Extreme Instability w/GTX1080 & Nvidia Drivers

[Derived]

Posted a thread on here a while back with similar issues, but they seemed to go away.

Now they're back, and worse than ever. Running my GTX1080 on 10.12.6 w/latest drivers, system essentially becomes unusable if left alone long enough to trigger the screen saver 8/10 times at least. Ironically if I leave the system alone overnight I never have this problem.

What happens is that the computer becomes almost completely non-responsive, all input locks up every 5-15 seconds for 2-5 seconds at a time. Any UI elements freeze in place, sometimes even the audio output stream cuts out until it comes back. This will happen constantly until rebooted. It's happening right now as I type this, every 3rd or 4th word it's locking up completely.

Took a look in the console log, and I found the kernel sending a "Sandbox violation: (DumpGPURestart (and some numbers here that vary over time)) deny (1) mach-lookup com.apple.lsd.modifydb" followed immediately by the DumpGPURestart process sending the message "LaunchServices: disconnect event received for service <private>"

This is happening hundreds of times every second. As I've typed this it's accumulated a few hundred thousand of each.

Every once in a while the kernel also sends the message "NVDA: failed to send buffer to GPU!" and "NVDA: Calling gpusKillClient for task <private>

In fact it even seems like it's writing messages for excessive log files now and clears them out a few thousand at a time but they keep coming back. Seems like the system isn't able to talk to the GPU properly, essentially.

I can't go on rebooting every time it does this, that's several times a day sometimes. Anyone have ideas about what I might be able to do to clean this up? I'm guessing reinstalling the driver is a first step, although this behavior has persisted across a few driver versions and at least 2 Mac OS updates. And, it stuck with it through a wipe and reinstall from Time Machine. Wondering if it's a preference file somewhere that needs to be deleted.

At this point I'm ready to give up and swap for a Vega...I don't need CUDA that bad....

EDIT: Seems if I manually trigger the screensaver, nothing happens. But...if I let it naturally sit long enough to trigger the screen saver...the screen saver never comes up, the screen just turns black and then the problems start. Very strange.

 

[Asgorath]

Posted a thread on here a while back with similar issues, but they seemed to go away.

Now they're back, and worse than ever. Running my GTX1080 on 10.12.6 w/latest drivers, system essentially becomes unusable if left alone long enough to trigger the screen saver 8/10 times at least. Ironically if I leave the system alone overnight I never have this problem.

What happens is that the computer becomes almost completely non-responsive, all input locks up every 5-15 seconds for 2-5 seconds at a time. Any UI elements freeze in place, sometimes even the audio output stream cuts out until it comes back. This will happen constantly until rebooted. It's happening right now as I type this, every 3rd or 4th word it's locking up completely.

Took a look in the console log, and I found the kernel sending a "Sandbox violation: (DumpGPURestart (and some numbers here that vary over time)) deny (1) mach-lookup com.apple.lsd.modifydb" followed immediately by the DumpGPURestart process sending the message "LaunchServices: disconnect event received for service <private>"

This is happening hundreds of times every second. As I've typed this it's accumulated a few hundred thousand of each.

Every once in a while the kernel also sends the message "NVDA: failed to send buffer to GPU!" and "NVDA: Calling gpusKillClient for task <private>

In fact it even seems like it's writing messages for excessive log files now and clears them out a few thousand at a time but they keep coming back. Seems like the system isn't able to talk to the GPU properly, essentially.

I can't go on rebooting every time it does this, that's several times a day sometimes. Anyone have ideas about what I might be able to do to clean this up? I'm guessing reinstalling the driver is a first step, although this behavior has persisted across a few driver versions and at least 2 Mac OS updates. And, it stuck with it through a wipe and reinstall from Time Machine. Wondering if it's a preference file somewhere that needs to be deleted.

At this point I'm ready to give up and swap for a Vega...I don't need CUDA that bad....

EDIT: Seems if I manually trigger the screensaver, nothing happens. But...if I let it naturally sit long enough to trigger the screen saver...the screen saver never comes up, the screen just turns black and then the problems start. Very strange.

What system are you running? Is this a Mac Pro? Sounds like you might have a bad board, or some other kind of hardware issue. Maybe try removing and re-installing the card to ensure it's plugged in properly, and then check the power cables. You should also try updating to 10.13 as the Pascal driver seem to be much more stable in that OS. FWIW I'm using a GTX 1080 and haven't had any problems at all.

