Q: 2011 MacBook Pro and Discrete Graphics Card

Correct me if I am wrong, but hypothetically, if Apple used the exact same Logic Board setup as before, BUT, they changed the solder material and the thermal paste, wouldn't that theoretically fix a huge portion of this issue?

My 8,2 Pro took over 3 1/2 years of intense daily usage for this problem to really begin to show and cause issues, constantly being used with both monitors and computer screens, and often running applications that would put it through cycles of intense on-and-off heat. I know others who have had it much, much sooner, and others later, and some 2011's are yet to have this happen but many eventually will. The problem is that MOST of these issues took over a year to show, with some exceptions.

I have found as a general rule that most people opting for the 15 (or former 17) Pro over the 13 Pro or Air tend to use those computers for more intensive applications (otherwise, the greater portability and much lower price of the 13 or Air would be a more logical purchase.) What that means is that people have been using these computers intensely for 3-4 years only to have the computer fail closer to the end of that timetable. Therefore, I don't think even short-term stress-testing will tell us how well Apple's fix corrects this issue in the future. Unfortunately, extended time is what is needed. Apple's program runs for 1 year, and unfortunately, I think it is going to take over a year to see how well this works.

I get mine back tomorrow. As I have a Retina Pro now, it will be used exclusively in clamshell mode, which to my understanding will force the auto graphics switching to run exclusively through the AMD GPU. We'll see how it goes... :-/

ZNickey wrote:

 

Correct me if I am wrong, but hypothetically, if Apple used the exact same Logic Board setup as before, BUT, they changed the solder material and the thermal paste, wouldn't that theoretically fix a huge portion of this issue?

 

My 8,2 Pro took over 3 1/2 years of intense daily usage for this problem to really begin to show and cause issues, constantly being used with both monitors and computer screens, and often running applications that would put it through cycles of intense on-and-off heat. I know others who have had it much, much sooner, and others later, and some 2011's are yet to have this happen but many eventually will. The problem is that MOST of these issues took over a year to show, with some exceptions.

 

I have found as a general rule that most people opting for the 15 (or former 17) Pro over the 13 Pro or Air tend to use those computers for more intensive applications (otherwise, the greater portability and much lower price of the 13 or Air would be a more logical purchase.) What that means is that people have been using these computers intensely for 3-4 years only to have the computer fail closer to the end of that timetable. Therefore, I don't think even short-term stress-testing will tell us how well Apple's fix corrects this issue in the future. Unfortunately, extended time is what is needed. Apple's program runs for 1 year, and unfortunately, I think it is going to take over a year to see how well this works.

 

I get mine back tomorrow. As I have a Retina Pro now, it will be used exclusively in clamshell mode, which to my understanding will force the auto graphics switching to run exclusively through the AMD GPU. We'll see how it goes... :-/

I think the question you are asking is can tests be devised that will simulate, in hours, years of operation?   It's my experience that such tests can be devised.  Further, I believe that over time, some of these tests can be improved such that test times can be reduced. 

We found that for some products certain combination of tests and test cycles were more effective than long duration "soak" type tests. 

Something associated with the solder used to connect the GPU to the circuit board seems to be a primary cause this recent epidemic.  You mention years of successful operation, while others experience repeated failures.  We know its no coincidence how certain operators and MBPs that frequently fail are brought together.   My bet is high failure groupings also have something in common like frequency of on/off cycles, mobility habits and mix of software.  Learning that will allow test engineers to compress that operational environment so the weakness can be detected in hours or minutes instead of weeks or months.

JroenVN wrote:

 

edit:

 

note that the chip they talk about in the post above is the optional AMD Radeon HD 6770M with 1 GB GDDR5 memory and normal configuration has the MD Radeon HD 6750M with 512 MB GDDR5 memory. (source: http://en.wikipedia.org/wiki/MacBook_Pro)

 

 

For the 6750 there is no speaking about tuning down the clockspeed.

