MacBook Pro (Retina, Mid 2012) freeze with Yosemite

Took my rMBP in to Apple last night. It passed all the tests, but I had crash logs and a radar I had done. My retina 2012 was still crashing when starting Photoshop even after the latest Yosemite updates. So, 3-5 days. Right now I'm working on my iMac 2011 also with a major defect (left side dark problem). That's another story. Anyhow, I expect this will be the end of the matter for my retina. If it starts photoshop without a problem then I should be good to go.

Hi Corporate_gadfly,

Interesting. I also digged in this direction and had also some "oddities".

Since I mainly work at home with my 15 retina as main machine (13 retina now in waiting to sort out the issue of the 15 one) with up to 2 or 3 screens attached all the day (TB display, mid2009 27 inchs imac as screen, Samsung TV), I also encountered hard shutdown or freezes when connecting a TB or mDP cable.

I also believed that my use of the machine could shorten its life (machine idling at 50°C without screens and 80°C with screens). More times istatmenu reported me CPU temps of 104°C :-/

As Ntennie, I'm convinced that one of the problem we're speaking about is not mobo or software related.

I really wonder if others on this thread can see a difference when bending the notebook. When some says it takes 100-200 card switch to crash the notebook, what if you bend a bit your notebook in the 2 directions and continue to switch the cards ? Does it also take 100-200 switch to crash ? Do you also feel your notebook more stable if you don't move it ?

Hi Corporate_gadfly,

Interesting. I also digged in this direction and had also some "oddities".

Since I mainly work at home with my 15 retina as main machine (13 retina now in waiting to sort out the issue of the 15 one) with up to 2 or 3 screens attached all the day (TB display, mid2009 27 inchs imac as screen, Samsung TV), I also encountered hard shutdown or freezes when connecting a TB or mDP cable.

I also believed that my use of the machine could shorten its life (machine idling at 50°C without screens and 80°C with screens). More times istatmenu reported me CPU temps of 104°C :-/

As Ntennie, I'm convinced that one of the problem we're speaking about is not mobo or software related.

I really wonder if others on this thread can see a difference when bending the notebook. When some says it takes 100-200 card switch to crash the notebook, what if you bend a bit your notebook in the 2 directions and continue to switch the cards ? Does it also take 100-200 switch to crash ? Do you also feel your notebook more stable if you don't move it ?

I used the late 2013 retina mbp since Dec. 2013. No problems whatsoever. Upgraded to Yosemite. No issues.

And then bam, 4-5 weeks ago, multiple crashes a day.

Don't know what I can conclude from that?

I have this issue too. And I went to genius bar, they tested it and told me "Passed MRI. Passed VST".

When I use mac os 10.9, sometime the system will froze for few seconds but not like Yosemite never response again.

Without fix this issue, I just disable auto switching graphic, and haven't duplicate this issue again.

I found apple setup a fix for graphic issue at http://www.apple.com/support/macbookpro-videoissues/, does anyone know this is related issue or not?

I went to the genius bar and my system passed the new VST.

I installed a clean copy of 10.10.2 in a separate partition. I installed only Adobe Flash and Microsoft Silverlight.

I was able to duplicate the problem:

1. The screen is totally frozen, except that the mouse pointer still moves.

2. Upon restart, I see the log is full of "GPU Restart" attempts.

-David

@dem107: Can you post a snippet of the log showing a few "GPU Restart" attempts with whatever log entries are in between? And indicate which logfile they are coming from? Do you always see these "GPU Restart" attempts after you experience the problem? Do you also see "GPU Panics"?

I'm wondering if this points toward a key difference, and while maybe not telling us the root cause, might tell us why we're experiencing different behaviors. On my unit, I never saw the locked-screen-with-moving cursor problem; my screen would just go black, sometimes it would reboot sometimes I'd have to force it, and after logging in I'd be notified that this was due to a GPU Panic, not a GPU reset. It sounds like you aren't typically seeing the GPU Panic, but are getting this flurry of GPU Resets.

