (again, leaving GPU aside, which is actually a different chip but on the same die) but is also very far from a representative -- as in usual -- high load computing task, especially on laptops.

x264 is stressful enough, yet still far from what Prime95 can stress, as you've discovered too. I for one regard x264 as more representative of usual high load on laptops, which is why I recommend it for this MacOH stress tool.

I only included Prime95 in MacOH to reveal if the machine copes because I agree with you that all machines should be able to cope with any type of load. Note, however, that throttling can also triggered by the motherboard.

On the TDP budget, as you call it: The power draw limits and allocation scheme between the CPU and GPU should be in the Intel specs (the full ones). I haven't checked but I'd be surprised if they weren't since motherboard manufacturers need that, particularly for dual GPU laptops (or could it be non-public information that Intel releases only to mobo manufacturers?).

The reason why you can't get full turbo or full max load on both CPU and GPU is beacuse they are actually two separate chips on the same die, yet the TDP is for the whole die. It's a physical power draw limitation limitation which has nothing to do with the cooling performance. If you were to push it beyond that limit (*), your chip can suffer irremdiable lattice changes, e.g. electromigration.

(*) the TDP is not the actual limit. Most chips can be pushed beyond the TDP (if the mobo allows) and be just fine. The actual limit is usually higher, sometimes much higher, but there always is one.

Cheers

I see, I'd just have to run software that uses, Intel Quick Sync, SSE, AVX, AES, FPU, ALU (I'm sure there are hundreds more lol) at the same time right? :rolleyes:

I've been playing around with the features of overclocking on my laptop for quite some time and was curious why I couldn't get the CPU and GPU to achieve full clocks at the same time. I had no idea about the damages that could occur, that is super interesting to know. (time to dig into that.. lol)

The motherboard can limit from what I found out as well as you've stated. I saw a custom TJMAX setting that I can set if I want to which is nice. Another thing I have been looking into is it seems Macbooks (haswell and ivybridge based ones) fully use Intels cTDP spec to help with power usage. Some person found that even if the temps are low (around 60c) the macbook still wants to cut clocks due to some other temp sensor which I believe is the power adapter. Since those overload so quickly under full CPU and GPU load as they only use 85 watts as you know. What do you think of that?

TDP limit for mobile chips seems to be 72 watts. I would agree with the TDP but after seeing the Configurable TDPs in my BIOS, you can set a TDP and it will stick to it no matter what. The power consumption increases in regards to "Long Duration Power Limit" and 'Short Duration Power Limit" in which it can draw up to that TDP for "n" amount of seconds (give temps are fine) then drop down. I really think the TDP is actually for the heatsink, it needs to be able to dissapate at least XX amount of watts. I know you probably are aware of all this but I'll type it out anyways for everyone else ;)
 

qnxor

macrumors member
Original poster
May 2, 2014
78
0
I see, I'd just have to run software that uses, Intel Quick Sync, SSE, AVX, AES, FPU, ALU (I'm sure there are hundreds more lol) at the same time right? :rolleyes:
For a true max load, yes, you would. That's why I said true max load may be impossible to reach, but Prime95 comes close to max power draw.

I've been playing around with the features of overclocking on my laptop for quite some time and was curious why I couldn't get the CPU and GPU to achieve full clocks at the same time. I had no idea about the damages that could occur, that is super interesting to know. (time to dig into that.. lol)
Electromigration is only one damaging effect that can occur in the lattice. Dig into wikipedia for more starting from that.

The motherboard can limit from what I found out as well as you've stated. I saw a custom TJMAX setting that I can set if I want to which is nice. Another thing I have been looking into is it seems Macbooks (haswell and ivybridge based ones) fully use Intels cTDP spec to help with power usage. Some person found that even if the temps are low (around 60c) the macbook still wants to cut clocks due to some other temp sensor which I believe is the power adapter. Since those overload so quickly under full CPU and GPU load as they only use 85 watts as you know. What do you think of that?

Things get specialized when it comes to motherboards and even more specialized when it comes to system manufacturers who design everything, including the case, power supply, dc regulator circuitry etc. When you have to deal with that kind of complexity in a small form factor like the laptops nowadays, the heat from many components affects the CPU temperature (they are in the same case, sometimes even under the same heatsink, like is the case with the Nvidia GPU and Intel CPU), and also vice versa (the CPU influences the temperature of other components). It's only natural to include more factors into the decision of what constitutes a threshold for throttling the CPU. I can't blame manufacturers for that, but I do blame some of them for not trying hard enough to design better cooling systems and/or better component layout and/or use more efficient components.