 

there is this: The more powerful Radeon HD 6750M in place here really stands out compared to the small entry-level 6490M graphics card found in the 2 GHz MacBook Pro (MC721LL/A).  There's a great deal of characteristics that separate the model in review here from the standard model: 480 shaders instead of 160, a 128-bit memory bus instead of 64-bit one and 1GB of GDDR5 RAM as opposed to the 256MB of the starter model.  On top of that, the Core i7-2720QM in place here runs at a  10% higher base clock rate and 14% higher maximum turbo clock rate.  Even the on-board graphics card in this laptop can reach a slightly higher turbo clock rate of 1.3 GHz than the one mounted on the 2 GHz i7-2635M CPU of the standard model.  Finally, the CPU in place here has two more Sandy-Bridge-core features not available in the standard model: VT-d virtualization and AES functions.  Since Mac OS X does not currently support AES, this becomes a moot point. The final distinguishing factor between the two MBP15 versions is the hard drive: In the case of the 2.2 GHz model at hand here, a larger 750 GB HDD takes care of all your storage needs. (source: http://www.notebookcheck.net/Review-Apple-MacBook-Pro-15-Early-2011-2-2-GHz-quad -core-glare-type-screen.50344.0.html)

 

http://www.notebookcheck.net/AMD-Radeon-HD-6750M.43958.0.html

It would be interesting to know if the operating system is using the GPU for bit slicing kind of computing. (Bit slicing is the term we used with parallel computing in the 1980s and may be out of favor today)  In brief the idea is to split calculations and send to highest efficiency processing units, then bring pieces back together for a final answer.  The GPU has significant processing power that depending on the operating system can be used to create solutions. 

Reviewing some of the reports made during this discussion and while researching web articles it seems there have been some variation in the CPU and GPU calculation split.  My guess is that Yosemite increased GPU use because there seems to be a spike in failure reports following Yosemite's release date.  There are people who have been following this problem a lot longer and they probably have a better insight.

In a former life time while working with main frame computers we came across versions of the macro code that would allow low end systems to out perform high end systems. (there was guy in Canada who created some phenomenal stuff)  The difference was the algorithm that split calculations.

ZNickey wrote:

 

Correct me if I am wrong, but hypothetically, if Apple used the exact same Logic Board setup as before, BUT, they changed the solder material and the thermal paste, wouldn't that theoretically fix a huge portion of this issue? ....

Would someone using Yosemite and some of the other versions of the operating system run an experiment to see if the discrete graphic processor is being left on after the application that turned it on is no longer active?

I do WebKit testing which leads to doing unconventional stuff.  With gfxCardStatus installed the state of the integrated or discrete processor is displayed.  While using WebKit to access an online presentation I noticed that "a" for AMD discrete appeared.  When the window for the online presentation was closed the discrete processor remained active.

The test  -  Try situations that activate the discrete processor and then test to see what inactivates the discrete processor.

Closing the window viewing the presentation did not deactivate the discrete processor.  I had to quit WebKit to inactivate the discrete processor.  Relaunching WebKit did not active the discrete processor until returning to the online presentation.

Has anyone else noticed this also?

My theory is that if Yosemite has this flaw it may account for the increase failures since Yosemite's release.

jimoase wrote:

 

ZNickey wrote:

 

Correct me if I am wrong, but hypothetically, if Apple used the exact same Logic Board setup as before, BUT, they changed the solder material and the thermal paste, wouldn't that theoretically fix a huge portion of this issue? ....

 

Would someone using Yosemite and some of the other versions of the operating system run an experiment to see if the discrete graphic processor is being left on after the application that turned it on is no longer active?

 

I do WebKit testing which leads to doing unconventional stuff.  With gfxCardStatus installed the state of the integrated or discrete processor is displayed.  While using WebKit to access an online presentation I noticed that "a" for AMD discrete appeared.  When the window for the online presentation was closed the discrete processor remained active.

 

The test  -  Try situations that activate the discrete processor and then test to see what inactivates the discrete processor.

