Automated tool to reveal throttling and overheating - GitHub

[whg]

Early 2011 MBP 15

Before my current 2012 rMBP I had the early 2011 MBP 15 that I gave to my son. Two years later, he bought his own rMBP (with which he is very happy) and I got the old one back. I recently upgraded it with 16 GB RAM and a 1 TB Samsung 850 EVO. Unfortunately, the cooling system is no longer OK and the fans reach max. RPM with the slightest load. I ordered a new cooling unit and arctic silver to get this issue fixed, hopefully. Here is the current status (see attachments). The GPU looks still OK? The temperature of the GPU was always much lower than the CPU by 10-15C.

 

[johnnnw]

Can't see any new results, but your old one was looking great anyway.

Post #186

 

[yjchua95]

Before my current 2012 rMBP I had the early 2011 MBP 15 that I gave to my son. Two years later, he bought his own rMBP (with which he is very happy) and I got the old one back. I recently upgraded it with 16 GB RAM and a 1 TB Samsung 850 EVO. Unfortunately, the cooling system is no longer OK and the fans reach max. RPM with the slightest load. I ordered a new cooling unit and arctic silver to get this issue fixed, hopefully. Here is the current status (see attachments). The GPU looks still OK? The temperature of the GPU was always much lower than the CPU by 10-15C.

The 2011 models are known for Radeongate, and that happens because of poor application of thermal paste. Besides, the laptop itself has a lousy thermal design and bad ventilation.

 

[matty1551]

Before my current 2012 rMBP I had the early 2011 MBP 15 that I gave to my son. Two years later, he bought his own rMBP (with which he is very happy) and I got the old one back. I recently upgraded it with 16 GB RAM and a 1 TB Samsung 850 EVO. Unfortunately, the cooling system is no longer OK and the fans reach max. RPM with the slightest load. I ordered a new cooling unit and arctic silver to get this issue fixed, hopefully. Here is the current status (see attachments). The GPU looks still OK? The temperature of the GPU was always much lower than the CPU by 10-15C.

Any chance you can run the same test with an external display hooked up? I really need results from someone with the same computer as me. Trying to prove my case to Apple.

 

[matty1551]

No worries, my pleasure. Fingers crossed with the Apple replacement/repair.

My visit didn't go well. I'm gonna try to keep this organized and concise but I've got a lot of info to share so here we go.

A tech called me and explained that my machine appeared normal. I asked what tests he ran and he listed all the various apple tests that he could. I asked if any of those allowed him to monitor the CPU frequency. Of course, the answer was "no". So I explained that if he did some real world testing with intel power gadget he should be able to see the difference in speed. I was told that he would do some more testing and he would call me back that same evening or this morning at latest.

That call never came so I called about noon to see if my computer was ready and went to pick it up. I didn't bother to ask questions because I already knew they would spew the same info the guy yesterday told me.

They left 2 logs on my desktop. I'm not sure if this was intentional or not. They are too long to post so I will link to both of them here:
CPU MultiTest w/o Display
CPU MultiTest w/ External Display

These seem easiest to read when opened in textedit. The only information of value I found in them was at the bottom it lists average processor power consumption and they are both a few watts lower w/ External Display. There is a correlation between Watts & GHz right? IMO this helps validate my complaint. I would guess the power consumption may drop even lower if they ran the test for longer. They only ran it for 12 minutes w/ external and 44 minutes w/o external.

So I guess it's time to call Apple and go up the chain? I mean is it worth all this? I've had to deal with multiple GPU failures with the unit so I almost feel at this point I shouldn't settle for anything less than perfection.

QNXOR, I will run a handbrake encode and send you the log file if you are still willing to turn it into a graph.

 

[qnxor]

I just released v1.4.0

https://github.com/qnxor/macoh/releases (download the zip)

Major change: I've coded my own FFT test to emulates a more humane CPU stress encountered more often in daily activities. It uses by default a number of threads equal to the number of physical cores (rather than logical cores). If your machine does not throttle in this test then you should be fine. If it throttles visibly in this test then it's not a good sign. You can change the number of threads from the menu. Set it to 2 to emulate a pretty light CPU stress -- if it still throttles with 2 threads then that's pretty bad. Remember to close all apps before running the test.