The rMBP late 2013 that I have, however, is pretty good in all respects in my opinion, and I was pleasantly surprised about it. Not many laptops can hold 3.3 GHz at full load in x264 even while on battery. It also runs impressively cool when not loaded: I never hear the fans unless I push it with Matlab, or other stressful apps like that, and I never had a problem keeping it on my lap. Even under full load with x264 and GPUTest the max temperature on the back is WAY lower than the Dell XPS 15 Late 2013 - about 47C measured on my rMBP vs 64C measured here (granted, they don't state their ambient temp, which is so important - in my case it was 24C). I'm hardly an Apple fan, but I can appreciate my rMBP.

TDP limit for mobile chips seems to be 72 watts. I would agree with the TDP but after seeing the Configurable TDPs in my BIOS, you can set a TDP and it will stick to it no matter what. The power consumption increases in regards to "Long Duration Power Limit" and 'Short Duration Power Limit" in which it can draw up to that TDP for "n" amount of seconds (give temps are fine) then drop down. I really think the TDP is actually for the heatsink, it needs to be able to dissapate at least XX amount of watts. I know you probably are aware of all this but I'll type it out anyways for everyone else ;)

Hmm, I'm not sure 72W is an official figure. The i7-4850HQ and most 4xx0HQ chips are 47W, which is the highest TDP in the mobile Haswell range AFAIK, I could be wrong. Indeed, there is a burst limit when the CPU is allowed to draw more power until it throttles. The relationship between heatsink and CPU power draw is a bit more complex :)

Edit: The i7-4940MX seems to be the only mobile Haswell cpu with a TDP higher than 47W (it's at 57W).
 
Last edited:

ha1o2surfer

macrumors 6502
Sep 24, 2013
425
46
Hmm, I'm not sure 72W is an official figure. The i7-4850HQ and most 4xx0HQ chips are 47W, which is the highest TDP in the mobile Haswell range AFAIK, I could be wrong. Indeed, there is a burst limit when the CPU is allowed to draw more power until it throttles. The relationship between heatsink and CPU power draw is a bit more complex :)

Edit: The i7-4940MX seems to be the only mobile Haswell cpu with a TDP higher than 47W (it's at 57W).

When you have an overclockable/Unlocked turbo bin mobile CPU (QM/XM) you can set the TDP to whatever you want and it will hold up to 4.1GHz turbo on all 4 cores for as long as you'd like under full load in my case.
 

qnxor

macrumors member
Original poster
May 2, 2014
78
0
When you have an overclockable/Unlocked turbo bin mobile CPU (QM/XM) you can set the TDP to whatever you want and it will hold up to 4.1GHz turbo on all 4 cores for as long as you'd like under full load in my case.

Well, the official TDP for those is still 47 and 57W respectively. Where did you find the 72W figure? Any Intel docs? In the bios, I believe you are changing power draw (or current) limit and cpu multipliers. The actual max power draw depends on the specific batch (not family) of the chip as it depends on the production characteristics for that batch. I don't think Intel guarantees more power draw than the spec'ed TDP (beyond the burst period) or else they'd get a ton of them returned for an exchange by all enthusiast overclockers. I also tested more than ten i7-3770K chips a while ago, some could go beyond 5.0GHz on all cores, others wouldn't budge more than 4.6 GHz.

In any case, does Apple make any laptops with QM or XM Haswell chips? I was under the impression that the best chip was HQ class.
 

yjchua95

macrumors 604
Apr 23, 2011
6,725
233
GVA, KUL, MEL (current), ZQN
Well, the official TDP for those is still 47 and 57W respectively. Where did you find the 72W figure? Any Intel docs? In the bios, I believe you are changing power draw (or current) limit and cpu multipliers. The actual max power draw depends on the specific batch (not family) of the chip as it depends on the production characteristics for that batch. I don't think Intel guarantees more power draw than the spec'ed TDP (beyond the burst period) or else they'd get a ton of them returned for an exchange by all enthusiast overclockers. I also tested more than ten i7-3770K chips a while ago, some could go beyond 5.0GHz on all cores, others wouldn't budge more than 4.6 GHz.