If that's true, then I think that could explain the locked-screen-with-cursor. It's not unusual of for drivers that deal with complex hardware—e.g. a GPU—to detect if the hardware is unable to communicate or is in an unexpected state or otherwise out of sync with the software, and for the software to try to address this by forcing a reset of the hardware. Those situations can arise due to software on the system, or they can arise due to issues with the hardware. I think this is what's happening when you see the "GPU Restart" in the log; it's not the problem, it's an attempt to fix the problem, and whatever caused the problem may be unavoidable (every chip on the market comes with a lengthy list of errata, and some of those include random glitches).

But all that's okay, because the driver in this case may be doing what it's supposed to do; detect the problem, and reset the system. Normally what happens in this case is that the hardware resets, the software reinitializes the hardware (and possibly itself), and then everything goes back to working normally. But what happens if the hardware doesn't reset properly. Again, this could be due to a hardware problem or a software problem. But what happens in this case is that the driver will just try to restart the hardware again, and you'll end up with a flurry of "Resets" in your log. Eventually, it may time out and force a reset to the entire system, but that depends on how the driver is written.

If we apply this to the Nvidia GPU, it matches the behavior that you and many on this list are reporting, the locked-screen-with-moving-cursor. On the Mac, compositing of windows to the frame buffer is handled entirely through the GPU, so if the GPU is being continually reset by the driver, screen updates would freeze. The exception to this is the cursor; while this _could_ be written to the frame buffer using the GPU, it's not required, so it sounds like Apple is handling this via software. So you see the cursor keep moving.

And this could also explain what frookt just verified, which is that when he booted his MacBook using Mavericks at the Apple Store, he experienced situations in which the screen would freeze for a few seconds; that could be a sign that the GPU was being reset, although in that case it reset properly.

What I'm leading up to here is the possibility that what we're seeing is being caused by two issues. There may be a Yosemite-specific software issue related to how the GPU is being reset that is causing this process to fail in Yosemite; as a trivial example, these hardware reset procedures often have timing elements, and it could be one of those timing elements got modified in the GPU driver used in Yosemite. BUT, the reason that we're only seeing this problem on a small number of MacBook Pros is due to a problem in hardware that is causing the GPU to glitch out in the first place. So it works like this:

• If you have a MacBook Pro in which the GPU doesn't glitch out, you don't get repeated reset attempts on Yosemite, so you never see the problem
• If you have a MacBook Pro in which the GPU does glitch, attempts to reset the GPU fail on Yosemite and you see the problem
• If you have a MacBook Pro in which the GPU does glitch, attempts to reset the GPU succeed in Mavericks, so you don't see the problem (except maybe a brief screen freeze)

What still doesn't make sense here is why swapping logic boards doesn't fix the problem by pushing your MacBook into the first bullet. Or why you'd see variations in behavior between user logins. Options are:

• Apple says they swapped the logic board but really didn't.
• Apple swapped with a "refurbed" logic board that came back from another user that had the problem
• That some batches of GPUs or logic boards have this glitchy problem, and it's just bad luck that you got another one with it (seems too coincidental)
• That whatever is glitching your GPU is not on the logic board, so it survives the logic board swap. One example is nrj45's GPU shorting theory, but there could be others.
• Something else

I'm just tossing this out as a theory. Interestingly, I also ran across a parallel discussion thread (fequent mid 2012 macbook pro retina kernel panic reboots possibly GPU related) that has been going on since late 2013 and still has activity. Most of these folks were reporting problems pre-Yosemite; it's not clear when some of these people report their computer "freezes" exactly what that means.

Here you go…

My early 2013 15" MBPr shows the symptom where the entire system becomes unresponsive except for the arrow cursor which can be moved. A force-power down is required to recover.