As a comparison: Handbrake uses a nr of threads equal to twice the number of phy cores (due to hyperthreading) which puts much more stress and heat. It also loads more CPU blocks than you normally can with most other apps. This type of stress is less common, unless you do a lot of encoding type of stuff. I run a lot of maths computations daily, all multi-threaded, all very intensive. I've almost never reached the level of stress with my maths computations that I reach with Handbrake.

Feel free to try both the new FFT test and the x264 test and post results. You may find that the x264 throttles whereas the FFT does not.

@whg: the first result looks pretty good (don't worry about 105 degrees, it's normal for the sandy bridge CPU it has). The second one is not ideal, the temperature keeps increasing as well (it didn't stabilize). If you look through this whole thread you'll see worse though (e.g. matty's).

@matty: First, @whg has the same CPU as you and his graph looks a lot better than yours (see above). You can run the new version of MacOH test yourself now, no need to send me anything. Sorry about the Apple employee experience, not unexpected, most of those guys really have close to zero knowledge (they just run some standard tests that were handed down to them). While you are in store, you may also want to simply run the same benchmark on a new laptop with similar specs and also on your laptop, so that they can see yours is twice as slow. You could simply run a Handbrake encoding of the same file with the same settings and record the time.

 

[qnxor]

v1.5.0 released

I just released v1.5.0: Added CPU power draw, stats, and Tmax and TDP overruns info:

https://github.com/qnxor/macoh/releases (download the zip)

It now shows the Tmax and TDP limits as well as the CPU power draw versus time, and tells you how much you are close/over the limits. Here is the x264 test on my rMBP Late 2013 with MacOH v1.5.0:

x264 test (8 threads), March 2015:

Notice how the power draw jumps above the TDP limit of 47W, then throttles and the power draw remains under 47W. This x264 test is very stressful, it uses 8 threads (twice the physical cores due to hyperthreading) and loads most of the CPU blocks simultaneously as I enabled many x264 encoding features. Most users are very unlikely to hit this kind of high loads in their daily activities, not even when gaming.

Below is the FFT test I coded recently (see v1.4.0 in the post above), which by default uses number threads = number of physical cores. It is still very stressful but more humane and realistic than the x264 test -- this is a more relevant test to run:

FFT test (4 threads), March 2015

... Also, I just discovered to my horror that my rMBP has deteriorated since last year ... either (i) the thermal paste has begun to dry up or (ii) there is a lot of dust. Below is the same x264 test run last year in May. Notice how slowly the temperature rises at the start (and also does not reach 100C), compared to how quickly it jumps to 100C above. This is normaly a sign that the thermal paste has dried up or lost thermal conductivity, not an effect of dust (dust only affects cooling after fans have kicked in).

x264 test (8 threads), May 2014

I'm curious to see graphs of the FFT test from you.

Enjoy.

 

[matty1551]

QNXOR, you are awesome! Thanks for updating this. Here are my x264 results. You can see that my long x264 test w/ external added about 8 minutes. This is very damning evidence right? I mean idk how apple can deny it if I show them this. Your FFT test did not work for me though. I copied the error at the bottom of this post.

x264 W/O External Display (Short)

x264 W/ External Display (Short)

x264 W/O External Display (Long)

x264 W/ External Display (Long)

Here is the FFT error:

Your choice: [Q=Quit] f


Max Temp = 100
number of nodes = 1
TDP(mWh)_0 = 45.00
Base Frequency = 2200.00(MHz)

Changing CPU priority may require your password.
Password:

Waiting 15 seconds to capture idle temperature ...
Starting FFT benchmark. May take a while and fans may go berserk ...

dyld: Library not loaded: /opt/local/lib/libgcc/libgomp.1.dylib
Referenced from: /Users/Matt/Downloads/macoh/fftwtest
Reason: image not found

Test finished. Cooling off for 15 seconds ...

 

[qnxor]

Yes, it looks bad when you add an external monitor ... it's not great even without the external monitor.

Can you download MacOH v1.5 that I released yesterday and re-run these? See the post just above yours here: https://forums.macrumors.com/threads/1731178/ ... it will also record and show power consumption and how much you are above/close to predefined temperature and power limits.