 

Closing the window viewing the presentation did not deactivate the discrete processor.  I had to quit WebKit to inactivate the discrete processor.  Relaunching WebKit did not active the discrete processor until returning to the online presentation.

 

Has anyone else noticed this also?

 

My theory is that if Yosemite has this flaw it may account for the increase failures since Yosemite's release.

Using an external drive with OS X10.9.5 I duplicated the 11 movie, google earth, iPhoto slide show, repair a hard drive test run yesterday with OS X 10.10.2. 

After 4 hours of running the battery stayed at full charge where it had discharged yesterday.

CPU temp peaked at 191 and held                    202 yesterday

GPU temp peaked at 158                                  162 yesterday

GPU diode temp peaked 165                            170 yesterday

Fan speed never exceeded 5300                      5,500...5,600 yesterday

Nothing exact about this and only conducted on one machine.  Doing the same tasks 10.9.5 is running cooler than 10.10.2. 

Conclusion:  If GPU heat translates into additional performance than 10.10.2 will bench mark faster than 10.9.5.

note: gfxCardStatus functions different when using 10.9.5 instead of displaying "i" or "a", as with 10.10.2,  it displayed "d" for dynamic

jimoase wrote:

 

ZNickey wrote:

 

Correct me if I am wrong, but hypothetically, if Apple used the exact same Logic Board setup as before, BUT, they changed the solder material and the thermal paste, wouldn't that theoretically fix a huge portion of this issue? ....

 

Would someone using Yosemite and some of the other versions of the operating system run an experiment to see if the discrete graphic processor is being left on after the application that turned it on is no longer active?

 

I do WebKit testing which leads to doing unconventional stuff.  With gfxCardStatus installed the state of the integrated or discrete processor is displayed.  While using WebKit to access an online presentation I noticed that "a" for AMD discrete appeared.  When the window for the online presentation was closed the discrete processor remained active.

 

The test  -  Try situations that activate the discrete processor and then test to see what inactivates the discrete processor.

 

Closing the window viewing the presentation did not deactivate the discrete processor.  I had to quit WebKit to inactivate the discrete processor.  Relaunching WebKit did not active the discrete processor until returning to the online presentation.

 

Has anyone else noticed this also?

 

My theory is that if Yosemite has this flaw it may account for the increase failures since Yosemite's release.

That would suggest that the discrete graphics card is a limited use item? It should not be used too often or continuously, otherwise, it will wear out quickly? That would be strange, no? What would be the problem of leaving the discrete card on or in use, apart from extra energy/battery consumption?

hansa69 wrote:

 

jimoase wrote:

 

ZNickey wrote:

 

Correct me if I am wrong, but hypothetically, if Apple used the exact same Logic Board setup as before, BUT, they changed the solder material and the thermal paste, wouldn't that theoretically fix a huge portion of this issue? ....

 

Would someone using Yosemite and some of the other versions of the operating system run an experiment to see if the discrete graphic processor is being left on after the application that turned it on is no longer active?

 

I do WebKit testing which leads to doing unconventional stuff.  With gfxCardStatus installed the state of the integrated or discrete processor is displayed.  While using WebKit to access an online presentation I noticed that "a" for AMD discrete appeared.  When the window for the online presentation was closed the discrete processor remained active.

 

The test  -  Try situations that activate the discrete processor and then test to see what inactivates the discrete processor.

 

Closing the window viewing the presentation did not deactivate the discrete processor.  I had to quit WebKit to inactivate the discrete processor.  Relaunching WebKit did not active the discrete processor until returning to the online presentation.

 

Has anyone else noticed this also?

 

My theory is that if Yosemite has this flaw it may account for the increase failures since Yosemite's release.

That would suggest that the discrete graphics card is a limited use item? It should not be used too often or continuously, otherwise, it will wear out quickly? That would be strange, no? What would be the problem of leaving the discrete card on or in use, apart from extra energy/battery consumption?