In any case, does Apple make any laptops with QM or XM Haswell chips? I was under the impression that the best chip was HQ class.

Apple only used QM chips in the Sandy Bridge and Ivy Bridge ones.

From Haswell onwards, it was all HQ.
 

ha1o2surfer

macrumors 6502
Sep 24, 2013
425
46
Well, the official TDP for those is still 47 and 57W respectively. Where did you find the 72W figure? Any Intel docs? In the bios, I believe you are changing power draw (or current) limit and cpu multipliers. The actual max power draw depends on the specific batch (not family) of the chip as it depends on the production characteristics for that batch. I don't think Intel guarantees more power draw than the spec'ed TDP (beyond the burst period) or else they'd get a ton of them returned for an exchange by all enthusiast overclockers. I also tested more than ten i7-3770K chips a while ago, some could go beyond 5.0GHz on all cores, others wouldn't budge more than 4.6 GHz.

In any case, does Apple make any laptops with QM or XM Haswell chips? I was under the impression that the best chip was HQ class QM and XM are Sandy and Ivy Bridge, HQ is Haswell.

Yes, Apple used a lot of the 3 series such as the 3612QM, 3720QM, 3740QM and so on.. although non of them perform to spec so it doesn't matter anyways lol.

I have found this limit by setting the TDP to 73 watts and watching Throttle Stop tell me something is throttling the TDP. Anything under and it wont throttle. The Long Duration Power Limit setting tells the CPU it can draw up to this much forever. You're right, it could be my machine can draw more power others and this could be all just my CPU.

For example, set LinPack to 5120MB and all 4 cores, if I monitor the power draw in Hardware monitor it says the package is drawing 72.99 watts and it will do that all day, 24/7 unless I specify something lower. I guess it's off topic but all I am trying to say is a CPU can at least reach it's max turbo boost speed on all 4 cores AT LEAST if given the thermal and power headroom. Anything above that is considered overclocking.

Sorry to derail this a little but, all this is super interesting to me and I spend time trying to figure out exactly what goes on so my nature is to shoot back facts and try to explain it so that I can be corrected if need be.

b5qmCrQ.jpg


What does yours get running the same test?
 

qnxor

macrumors member
Original poster
May 2, 2014
78
0
Your 72W figure is then most probably specific that that particular production batch, not the the CPU model.

In any case, could one then overclock earlier MBPs which used QM? How exactly do you set o/c settings (multipliers, current/power limits, etc)? I presme via software while the OS is running, given there is no BIOS in Apple laptops of the traditional kind, or is there?

Disclaimer: I used a mac for the first time 3 months ago when I got my rMBP, when I posted this thread actually.
 

marvinmac

macrumors newbie
Aug 29, 2014
3
0
Hi and thanks for this tool,

I have an issue when runnning this on a macbook pro late 2013.

When I run the command "bash macoh.sh" and go to the menu I select either "x" or "1" I get the following error:

Fetching and installing Intel Power Gadget into /Applications ...
######################################################################## 100.0%
[/Users/myusername/macoh/tmp/ipg.zip]
End-of-central-directory signature not found. Either this file is not
a zipfile, or it constitutes one disk of a multi-part archive. In the
latter case the central directory and zipfile comment will be found on
the last disk(s) of this archive.
unzip: cannot find zipfile directory in one of /Users/myusername/macoh/tmp/ipg.zip or
/Users/myusername/macoh/tmp/ipg.zip.zip, and cannot find /Users/myusername/macoh/tmp/ipg.zip.ZIP, period.

I check the path and the file ipg.zip is indeed there but maybe corrupt since i try to open it with a GUI unarchiver and find an "ipg.zip.cpgzip" and when I decompress that one I get the "ipg.zip" file again and so on.

Any ideas?

Thanx.
 

qnxor

macrumors member
Original poster
May 2, 2014
78
0
@marvinmac: Intel changed the URL of the IPG tool and the old one leads to a generic "page not found" message, but the server does not issue a 404 error so downloading it looks like it succeeds ...

I've updated the .conf file with the correct link. You can do either of the following two things:

- get the MacOH tool again from GitHub, or
- edit macoh.conf and replace the url_ipg="..." string with
Code:
url_ipg="https://software.intel.com/sites/default/files/managed/59/39/IntelPowerGadgetMac3.0.1.zip"
 

marvinmac

macrumors newbie
Aug 29, 2014
3
0
@marvinmac: Intel changed the URL of the IPG tool and the old one leads to a generic "page not found" message, but the server does not issue a 404 error so downloading it looks like it succeeds ...