Sun Jan 25 06:14:31 2015

Event: GPU Reset

Data/Time: Sun Jan 25 06:14:31 2015

Application:

Path:

OS Version: Mac OS X Version 10.10.1 (Build 14B25)

Graphics Hardware: Intel HD Graphics 4000

Signature: 801

Report Data:

===========================================================================

Intel GPU Hang Summary

version 1.0.0

===========================================================================

-- Summary of Status for ALL Rings --

MAIN Ring is idle

MEDIA Ring is NOT enabled

BLT Ring is NOT enabled

===========================================================================

Essential Ring Registers

===========================================================================

-- MAIN Ring --

RING_HEAD = 0x00000048 (offset = 0x00000048, wrap count = 0, wait_for_cond = 0x0)

RING_TAIL = 0x00000048 (offset = 0x00000048)

RING_START = 0x10204000 (offset = 0x10204000)

RING_CTRL = 0x0000f001 (enabled, auto report disabled (0x00), not waiting, semaphore not waiting, length = 0x010 4KB pages)

-- MEDIA Ring --

RING_HEAD = 0x00000000 (offset = 0x00000000, wrap count = 0, wait_for_cond = 0x0)

RING_TAIL = 0x00000000 (offset = 0x00000000)

RING_START = 0x00000000 (offset = 0x00000000)

RING_CTRL = 0x00000000 (disabled, auto report disabled (0x00), not waiting, semaphore not waiting, length = 0x001 4KB pages)

-- BLT Ring --

RING_HEAD = 0x00000000 (offset = 0x00000000, wrap count = 0, wait_for_cond = 0x0)

RING_TAIL = 0x00000000 (offset = 0x00000000)

RING_START = 0x00000000 (offset = 0x00000000)

RING_CTRL = 0x00000000 (disabled, auto report disabled (0x00), not waiting, semaphore not waiting, length = 0x001 4KB pages)

- Common Slice INST_DONE register -

All Common Slice engines are done.

- INSTDONE_CS (=0xffffffff): Not done =

===========================================================================

MMIO Register Dump

===========================================================================

MAIN MMIO Registers

the same situation for me (I sent many material to apple support with no significant response)

Mon Feb 23 21:12:50 2015

Event: GPU Reset

Data/Time: Mon Feb 23 21:12:50 2015

Application:

Path:

OS Version: Mac OS X Version 10.10.2 (Build 14C109)

Graphics Hardware: Intel Iris Pro

Signature: 801

Report Data:

===========================================================================

Intel GPU Hang Summary

version 1.0.0

===========================================================================

-- Summary of Status for ALL Rings --

MAIN Ring is idle

MEDIA Ring is idle

BLT Ring is idle

VEBOX Ring is idle

===========================================================================

Essential Ring Registers

===========================================================================

-- MAIN Ring --

RING_HEAD = 0x00000000 (offset = 0x00000000, wrap count = 0, wait_for_cond = 0x0)

RING_TAIL = 0x00000000 (offset = 0x00000000)

RING_START = 0x10205000 (offset = 0x10205000)

RING_CTRL = 0x0000f001 (enabled, auto report disabled (0x00), not waiting, semaphore not waiting, length = 0x010 4KB pages)

-- MEDIA Ring --

RING_HEAD = 0x00000000 (offset = 0x00000000, wrap count = 0, wait_for_cond = 0x0)

RING_TAIL = 0x00000000 (offset = 0x00000000)

RING_START = 0x10225000 (offset = 0x10225000)

RING_CTRL = 0x0000f001 (enabled, auto report disabled (0x00), not waiting, semaphore not waiting, length = 0x010 4KB pages)

-- BLT Ring --

RING_HEAD = 0x00000000 (offset = 0x00000000, wrap count = 0, wait_for_cond = 0x0)

RING_TAIL = 0x00000000 (offset = 0x00000000)

RING_START = 0x10215000 (offset = 0x10215000)

RING_CTRL = 0x0000f001 (enabled, auto report disabled (0x00), not waiting, semaphore not waiting, length = 0x010 4KB pages)

-- VEBOX Ring --

RING_HEAD = 0x00000000 (offset = 0x00000000, wrap count = 0, wait_for_cond = 0x0)

RING_TAIL = 0x00000000 (offset = 0x00000000)

RING_START = 0x10235000 (offset = 0x10235000)

RING_CTRL = 0x0000f001 (enabled, auto report disabled (0x00), not waiting, semaphore not waiting, length = 0x010 4KB pages)

- Common Slice INST_DONE register -

All Common Slice engines are done.