 

[Derived]

10.12.6...it is indeed a Mac Pro, hence why I asked here. It's a 4,1 flashed to a 5,1. 2x 3.46 6 core, 128 gb, 1 tb PCIE SSD.

The issue happened quite often earlier in the year, then didn't really happen at all after a new version of OS X and a new version of the driver. Then, lately, it's started up again. I doubt it's the board or card, the machine runs fine otherwise. Can run furmark with no issues. It only seems to be triggered by letting the system attempt to go to screen saver.

I will probably end up upgrading soon, just haven't had the time to deal with it lately.

 

[Asgorath]

10.12.6...it is indeed a Mac Pro, hence why I asked here. It's a 4,1 flashed to a 5,1. 2x 3.46 6 core, 128 gb, 1 tb PCIE SSD.

The issue happened quite often earlier in the year, then didn't really happen at all after a new version of OS X and a new version of the driver. Then, lately, it's started up again. I doubt it's the board or card, the machine runs fine otherwise. Can run furmark with no issues. It only seems to be triggered by letting the system attempt to go to screen saver.

I will probably end up upgrading soon, just haven't had the time to deal with it lately.

There are plenty of people who use Hackintoshes (myself included) who post here, just wanted to be sure. GPU restarts and submission failures do often indicate a hardware-level problem, so it's always worth double-checking the basic stuff.

What screen saver are you using, out of curiosity?

 

[Derived]

Of course, wasn't thinking about that I suppose. For some reason I had thought I mentioned the machine in the first post but I may have forgotten due to the trouble typing.

I'm using a standard built-in screensaver, "floating" - the fact that it doesn't happen every time is also strange. I've spoken with MacVidCards and he also said something must be very wrong, but hard to say what it might be. I'm thinking it has to be the driver install.

The times it doesn't go completely off the rails coming back from the screensaver, it's still quite sluggish for a minute or two. Although as far as I can tell this doesn't happen if I use a hot corner to manually trigger it.

I should try my 680 with both Mac and Nvidia drivers to test it out some more, just don't have the time to do all the swapping. Might have the chance this weekend, will definitely update with that.

 

[Squuiid]

The times it doesn't go completely off the rails coming back from the screensaver, it's still quite sluggish for a minute or two. Although as far as I can tell this doesn't happen if I use a hot corner to manually trigger it.

And this is why I sold my GTX 1070 FE on eBay and put the GTX 680 back in. As soon as I did this issue, and others, went away.

Pascal is not usable in macOS in my opinion. Sure it 'sorta' works, but it's buggy as hell.

I have a pre-order in for a Sapphire RX 580 Pulse 8GB and feedback for this card has been very good.

I do have a GTX 1080 FE in my other cMP but run Windows 10 on it exclusively. The 1080 runs very nicely in Windows.

 

[Asgorath]

And this is why I sold my GTX 1070 FE on eBay and put the GTX 680 back in. As soon as I did this issue, and others, went away.

Pascal is not usable in macOS in my opinion. Sure it 'sorta' works, but it's buggy as hell.

I have a pre-order in for a Sapphire RX 580 Pulse 8GB and feedback for this card has been very good.

I do have a GTX 1080 FE in my other cMP but run Windows 10 on it exclusively. The 1080 runs very nicely in Windows.

FWIW the Pascal drivers appear to be much better in High Sierra.

 

[William Payne]

What happens if you disable the screensaver so that if never goes to it?

 

[Asgorath]

What happens if you disable the screensaver so that if never goes to it?

Alternatively, what if you use a different screen saver? Is the problem specific to that one, or more generic?

 

[bsbeamer]

Do you happen to have any update on this?
What version of CUDA and NVIDIA web drivers are you using?
What OS update?

Considering upgrading to a GTX 1080 FE before end of this year and wondering if this is still persisting, or if there are other major issues I should be aware of.

Was running into a similar issue when I would forget to disable a screensaver during long CUDA-based video renders with my GTX 680 card (official Mac version). The renders would suddenly freeze or the system would lockup if the screensaver started. If disabling the screensaver, everything was perfectly fine. Sort of think it had something to do with the displayport monitor (Apple Cinema 27") sending a signal to the card/system or drastically changing the power load, but I've been unable to confirm this. Disabled the screensaver permanently. Only enabled through a hot corner action. Monitor now auto turns off after XX minutes and greatly fixed this issue.