There are two graphic processors units (GPUs) because Apple is trying to deliver a higher performance machine.  When the 2011 MBP came to the market it was faster than previous big box Mac Pros.  To accomplish that feat and have better battery life besides Apple installed two GPUs.  One for light lifting and one for heavy lifting, if needed.  In effect Apple was repeating what Intel did back in the 386 central processor (CPU) days, they were adding a math co-processor.  Only in this case the math co-processor also had a video driver section that could also be switch in and out.

The video section is used for driving external monitors etc..

But .... there is always a BUT... when the discrete graphic processor is turned on it uses significantly more electricity/power so if running on a battery it would be possible to discharge a battery in minutes verse hours if the discrete GPU is used a lot.  So the operating system has an algorithm that decides which a calculation needs a co-processor.

When the co-processor is used a lot, temps go up, fans begin to whir, the bottom of the machine gets hot.  So Apple purposely turns the co-processor, aka GPU, on and off  often during normal operation in an attempt to get performance without wearing the battery down too fast or creating too much heat.  It's that constant on/off cycling that causes expansion and contraction of the GPU chip and the repeated stressing of the solder joints that seems to be cracking the connections.

Leaving the processor on or off would probably result in a longer life span for those solder joints, BUT there are consequences.  An ideal test for Apple would be a software sequence that caused the operating system to repeatedly turn on and off the discrete processor on some varying cycle rate until they found the frequency that causes the most movement in the shortest period, then subject the board to that test while doing thermo chamber temps cycles.  I think Apple would discover a worst cast pattern from this kind of test sequence and then be able to make an optimize version of the software and thermo chamber sequences.

Since Apple is continuing to use co-processors this insight would probably drop their custom plug and play failure rate significantly for all their products.  In a former life time this is the kind of stuff that made my phone ring.

jimoase wrote:

 

There are two graphic processors units (GPUs) because Apple is trying to deliver a higher performance machine.  When the 2011 MBP came to the market it was faster than previous big box Mac Pros.  To accomplish that feat and have better battery life besides Apple installed two GPUs.  One for light lifting and one for heavy lifting, if needed.  In effect Apple was repeating what Intel did back in the 386 central processor (CPU) days, they were adding a math co-processor.  Only in this case the math co-processor also had a video driver section that could also be switch in and out.

Agree with most of what you wrote, but a small correction.

Apple dit not "install" 2 GPUs.

1 Is integrated in the CPU, "Intel HD Graphics"

They only added the AMD as extra for mentioned purpose.

D3us wrote:

 

jimoase wrote:

 

There are two graphic processors units (GPUs) because Apple is trying to deliver a higher performance machine.  When the 2011 MBP came to the market it was faster than previous big box Mac Pros.  To accomplish that feat and have better battery life besides Apple installed two GPUs.  One for light lifting and one for heavy lifting, if needed.  In effect Apple was repeating what Intel did back in the 386 central processor (CPU) days, they were adding a math co-processor.  Only in this case the math co-processor also had a video driver section that could also be switch in and out.

 

Agree with most of what you wrote, but a small correction.

Apple dit not "install" 2 GPUs.

1 Is integrated in the CPU, "Intel HD Graphics"

They only added the AMD as extra for mentioned purpose.

We might be splitting hairs here.  I think there are separate chips for the CPU, the Intel GPU and the AMD GPU.  Here is what steers me to think that:

They have distinct names and operate on different bus structures.  An internet search will probable find separate specifications and maybe individual pictures.  I have seen specs for the AMD GPU and Intel CPU.  I have not looked for the Intel GPU.

My son works for Intel and when I mentioned my graphic problem his first comment was something about probably a problem with an AMD processor.  This is when I first had the problem.  Neither of us had done any research at that time.  My son is involved with database network solutions so he is not associated with hardware engineering.

<Image Edited by Host to Remove Personal Information>

jimoase wrote:

 

D3us wrote:

 

jimoase wrote:

 

There are two graphic processors units (GPUs) because Apple is trying to deliver a higher performance machine.  When the 2011 MBP came to the market it was faster than previous big box Mac Pros.  To accomplish that feat and have better battery life besides Apple installed two GPUs.  One for light lifting and one for heavy lifting, if needed.  In effect Apple was repeating what Intel did back in the 386 central processor (CPU) days, they were adding a math co-processor.  Only in this case the math co-processor also had a video driver section that could also be switch in and out.