- INSTDONE_CS (=0xffffffff): Not done =

> My 2012 non-retina MBP is on repair since last week, I was told I need a new logic board.

> This was before the repair program was announced but the non-retinas are not included so

>I probably will have to pay for something that doesn't fix the problem anyhow.

Got my 2012 non-retina MBP back yesterday, new logic board, not covered by the program

so I had to pay. Let's hope it fixes the problems but I have I've wasted a lot of money.

I just got my mid 2012 15" retina Mac Book Pro back from having the logic board replaced, and the problem still exists.

I was using Safari and all of a sudden everything froze, cursor and all.

And then it shutdown.

I don't have Apple Care, so I'm not sure what else they can really do. My MBP is eligible for the free repair, but I'm not sure what else they can do besides replace the logic board.. again.

WOW! I didn't realize there were so many here with this problem so this reply is for everyone as here's the fix that worked for me!

Download and install CUDA here: http://www.nvidia.com/object/macosx-cuda-6.5.46-driver.html

It will install a new Preference Pane in your System Preferences. Go to the CUDA Preference Pane to make sure you have the newest update.

Go to the Energy Saver Preference Pane and turn OFF Automatic Graphics Switching.

If you've had problems with Photoshop and had to turn off the "Graphic Processor Settings", go ahead and check "Use Graphics Processor" and click Advanced Settings.

Set the Drawing Mode to Advanced and check all except "Use OpenCL".

Click Ok to close both dialogs, Quit Photoshop and Restart your computer. Problem should be solved.

I've had no more problems since doing this on my mid-2012 MBP(r) on Dec. 27, 2014. Photoshop CS6 runs faster than it ever has!

Sorry I took so long to let this out!

WOW! I didn't realize there were so many here with this problem so this reply is for everyone as here's the fix that worked for me!

Download and install CUDA here: http://www.nvidia.com/object/macosx-cuda-6.5.46-driver.html

It will install a new Preference Pane in your System Preferences. Go to the CUDA Preference Pane to make sure you have the newest update.

Go to the Energy Saver Preference Pane and turn OFF Automatic Graphics Switching.

If you've had problems with Photoshop and had to turn off the "Graphic Processor Settings", go ahead and check "Use Graphics Processor" and click Advanced Settings.

Set the Drawing Mode to Advanced and check all except "Use OpenCL".

Click Ok to close both dialogs, Quit Photoshop and Restart your computer. Problem should be solved.

I've had no more problems since doing this on my mid-2012 MBP(r) on Dec. 27, 2014. Photoshop CS6 runs faster than it ever has!

Sorry I took so long to let this out!

@edoardopa and @Barry Fass-Holmes, I have three questions about the GPU Reset logs you posted:

• Are you sure that this GPU Reset is associated with your screen lockup? In other words, the timestamp matches?
• Like @dem107, do you see multiple "GPU Reset" log entries associated with a single lockup? @dem107 reported that on restart after his problem, the "log is full of 'GPU Restart' attempts"
• If so, what is the time delta between these multiple "GPU Restart" log entries that seem to be associated with the lockup?

The thing that I found unexpected about your GPU Reset logs was that they weren't for the Nvidia GPU. @Barry Fass-Holmes has his happening on the integrated Intel GPU (on what I assume is a mid-2012 Retina MacBook Pro), the Intel HD Graphics 4000. @edoardopa has his happening on the Intel Iris Pro, which I assume means he has a late-2013 Retina MacBook Pro.

