Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Mac Pro (Late 2013) GPU (Driver) Issues
- Thread starter bax2003
- Start date
- Sort by reaction score
D300 system in for repair at apple, they found fault with the cards and want to replace it with like for like at a cost. Really anoyyed i have to pay for this. How likely is it the fault will resurface? They will only replace for the same card and won't upgrade the card to D500 or D700. Seems a bit sus they don't sell them with these cards anymore, makes me think its old stock and the fault will re-appear
My Mac Pro 2013 has the exactly same frozen issue in this thread.
I like to know if I can disable one of the dual D300 card, or just unplug the fault card from machine, if only one card is fault. In my usage case, it always use only one graphics card. (I bought the Mac Pro for the reason of four TB interface)
Anyone know this?
Or can I disable some fancy function in graphics card if I do not need it? Hopefully it can work stably in this downgraded way.
I like to know if I can disable one of the dual D300 card, or just unplug the fault card from machine, if only one card is fault. In my usage case, it always use only one graphics card. (I bought the Mac Pro for the reason of four TB interface)
Anyone know this?
Or can I disable some fancy function in graphics card if I do not need it? Hopefully it can work stably in this downgraded way.
Last edited:
You display output is connected to the CARD B + PCIe Flash connector, card A i for computing only. If you disconnect card B - no video or boot, if you disconnect card A - Mac will boot with fan on full blast with GFX in some kind of safe non-accelerated mode.
Last edited:
Thanks for the info.
I read your signature below the msg, your MP is w/ D500s card. But seems that only D300s is impacted by this issue. Did your D500s have this issue? or you bought two new D500s cards or new MP to replace the faulty one?
like to know if it is possible to replace the D300s cards with any fine working cards, D500, D700. maybe replace only the card which connects the display output is enough, right? Wont spend much on this 4years old model, but it is really annoy for this moment, and need fixed.
In few pages before I explained in details what happened with my nMP with D300s and how I ended up with D500s 6c model:
https://forums.macrumors.com/thread...u-driver-issues.1860297/page-46#post-24816624
Most of the reported problems are related to models with D300s. One of the reasons is probably because this is the most sold model. D300s also use a bit older chips than in D500s and D700s. You also should be aware of the fact that there are reported problems with every model.
Changing the chips could work but only if you change both cards. If I had at my disposal two nMPs, one with D300s and the other with D500s I would test this theory this very second. It could also be the case that every model´s logic board expects certain Dxx model to be plugged in.
I think that acceleration could be disabled via XCode Developer Tool named Quartz Debug and its settings: (without typing terminal commands)
- Deselected Enable Quartz Extreme
- Checked the Disable 2D Acceleration
- Checked the Disable radio button under Quartz GL
cannot agree more.
I have disabled the hardware acceleration in google Chrome browser. but it is only for Chrome.
For the safari and some other applications which depends on safari engine, I still do not know how to disable the hardware acceleration on it.
In my experience, the freeze issue likely happens on chrome/safari or other webpage app, and the waking up from sleep, which I now totally not make nMP into sleep mode.
I am starting to suspect the same with my system - had posted previously about beach balling and lockups.
D300 running High Sierra. Reinstalled from scratch. Temps all good.
Running internal GPU through Thunderbolt Display, maxed out CPU w/“yes” (99% load), locks up within minutes.
Running internal GPU in safe mode, seems to work ok without locking up with same CPU stress.
Running RX 580 eGPU via Akitio Node w/TB2 converter, NOT in safe mode through ultra wide LG display, no lock ups so far, even if I run Safari session with same CPU stress. Over 30 minutes now.
Makes me tentatively think it is a driver and/or hardware issue with the D300 graphics cards.
Will let run for a few hours and report back. Next step will be to restore the drive and test again with eGPU.
D300 running High Sierra. Reinstalled from scratch. Temps all good.
Running internal GPU through Thunderbolt Display, maxed out CPU w/“yes” (99% load), locks up within minutes.