 

Agree with most of what you wrote, but a small correction.

Apple dit not "install" 2 GPUs.

1 Is integrated in the CPU, "Intel HD Graphics"

They only added the AMD as extra for mentioned purpose.

 

D3us ... you are right... the Intel GPU is a separate chip die mounted within the same chip as the CPU die.  Found a picture of a different CPU with a the GPU we are discussing as an example.   Thanks for pointing that out.   Poor day when you can't learn something ~ Vance Grannis, So St Paul, MN

Just thought this interesting and wanted to share...

I have been working for about 8 months with Executive Relations. After many round trips to Depot and the Apple Store and three logic board replacements it died again. At that point they told me my machine was no longer going to be fixed by Apple and that "I should consider recycling it and buying a new one."

Thank god I held onto it or I would not have been able to partake in the new Repair Program.

Now, here is where it gets interesting.

Rather than just send it in for repair, Apple has asked me to send my 17" MBP to their HQ so engineers can study it. They've had it since Wednesday morning. I am supposed to hear back by tomorrow what they are going to do.

Anyone else been asked to send their machine to California Headquarters?

@Hal,

Like death itself there can be many reasons for a machine dying.  Did you have problems with the graphics, i.e., did your MBP qualify for the program that began on 20  Feb. in the US and Canada and on 27 Feb. elsewhere?  I'm assuming that all motherboards exchanged under the program will be returned to Apple and that some or all of them may be analyzed.  I'm mot sure what more they would learn from other parts of the machine that are not on the motherboard, especially since Apple asked me for my password when I broughit my late 2011 MBP 17" to them and thus would have access to logs.

I got mine back today and it works fine thus far. VERY appreciative of Apple's willingness to go they extra mile beyond contractual terms of a warranty.

Interestingly, when it went to Apple it was a 2.3 i7...when it came back it was the 2.2. Not a big deal given the $300-600 repair was free and it's not the machine I currently use for heavy lifting tasks, but I was a little surprised.

So a FYI if you have a 2.3 GHz machine, it may come back to you as a 2.2!!! (then again a working 2.2 beats a 2.3 stuck at that ******* gray screen any day of the week!)

@jimoase,

Doesn't "d" stand for "discrete", I.e. The AMD GPU In gfxCard?  I believe the program gives you the option of forcing the laptop to use either the integrated or the discrete GPU, or to allow automatic switching.

Regarding Apple's continued use of a co-processor GPU on later models that apparently don't have the problems of the models that qualify for the repair program, I would assume therefore that those models don't have the same problems, not that the switching is taking place less often than considerations of energy consumption would dictate.  I don't know how long it took Apple to diagnose the cause of the problem, but thereafter the question of what to do about it remained.  Evident "solutions," I.e.,  changing the switching algorithm, or  throttling the GPU, might have appeared preferable to a total recall in light of estimates of the number of machines affected, even while engineering changes were made in future models.

Jimoase, wasn't really haisplitting, is important to know for what/how (it) can happen.

Most consumer cpus have integrated graphics these days.

Except high-end and server versions.

The dies got smaller, faster, resulting in cooler running cpus.

But... the gain in lower temps are lost again in adding the extra GPU core.

Then Haswell came, dies got smaller again, more instructions, faster cpus, again running cooler.

The even smaller die gave extra room so they integrated the voltage regulators on the cpu.

This ment cheaper to make pcbs due lesser components, smaller ones possible (like brix systems).

But... yup, a big but again, these voltage regulators generate heat too.

Intel might not agree on this, but the Haswells were a bit harder to overclock then Ivy Bridge as they got hotter allready at lower frequency.

Not much, but notable, this maybe is or might be improved in later stepping.

Or next generation cpus.