At least for people with mid-2012 Retina MacBook Pros, If it's true that the GPU Resets are happening on the integrated Intel GPU, it seems unlikely that @JWDemon's CUDA driver update would address the problem, although turning off automatic graphics switching obviously would, simply because it forces use of the external GPU.

For whoever this may concern: The Graphics freeze ups are being caused by the switching from the Intel GPU to the Nvidia GPU. After placing my post last night I installed the newest CUDA update, which is this same one that I posted, and then I tested to see if the problem would still occur. I turned on Automatic Graphics Switching, Opened Photoshop and the system froze immediately.

No, this fix does not address the Intel GPU problem but the Intel GPU IS the problem. I tried just turning off the Automatic Graphics Switching but that didn't stop the freeze ups until I installed Nvidia's CUDA. Yes, it will use a bit more battery power but it's not excessive. As stated, I have had no freeze ups or any other GPU problems with this setup for the last 2 months.

To make the decision to use this procedure, you only need to ask yourself one question, "Do you want to continue fighting with your computer or do you want to use it?"

BTW, this has also sped up my computer unbelievably.

@ntennies: I can reply to your questions to edoardopa and @Barry Fass-Holmes

I had both Logic Board and Display replaced on my mid-2012 Retina MacBook under the Apple Care Plan.

I took the Mac home and after an hour of very light usage (just Safari and Firefox running together) I had a freeze when attempting to watch a video on Yahoo.

1) Yes, the freeze and the GPU Reset occurred at exact the same time.

2) Yes, you can see in the attached picture multiple GPU resets, 20 restarts in 19 seconds, about 1 attempt per second

3) The delta time is 1 second circa.

All the attempts were for the Intel HD Graphics 4000

Hope this help.

Here below you can read the detail from console of the 1st GPU Reset listed in the picture

Mon Feb 23 18:43:15 2015

Event: GPU Reset

Data/Time: Mon Feb 23 18:43:15 2015

Application:

Path:

OS Version: Mac OS X Version 10.10.2 (Build 14C109)

Graphics Hardware: Intel HD Graphics 4000

Signature: 801

Report Data:

===========================================================================

Intel GPU Hang Summary

version 1.0.0

===========================================================================

-- Summary of Status for ALL Rings --

MAIN Ring is idle

MEDIA Ring is NOT enabled

BLT Ring is NOT enabled

===========================================================================

Essential Ring Registers

===========================================================================

-- MAIN Ring --

RING_HEAD = 0x00000048 (offset = 0x00000048, wrap count = 0, wait_for_cond = 0x0)

RING_TAIL = 0x00000048 (offset = 0x00000048)

RING_START = 0x10204000 (offset = 0x10204000)

RING_CTRL = 0x0000f001 (enabled, auto report disabled (0x00), not waiting, semaphore not waiting, length = 0x010 4KB pages)

-- MEDIA Ring --

RING_HEAD = 0x00000000 (offset = 0x00000000, wrap count = 0, wait_for_cond = 0x0)

RING_TAIL = 0x00000000 (offset = 0x00000000)

RING_START = 0x00000000 (offset = 0x00000000)

RING_CTRL = 0x00000000 (disabled, auto report disabled (0x00), not waiting, semaphore not waiting, length = 0x001 4KB pages)

-- BLT Ring --

RING_HEAD = 0x00000000 (offset = 0x00000000, wrap count = 0, wait_for_cond = 0x0)

RING_TAIL = 0x00000000 (offset = 0x00000000)

RING_START = 0x00000000 (offset = 0x00000000)

RING_CTRL = 0x00000000 (disabled, auto report disabled (0x00), not waiting, semaphore not waiting, length = 0x001 4KB pages)

- Common Slice INST_DONE register -

All Common Slice engines are done.

- INSTDONE_CS (=0xffffffff): Not done =

===========================================================================

MMIO Register Dump

===========================================================================