I've updated the .conf file with the correct link. You can do either of the following two things:

- get the MacOH tool again from GitHub, or
- edit macoh.conf and replace the url_ipg="..." string with
Code:
url_ipg="https://software.intel.com/sites/default/files/managed/59/39/IntelPowerGadgetMac3.0.1.zip"

Thanks for the reply qnxor! I redownloaded the tool. It now installs Intel Power Gadget succesfully (when i selext "x" from the menu), but then later on it fails to install ImageMagik. Here is the message below:

Fetching ImageMagick into /Users/myusername/macoh/bin ...
######################################################################## 100.0%
tar: Unrecognized archive format
tar: Error exit delayed from previous errors.
 

qnxor

macrumors member
Original poster
May 2, 2014
78
0
@marvinmac: Same problem with ImageMagick (updated to 13.2 and they wiped 13.1). Same two solutions as above. Looking forward to seeing your results on the rMBP.
 

ha1o2surfer

macrumors 6502
Sep 24, 2013
425
46
Your 72W figure is then most probably specific that that particular production batch, not the the CPU model.

In any case, could one then overclock earlier MBPs which used QM? How exactly do you set o/c settings (multipliers, current/power limits, etc)? I presme via software while the OS is running, given there is no BIOS in Apple laptops of the traditional kind, or is there?

Disclaimer: I used a mac for the first time 3 months ago when I got my rMBP, when I posted this thread actually.

lol I rarely use my macbook.. so yeah.. lol and yeah, all in the BIOS. so no go for any macbooks. They couldn't handle the extra heat anyways.

For example, If my laptop is already overheating at 100c setting a TDP higher will just cause it to throttle sooner and essentially make it useless. lol It's really shame because these CPUs, when cooled properly, are amazing performers. My 3840qm is neck and neck with a stock 3770k.
 
Last edited:

grame

macrumors member
Nov 12, 2007
89
0
My 2014 2.2 base model. Is this good or bad?
 

Attachments

  • Screen Shot 2014-08-30 at 09.35.19.png
    Screen Shot 2014-08-30 at 09.35.19.png
    280 KB · Views: 200

johnnnw

macrumors 65816
Feb 7, 2013
1,214
21
bash: macoh.sh: No such file or directory

When I try and run it after downloading from github
 

johnnnw

macrumors 65816
Feb 7, 2013
1,214
21
Thanks ^, Got it working.

Can someone interpret this chart? 2012 Macbook Pro, Yosemite 10.10

20140830_172309_x264.png


20140830_172309_prime95.png
 
Last edited:

marvinmac

macrumors newbie
Aug 29, 2014
3
0
My test results

Hi guys. Here are my results on a macbook pro late 2013 16GB RAM. Regarding ambient temp I don't see any reading on the istat menus 5.01 (568).

What are your thoughts?
 

Attachments

  • 20140930-113133-x264.png
    20140930-113133-x264.png
    69.4 KB · Views: 191
  • 20140930-113133-prime95.png
    20140930-113133-prime95.png
    76 KB · Views: 161
  • 20140930-144947-gputest.png
    20140930-144947-gputest.png
    93.7 KB · Views: 181
  • 20140930-113133-x264-long.png
    20140930-113133-x264-long.png
    74.5 KB · Views: 146

NazgulRR

macrumors 6502
Oct 4, 2010
423
83
2013 Macbook Air 13" with Intel® Core™ i5-4250U Processor (3M Cache, up to 2.60 GHz) here.

How come the CPU never raises above 2.30GHz? Shouldn't it boost to 2.60GHz??

Also, that is some heavy throttling on the CPU when GPU is in use, no?

I re-ran the two tests twice with 12hours in between them, fresh boot, ambient T around 22C.

20141231-124809-x264.png


20141231-124809-gputest.png
 

WiZARD7

macrumors newbie
Feb 3, 2015
1
0
thanks for this great "app"

Here are my results from early 2011 macbook pro 13" with some dust, and the factory thermal grease. And then some cleaning, and applying Arctic Silver MX-4 instead the 4 yrs old factory stuff...
 