Running internal GPU in safe mode, seems to work ok without locking up with same CPU stress.
Running RX 580 eGPU via Akitio Node w/TB2 converter, NOT in safe mode through ultra wide LG display, no lock ups so far, even if I run Safari session with same CPU stress. Over 30 minutes now.
Makes me tentatively think it is a driver and/or hardware issue with the D300 graphics cards.
Will let run for a few hours and report back. Next step will be to restore the drive and test again with eGPU.
What is the internal GPU safe mode?
My D300 MP is connected with four DELL U2515h monitors, works on some java data applications. Totally no GPU acceleration is needed. like to know how to make it in safe mode.
Standard Mac safe mode - boot holding the shift key down. Suspect it loads a minimal graphics driver without acceleration.
Then it should be possible to delete all but the minimal graphics driver via the terminal (or after you boot up via safe mode or via an external drive). Does anyone know how?
Don’t know why you would want to do this - the barebones driver is really minimal, with terrible screen response.
Just find a way to disable gpu acceleration (OpenCL) on mac.
https://www.phaseone.com/zh-CN/Search/Article.aspx?articleid=1720
In this page, there is a section, posted as below.
How to completely disable OpenCL, in case of a crash at startup
• Windows: run the script DONOT_useopencl.reg and restart Capture One (the download also contains a restore script to re-enable OpenCL useopencl.reg) Download here
• Mac: Run this script “ToggleOpenCL.app” - disable OpenCL and restart Capture One Download here
• Mac 10.11 UPDATE: Run this script instead “ToggleOpenCLMacOS10.11.app” - disable OpenCL and restart Capture One Download here
Since my MP is running on 10.13, not sure which script should be used. I tried both scripts for mac.
No freezing issue for 30 hours after I toggle off OpenCL with the script.
You guys may have a try.
[doublepost=1507990156][/doublepost]BTW, OpenGL is another GPU graphics interface which Mac OS is used in core level. OpenGL could not be disabled in Mac.
https://www.phaseone.com/zh-CN/Search/Article.aspx?articleid=1720
In this page, there is a section, posted as below.
How to completely disable OpenCL, in case of a crash at startup
• Windows: run the script DONOT_useopencl.reg and restart Capture One (the download also contains a restore script to re-enable OpenCL useopencl.reg) Download here
• Mac: Run this script “ToggleOpenCL.app” - disable OpenCL and restart Capture One Download here
• Mac 10.11 UPDATE: Run this script instead “ToggleOpenCLMacOS10.11.app” - disable OpenCL and restart Capture One Download here
Since my MP is running on 10.13, not sure which script should be used. I tried both scripts for mac.
No freezing issue for 30 hours after I toggle off OpenCL with the script.
You guys may have a try.
[doublepost=1507990156][/doublepost]BTW, OpenGL is another GPU graphics interface which Mac OS is used in core level. OpenGL could not be disabled in Mac.
Last edited:
From my research into this over the past 18 months, the general opinion from people who've had the issue resolved on the quad core D300 model is that the circular interconnect board needs to be replaced. The problem is that it isn't listed in their repair guidelines to replace this part, so thy usually won't order it in for you or agree to replace it until they've replaced the GPU's several times first.
Maybe you could try mentioning to them you know there's an extended repair program for the other 2 models, and you're baffled as to why there are so many cases of the exact same parts failing on D300 computers yet that model isn't covered under the extension.
Do you know when your computer was manufactured?
Last edited:
iigsie,
You're saying my card isn't bad but just the interconnect board? Do you have any links to support this claim?
You're saying my card isn't bad but just the interconnect board? Do you have any links to support this claim?
From my experience I just paid £650 to have my nMP repaired at an Apple store.
I have the standard D300 spec nMP. It exhibited all the symptoms of the recall on the D500 and D700 cards. It progressively got worse until it wouldn't start up. Lights appeared around the usb and thunderbolt ports but it wouldn't start up when pressing the power button.
They changed the power supply and the interconnect board and graphic card B.