The FFT test will be more relevant (unless your daily work involve heavy computation/video encoding/etc).
I'll look into the FFT error, thanks for pinpointing it.

 

[mamarrash]

Waiting 15 seconds to capture idle temperature ...
Starting FFT benchmark. May take a while and fans may go berserk ...

dyld: Library not loaded: /opt/local/lib/libgcc/libgomp.1.dylib
Referenced from: /Users/Matt/Downloads/macoh/fftwtest
Reason: image not found

Test finished. Cooling off for 15 seconds ...

I also got this error

 

[qnxor]

Working on it https://github.com/qnxor/macoh/issues/8 ... I'll post back shortly

 

[qnxor]

v1.5.1 released

FFT error fixed in v1.5.1 hopefully, download it from:

https://github.com/qnxor/macoh/releases

I also fixed the number of default threads in the FFT test to phy_cores - 1 (instead of phy_cores).

I noticed a very interesting behaviour different between Mavericks and Yosemite


Below is Mavericks 10.9.5, the FFT test with 4 threads. It stays below the 47W TDP limit and does not throttle - pretty constant 3.3 GHz, and gets 11.0 FFTs/sec performance. Activity monitor shows 360% load, so less than 4 full cores (?!).

Below is Yosemite 10.10.3 beta, the FFT test with 4 threads. It goes above the 47W TDP limit, then throttles to about 3.0-3.1 GHz, but gets 13.3 FFTs/sec performance (!!!) ... Activity monitor shows 398% load, so 4 full cores.

Below is Yosemite 10.10.3 beta, the FFT test with 3 threads. It stays below the 47W TDP limit, does not throttle - pretty constant 3.3 GHz, and gets 11.6 FFTs/sec performance! ... still higher than Mavericks with 4 threads (?!?!). Activity monitor shows 301% load (3 full cores).

Both OS's were completely at idle when performing these tests.

It looks like either Mavericks is slower than Yosemite, or I've hit some sort of bug ...

I'm curious to see the same FFT test done by others with 3 and 4 threads on quad core CPUs.

Cheers

p.s. EDIT: After more careful testing, Both Mavericks and Yosemite when idle consume the same power and have the same temperature. When stressed, Yosemite performs better in my case, even though it throttles more ...

 

[mamarrash]

I have several results from today. The first batch is with external thunderbolt display connected. It affected greatly the results. I got throttle in all the tests.
(see the two firsts attached images)

I only could run the FFT test with the update. The MBP was without TB display.
The results i got then were very different. Almost no throttle in the GPUtest, and also no throttle in the x264 test.

(the other three images)

MBP retina mid 2012 - Yosemite 10.10.2 -Ambient temp 24° C

 

[mamarrash]

X264 test and FFT test 4 threads after 15 minutes of "cooling off" the machine a little bit and switching to integrated only. I believe that the GPU on affects in a great deal the outcoming of the results.

Things i´ve noticed.

My temp at idle with the integrated GPU is 50-55°C. With the GPU goes above 60°C

The ramp of temp in the second x264 test is not abrupt as the previously tests. But despite being almost no throttle, is not the same as a year before. It´s aging

 

[qnxor]

@mammarash these look good, and I see no difference between the new test and the one form 1 year ago -- both temp and freq look the same to me. It hasn't "aged". Mine however did change a lot ...

Your graph are missing some labels and info. EDIT: Fixed in v1.5.2:

https://github.com/qnxor/macoh/releases

Cheers

 

[mamarrash]

I see a steep curve between the idle temp and max temp, and it´s reached in less than a minute vs the year before chart, where the temp curve is less pronounced and less steep. But I think it´s OK too. It´s nearly a 3 years old machine.