Attachments

  • 20150203-164837-prime95.png
    20150203-164837-prime95.png
    65.9 KB · Views: 154
  • 20150203-164837-x264.png
    20150203-164837-x264.png
    65.5 KB · Views: 153
  • 20150203-181524-prime95.png
    20150203-181524-prime95.png
    66.2 KB · Views: 135
  • 20150203-181524-x264.png
    20150203-181524-x264.png
    56.6 KB · Views: 145

matty1551

macrumors 6502
Jul 7, 2009
289
30
First off, thank you very much for coding this. I have been suspecting my MBP has been throttling a lot when using it in clamshell mode and I think the test results show that. The longer an intensive process, the more it seems to throttle I'm hoping the more informed MacRumors community can confirm by interpreting my results for me.

Early 2011 15" MBP 2.2GHz i7 2720QM

Internal Display 5 Minute
166gg35.jpg


Clamshell Mode 5 Minute
2l9qsqo.png


Internal Display 15 Minute
11uhuz4.png


Clamshell Mode 15 Minute
ei176c.png
 
Last edited:

casperes1996

macrumors 604
Jan 26, 2014
7,599
5,771
Horsens, Denmark
First off, thank you very much for coding this. I have been suspecting my MBP has been throttling a lot when using it in clamshell mode and I think the test results show that. The longer an intensive process, the more it seems to throttle I'm hoping the more informed MacRumors community can confirm by interpreting my results for me.

Early 2011 15" MBP 2.2GHz i7 2720QM

Internal Display 5 Minute
Image

Clamshell Mode 5 Minute
Image

Internal Display 15 Minute
Image

Clamshell Mode 15 Minute
Image


To me this just looks peculiar. It's definitely not heat based throttling. The rapid pace with which it goes down and back up, to and above the frequency it ran stable at with the built in display is quite weird. My first thought was that perhaps this had to do with a difference in power draw, as I can't at all see this being because of the thermal difference that is almost non existent, but I can't really imagine A) that the difference would matter this much, and B) That the graphs would look like this.
How is the performance in clamshell over normal? Try a Geekbench or Cinebench or something. Because in clamshell, yes the frequency goes way down, but it also goes above what it laid stable at in it's regular mode.
 

matty1551

macrumors 6502
Jul 7, 2009
289
30
To me this just looks peculiar. It's definitely not heat based throttling. The rapid pace with which it goes down and back up, to and above the frequency it ran stable at with the built in display is quite weird. My first thought was that perhaps this had to do with a difference in power draw, as I can't at all see this being because of the thermal difference that is almost non existent, but I can't really imagine A) that the difference would matter this much, and B) That the graphs would look like this.
How is the performance in clamshell over normal? Try a Geekbench or Cinebench or something. Because in clamshell, yes the frequency goes way down, but it also goes above what it laid stable at in it's regular mode.


Thanks for your response. I have an appt with the genius bar next week for this issue.

All my cpu based tasks take a lot longer in clamshell mode. For instance handbrake encodes can take a lot longer than they would without the display hooked up. Same thing happens with audio encoding with programs like audacity and logic. As soon as I disconnect the display, clock jumps back up to 3ghz and stays there for the most part.

Right now a 45min audio encode from aiff to 320kbps mp3 took wayyy longer. The cpu sat at .8ghz for the majority of the encode in clamshell.

3:20 Internal Display
11:30 Clamshell Mode
 

casperes1996

macrumors 604
Jan 26, 2014
7,599
5,771
Horsens, Denmark
Thanks for your response. I have an appt with the genius bar next week for this issue.

All my cpu based tasks take a lot longer in clamshell mode. For instance handbrake encodes can take a lot longer than they would without the display hooked up. Same thing happens with audio encoding with programs like audacity and logic. As soon as I disconnect the display, clock jumps back up to 3ghz and stays there for the most part.

Right now a 45min audio encode from aiff to 320kbps mp3 took wayyy longer. The cpu sat at .8ghz for the majority of the encode in clamshell.

3:20 Internal Display
11:30 Clamshell Mode

Doubting they'll give you many of the technical details, but I'm quite curious now, so if the Genius guys say anything interesting, please let me know.
That's quite a severe difference, and simple makes zero sense to me.
 

matty1551

macrumors 6502
Jul 7, 2009
289
30
Doubting they'll give you many of the technical details, but I'm quite curious now, so if the Genius guys say anything interesting, please let me know.
That's quite a severe difference, and simple makes zero sense to me.

I definitely will post back after my visit. Appreciate your input so far.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.