I will test it out tomorrow and see if I get any GPU restarts
Summary: if you get WindowServer lockups on High Sierra - it would be worth trying a downgrade to Sierra ... IS INCORRECT - see correction below.
I was getting the crashes described at the top of the thread on my late 2013 Mac Pro with the D300 graphics card, on El Capitan. About once a day on average.
I upgraded to High Sierra, and continued to get WindowServer crashes, but with a different error message. As before, the WindowServer would lock, with the SBOD. I could log in from another machine. The output of "log show" found that, at the time that WindowServer locked, I would get these messages, repeating until reboot:
WindowServer: (CoreDisplay) [ERROR] - Fallback GPU Display Pipe transaction submission failed
There was no obvious pattern to the lock-up - I replicated it on a fresh install with Apple standard software; it occurred using Safari and Virtualbox, I suppose coincidentally. It would only occur if I was using the GUI.
Apple claimed to have fixed this with a re-install of High Sierra, but I suspect they just didn't wait long enough for the crash to recur. However, they did say that Apple higher-up support said that there might be an incompatibility of the GPUs with High Sierra, and recommended downgrading to Sierra if the problem recurred. I asked them to do that downgrade, and since they did, I have not had the problem (now 4 days).
CORRECTION - STILL CRASHES UNDER SIERRA
It crashed again after 6 days with the same symptoms. Result of `log show` revealed new error messages at the time of the freeze:
2017-11-06 16:03:22.914095+0000 0x460 Fault 0x0 0 kernel: (IOAcceleratorFamily2) virtual IOReturn IOAccelEventMachine2::waitForStamp(int32_t, stamp_t, stamp_t *): initial wait for 1 second expired. Continue wait for 4 seconds. stamp 44765 (gpu_stamp=44764)
2017-11-06 16:03:24.034437+0000 0x70d18 Default 0x0 265 CommCenter: [com.apple.CommCenter.default] #I CSIAppInfo.ApplicationActivationObserver: handleLSNotitifcation_sync: Application exited: <private>
2017-11-06 16:03:24.198290+0000 0x73607 Default 0x0 2079 diskimages-helper: (Foundation) *** -[NSMachPort handlePortMessage:]: dropping incoming DO message because the connection is invalid
2017-11-06 16:03:24.369411+0000 0xdc5 Default 0x0 276 accountsd: (AccountsDaemon) [com.apple.accounts.core] "Idle-exit-preventing transaction ended!"
2017-11-06 16:03:26.915089+0000 0x460 Fault 0x0 0 kernel: (IOAcceleratorFamily2) virtual IOReturn IOAccelEventMachine2::waitForStamp(int32_t, stamp_t, stamp_t *): timeout waiting for stamp 44765 (gpu_stamp=44764)
2017-11-06 16:03:27.034800+0000 0x5c5 Fault 0x0 0 kernel: (IOAcceleratorFamily2) virtual void IOAccelEventMachineFast2::checkGPUProgress() - Signaling hardware error on channel 0..
2017-11-06 16:03:27.034822+0000 0x5c5 Fault 0x0 0 kernel: (IOAcceleratorFamily2) void IOAccelEventMachine2::signalHardwareError(eRestartRequest, int32_t): GPURestartSignaled stampIdx=0 type=2 prevType=0 numStamps=9
This time, when I contacted Apple support, they forwarded me to an advanced support person, who told me to bring it back to the Apple store. I wrote up the problems I was having in great detail and left this report on the machine - see : https://gist.github.com/c93717cac11cfa5f8cae9ddc662c612e
Apple said they were testing the memory and the GPUs. They told me that my write-up had been very helpful in diagnosing the problem. Opening Safari when I got the machine back suggested they were using very long videos playing (or being recorded) on YouTube, to replicate the problem. They ended up replacing both the GPUs with new AMD FirePro D300s. I'll report back if the crash recurs.
Cheers,
Matthew
I was getting the crashes described at the top of the thread on my late 2013 Mac Pro with the D300 graphics card, on El Capitan. About once a day on average.