 

[h9826790]

Do you happen to have any update on this?
What version of CUDA and NVIDIA web drivers are you using?
What OS update?

Considering upgrading to a GTX 1080 FE before end of this year and wondering if this is still persisting, or if there are other major issues I should be aware of.

Was running into a similar issue when I would forget to disable a screensaver during long CUDA-based video renders with my GTX 680 card (official Mac version). The renders would suddenly freeze or the system would lockup if the screensaver started. If disabling the screensaver, everything was perfectly fine. Sort of think it had something to do with the displayport monitor (Apple Cinema 27") sending a signal to the card/system or drastically changing the power load, but I've been unable to confirm this. Disabled the screensaver permanently. Only enabled through a hot corner action. Monitor now auto turns off after XX minutes and greatly fixed this issue.

Don't know about Sierra, but with High Sierra, my 1080Ti (reference PCB) work flawlessly.

 

[ticotoo]

Posted a thread on here a while back with similar issues, but they seemed to go away.

Now they're back, and worse than ever. Running my GTX1080 on 10.12.6 w/latest drivers, system essentially becomes unusable if left alone long enough to trigger the screen saver 8/10 times at least. Ironically if I leave the system alone overnight I never have this problem.

What happens is that the computer becomes almost completely non-responsive, all input locks up every 5-15 seconds for 2-5 seconds at a time. Any UI elements freeze in place, sometimes even the audio output stream cuts out until it comes back. This will happen constantly until rebooted. It's happening right now as I type this, every 3rd or 4th word it's locking up completely.

Took a look in the console log, and I found the kernel sending a "Sandbox violation: (DumpGPURestart (and some numbers here that vary over time)) deny (1) mach-lookup com.apple.lsd.modifydb" followed immediately by the DumpGPURestart process sending the message "LaunchServices: disconnect event received for service <private>"

This is happening hundreds of times every second. As I've typed this it's accumulated a few hundred thousand of each.

Every once in a while the kernel also sends the message "NVDA: failed to send buffer to GPU!" and "NVDA: Calling gpusKillClient for task <private>

In fact it even seems like it's writing messages for excessive log files now and clears them out a few thousand at a time but they keep coming back. Seems like the system isn't able to talk to the GPU properly, essentially.

I can't go on rebooting every time it does this, that's several times a day sometimes. Anyone have ideas about what I might be able to do to clean this up? I'm guessing reinstalling the driver is a first step, although this behavior has persisted across a few driver versions and at least 2 Mac OS updates. And, it stuck with it through a wipe and reinstall from Time Machine. Wondering if it's a preference file somewhere that needs to be deleted.

At this point I'm ready to give up and swap for a Vega...I don't need CUDA that bad....

EDIT: Seems if I manually trigger the screensaver, nothing happens. But...if I let it naturally sit long enough to trigger the screen saver...the screen saver never comes up, the screen just turns black and then the problems start. Very strange.

[doublepost=1513863983][/doublepost]Its not the NVIDIA drivers. I have been running a gtx1080 in my 2010 MacPro and various versions of CUDA & NVIDIA drivers and Sierra for about 7 months without any of the problems you are talking about, other not having support for the boot screens. Does it happen with your original graphics card ? How are you powering your gtx1080? I would do a recovery boot reinstall & see if it still happens or boot from a drive with a clean install of Sierra or High Sierra to see what happens. You many be getting a kext caused crash. If its still happening, then I would suspect a hardware/power problem.

 

[bsbeamer]

Anyone with a GTX1080 notice a bit of "knock" noise every so often? MacPro 5,1 is basically silent with mainly SSDs and can hear the GTX1080 running much more than with the previous GTX680 (official mac version).

Using a dual 6-pin to 8-pin power adapter at the moment:
https://www.ebay.com/itm/NVIDIA-TIT...RAPHIC-CARD-POWER-CABLE-ORIGINAL/202091478316

Ordered one of these dual 6-pin to 8-pin (without adapter) and to see if it makes any difference:
https://www.moddiy.com/products/App...tandard-PCI%2dE-8%2dPin-Video-Card-Cable.html

 

[Dr. Stealth]

Another Happy User of dual 1080 TI's and the Web Drivers in High Sierra....... It works Very Nice. I think Nvidia has left Sierra in the past, their efforts are geared towards the High Sierra drivers now.