I was worried about the first results and the throttling of the GPU. Perhaps it is connected with the fact that i was running an external display also (TB display)

The issues with the text are finally solved!. I run a FFT test with 3 threads. As you can see, my idle temp is now 70 °C (TB display attached, GPU switch on by default as always when you plug in an external display. I guess the Nvidia GPU is a little oven )

No throttling. 9.23 ffts/sec

And after. 4 threads. No throttling also. 10.98 ffts/sec

 

[qnxor]

@mamarrash, honestly I see both now and 1y ago reach max temp 105C in about 40-45 seconds in your case int he x264 test (also take into account that one starts at 45C idle, the other at 52C idle, probably different room temperature or background processes), can't really tell the difference in the rest either. It's an extremely stable performance for 1 year. If you look at mine from 1y ago and now, you'll see a huge difference -- last year it wasn't throttling at all, now it throttles pretty quickly and the performance dropped too from 42 fps to 36 fps ... I'm not a happy bunny

Nice to see that the newer generation CPU manages to get better performance at a lower clock speed. My Haswell based CPU with 4 threads runs at an average of 3.1 GHz but gets 13.3 FFTs/s (graph), whereas your Sandy Bridge based CPU runs at a higher 3.35 GHz on average but only gets 10.85 FFTs/sec (graph) ... well done to Intel, goes to show once again that clock speed is not the only answer

 

[mamarrash]

MBP late 2013 BTO 2.6. At idle. Yosemite 10.10.02
Room Temp 24°C. Almost same results june last year

FFT 4 threads: No throttling. 14.28 ffts/sec
FFT 3 threads: No throttling. 12.31 ffts/sec
x264: Almost identical last year result
GPUtest discrete: same

 

[qnxor]

MBP late 2013 BTO 2.6. At idle. Yosemite 10.10.02
Room Temp 24°C. Almost same results june last year

FFT 4 threads: No throttling. 14.28 ffts/sec
FFT 3 threads: No throttling. 12.31 ffts/sec
x264: Almost identical last year result
GPUtest discrete: same

Not severely, but it clearly throttles in the FFT 4 threads test (from 3.6 to 3.4). It also throttles in the 3 threads test but not as much (drops to about 3.5). For that CPU however, these results are good. The x264 will throttle on your CPU (it's basically an overclocked version of the 4850HQ that I have, which last year was not throttling in the x264 tes and was getting 43 fps)

 

[mamarrash]

Automated tool to reveal throttling and overheating - GitHub

You are right. It throttles in the FFT 4 and in the x264. But it's normal behaviour for an overclocked chip. Altough i'm getting 14.28 ffts in the test

 

[qnxor]

Yes you do . It's a neat little script ... you can now see how the CPU throttles to bring down tmperature or power draw.

 

[mamarrash]

Automated tool to reveal throttling and overheating - GitHub

Yes you do . It's a neat little script ... you can now see how the CPU throttles to bring down tmperature or power draw.

It's a great set of tests. It accurate displays the effects of temperature, design, and the GPU proximity over the CPU performance.

Thanks for this awesome tool!!

PS: I don't regret buying the BTO Intel, but I do know now that it'll never achieve (in a real scenario) the 3.8 Max freq that is in the specs. In the other hand, I'm not sure either if the i7-4850HQ achieves the 3.5 Turbo Max freq advertised.

i7-4960HQ

i7-4850HQ

 

[qnxor]

It's a great set of tests. It accurate displays the effects of temperature, design, and the GPU proximity over the CPU performance.

Thanks for this awesome tool!!

PS: I don't regret buying the BTO Intel, but I do know now that it'll never achieve (in a real scenario) the 3.8 Max freq that is in the specs. In the other hand, I'm not sure either if the i7-4850HQ achieves the 3.5 Turbo Max freq advertised.

I didn't say it does. What I said was that the 4850 can actually get better performance under (high) stress than the 49xx because the latter heats up quite a bit more and throttles. See for instance my x264 results from last year, it was getting better results in x264 than most 49xx.

 

[mamarrash]

Automated tool to reveal throttling and overheating - GitHub

I didn't say it does. What I said was that the 4850 can actually get better performance under (high) stress than the 49xx because the latter heats up quite a bit more and throttles. See for instance my x264 results from last year, it was getting better results in x264 than most 49xx.

I didn't said you said that . As a matter of fact I agree with you with the fact that the 4850 can actually perform better under high stress than 4960.

I was only pointing out that the specs for both processors were deceiving in the turbo Max freq advertised. They not reach those freqs in real scenarios

 

[qnxor]

The max turbo freq is reached if only 1 core is active. This applies for all CPUs (unless limits are removed via BIOS, which is not the case on MBPs). 2 cores usually loses 0.1 GHz, 4 cores loses 0.2 GHz and so on, tends to be 0.1GHz*log2(cores).