I upgraded to High Sierra, and continued to get WindowServer crashes, but with a different error message. As before, the WindowServer would lock, with the SBOD. I could log in from another machine. The output of "log show" found that, at the time that WindowServer locked, I would get these messages, repeating until reboot:
WindowServer: (CoreDisplay) [ERROR] - Fallback GPU Display Pipe transaction submission failed
There was no obvious pattern to the lock-up - I replicated it on a fresh install with Apple standard software; it occurred using Safari and Virtualbox, I suppose coincidentally. It would only occur if I was using the GUI.
Apple claimed to have fixed this with a re-install of High Sierra, but I suspect they just didn't wait long enough for the crash to recur. However, they did say that Apple higher-up support said that there might be an incompatibility of the GPUs with High Sierra, and recommended downgrading to Sierra if the problem recurred. I asked them to do that downgrade, and since they did, I have not had the problem (now 4 days).
CORRECTION - STILL CRASHES UNDER SIERRA
It crashed again after 6 days with the same symptoms. Result of `log show` revealed new error messages at the time of the freeze:
2017-11-06 16:03:22.914095+0000 0x460 Fault 0x0 0 kernel: (IOAcceleratorFamily2) virtual IOReturn IOAccelEventMachine2::waitForStamp(int32_t, stamp_t, stamp_t *): initial wait for 1 second expired. Continue wait for 4 seconds. stamp 44765 (gpu_stamp=44764)
2017-11-06 16:03:24.034437+0000 0x70d18 Default 0x0 265 CommCenter: [com.apple.CommCenter.default] #I CSIAppInfo.ApplicationActivationObserver: handleLSNotitifcation_sync: Application exited: <private>
2017-11-06 16:03:24.198290+0000 0x73607 Default 0x0 2079 diskimages-helper: (Foundation) *** -[NSMachPort handlePortMessage:]: dropping incoming DO message because the connection is invalid
2017-11-06 16:03:24.369411+0000 0xdc5 Default 0x0 276 accountsd: (AccountsDaemon) [com.apple.accounts.core] "Idle-exit-preventing transaction ended!"
2017-11-06 16:03:26.915089+0000 0x460 Fault 0x0 0 kernel: (IOAcceleratorFamily2) virtual IOReturn IOAccelEventMachine2::waitForStamp(int32_t, stamp_t, stamp_t *): timeout waiting for stamp 44765 (gpu_stamp=44764)
2017-11-06 16:03:27.034800+0000 0x5c5 Fault 0x0 0 kernel: (IOAcceleratorFamily2) virtual void IOAccelEventMachineFast2::checkGPUProgress() - Signaling hardware error on channel 0..
2017-11-06 16:03:27.034822+0000 0x5c5 Fault 0x0 0 kernel: (IOAcceleratorFamily2) void IOAccelEventMachine2::signalHardwareError(eRestartRequest, int32_t): GPURestartSignaled stampIdx=0 type=2 prevType=0 numStamps=9
This time, when I contacted Apple support, they forwarded me to an advanced support person, who told me to bring it back to the Apple store. I wrote up the problems I was having in great detail and left this report on the machine - see : https://gist.github.com/c93717cac11cfa5f8cae9ddc662c612e
Apple said they were testing the memory and the GPUs. They told me that my write-up had been very helpful in diagnosing the problem. Opening Safari when I got the machine back suggested they were using very long videos playing (or being recorded) on YouTube, to replicate the problem. They ended up replacing both the GPUs with new AMD FirePro D300s. I'll report back if the crash recurs.
Cheers,
Matthew
Last edited:
Sure, but I hope that it is a good sign that somebody at Apple seems to know that High Sierra is crashing the GPU, and therefore, that there might be an OS or hardware fix available, at some point.
No, not all D300s are affected, it is totally random. Again, the most sold model is QC D300 so it is very hard to talk about percentages but the fact is that is too much of those nMPs with the same problems.