MP 1,1-5,1 Help please - upgraded 4, 1 crashing to login screen - GPU related?

[vel0city]

Hi all.

Long time lurker, first time poster.

Earlier this year I had my 4, 1 upgraded with new CPU (2 x 3.46 GHz 6-Core) and GPU (GTX 1080 Ti)

Running 10.12.6, 64GB Ram, two Apple Cinema Displays, 30" and 23".

Since the upgrade I've had random, intermittent crashes which dump me back to the login screen. This could be triggered by anything - double clicking an image in iMessage, hitting spacebar to Quicklook in Finder, rendering an image in Cinema 4D, or simply by watching a Youtube video and not clicking on anything.

The nVidia drivers and Window Server are always mentioned in the crash logs. My nVidia web driver and CUDA are up to date.

Does anyone have any diagnostic ideas or solutions please? Here's the latest from the crash logs.

Thanks so much for any pointers. Any advice is truly appreciated.

Process: WindowServer [257]
Path: /System/Library/PrivateFrameworks/SkyLight.framework/Versions/A/Resources/WindowServer
Identifier: WindowServer
Version: 600.00 (15)
Code Type: X86-64 (Native)
Parent Process: launchd [1]
Responsible: WindowServer [257]
User ID: 88

Thread 0 Crashed:: Dispatch queue: com.apple.main-thread
0 com.nvidia.web.GeForceGLDriverWeb 0x00000001779f4a0e 0x177484000 + 5704206
1 com.nvidia.web.GeForceGLDriverWeb 0x00000001779f4760 0x177484000 + 5703520
2 com.nvidia.web.GeForceGLDriverWeb 0x00000001777e099a 0x177484000 + 3525018
3 com.nvidia.web.GeForceGLDriverWeb 0x00000001777fcc56 0x177484000 + 3640406
4 libGPUSupportMercury.dylib 0x00007fff9ab5fcbd gldLoadFramebuffer + 226
5 com.nvidia.web.GeForceGLDriverWeb 0x00000001777fed6a gldLoadFramebuffer + 229
6 GLEngine 0x00007fff91d9dbeb gleCheckFramebufferStatus + 103
7 GLEngine 0x00007fff91cdb1ec glCheckFramebufferStatusEXT_Exec + 153
8 com.apple.SkyLight 0x00007fff9eb564dc WSGLFramebufferObjectCheckStatus + 42
9 com.apple.SkyLight 0x00007fff9eadce93 CaptureSurfaceGL:opulate(WSCompositeSourceLayer*) + 171
10 com.apple.SkyLight 0x00007fff9e9cafd9 WSFlattenedRepresentation::Update() + 367
11 com.apple.SkyLight 0x00007fff9ea3485b WSCALayerBackingUpdateFlattening + 181
12 com.apple.SkyLight 0x00007fff9eb542e3 CGXNextSurface + 3421
13 com.apple.SkyLight 0x00007fff9ea11f88 generate_layers_for_window_surfaces(CGXRedrawState*, CGXWindow*, CGSOrderOp, unsigned int, int*, WSCompositeSourceLayer**, CGSRegionObject*) + 5795
14 com.apple.SkyLight 0x00007fff9ea12e9b generate_layers_for_window(CGXRedrawState*, CGXWindow*) + 3104
15 com.apple.SkyLight 0x00007fff9ea0f584 CGXUpdateDisplay + 15282
16 com.apple.SkyLight 0x00007fff9ea0b787 update_display_callback(void*, double) + 216
17 com.apple.SkyLight 0x00007fff9ea733cd run_timer_pass + 706
18 com.apple.SkyLight 0x00007fff9eae5c7e CGXRunOneServicesPass + 152
19 com.apple.SkyLight 0x00007fff9eae7444 SLXServer + 3853
20 WindowServer 0x000000010cfc0dde 0x10cfc0000 + 3550
21 libdyld.dylib 0x00007fffa269e235 start + 1

 

[bsbeamer]

Why have you not updated to 10.13.X? It fixed several things, including a lot of issues, firmware updates, allows installation of newer NVIDIA Web Drivers, also allows installation of newer NVIDIA CUDA drivers. Not saying it would magically fix everything, but may be a starting point.

 

[bookemdano]

Do you have another GPU installed along with the 1080? If so, that is likely the cause.

 

[vel0city]

Do you have another GPU installed along with the 1080? If so, that is likely the cause.

Yes I do have the stock GT 120 still installed. The company who did the upgrade (Create Pro) advised me to keep it installed for OS updates and the boot screen. Could you expand on why you think it might be a problem please? Thanks.

@bsbeamer - thanks for the suggestion. I did think my CUDA drivers were up to date, didn't realise that newer versions were available.

 

[bsbeamer]

Remove the GT120 from the system and see if you continue to have issues. You do not need boot screens to troubleshoot this issue. Dual GPUs that are not nearly identical often can cause a driver conflict in the OS, especially if you're utilizing the GT120 for display output. The GTX 1080 Ti can easily drive those two displays without the need for a 2nd GPU (a common misperception). Also make sure FileVault is disabled or you won't be able to boot properly.

Once GT120 is removed try only connecting one monitor temporarily, ideally via DP (or DP>mDP) to see if issues persist. Avoid signal converters if possible - they often introduce additional issues that need to be separately addressed. (DP>mDP is an adapter, not converter.)

Latest CUDA updates only work properly when using the latest NVIDIA Web Drivers. Latest CUDA is currently version 396.148 and only works properly with 10.13:
https://www.nvidia.com/object/macosx-cuda-396.148-driver.html

 

[vel0city]

Thank you bsbeamer, seriously appreciated.

 

[bsbeamer]

Worth asking - how are you powering the GTX 1080 Ti? Most models I've seen are 8 pin + 6 pin, which is not a "standard" power cable setup. There are ways to get around this (and several people have successfully), but it's not as simple as the dual mini 6-pin to standard 8-pin that many GPUs use.

 

[vel0city]

Have to say I'm not sure, it was installed by this company: https://create.pro/

You send them your 4,1 or 5,1 and they return it to you with your requested upgrades.

As far as I'm aware, they disconnected the CD ROM (doesn't work any more) and the GPU takes its power from there.

 

[bsbeamer]

If removing the GT120 from the system does not fix your issues, suggest you contact them. With the money you've paid, they should guarantee their work and offer support. Hopefully they also made your machine ID as a 5,1...

 

[vel0city]

Yes, it IDs as a 5,1. Their support blamed Nvidia drivers, not much help.

 

[bookemdano]

The dual GPU thing is the problem. There have been several other threads over the past few months with people exhibiting the exact same symptom (crashing to the login screen). The GT120 seems to run OK in conjunction with an unflashed ATI card (like the RX 580), but two NVIDIA cards (one with an EFI, one without) has been problematic. It's too bad the firm you bought the upgrade from wasn't aware of this issue. I don't think there is a solution at present other than removing the GT120 from the machine.

 

[vel0city]

Thank you @bookemdano, as soon as I'm done working for the day (the Mac in question is my main work machine) I'll pull the GT120.

I really appreciate all the replies, thanks a lot.

 

[h9826790]

Don't expect too much from Create Pro.

https://forums.macrumors.com/thread...-over-mac-gpu-development-create-pro.1813249/

 

[vel0city]

Yeah, have to say I'm disappointed in Create Pro's aftercare. I never had a single crash on my 4,1 before they got their hands on it. It was absolutely rock solid stable. As soon as I got it back from them I invested in Redshift and Octane and got tied into CUDA dependancy quickly for my client work. When the Mac started to crash it was too late to go back.

Create Pro told me that most/all of their customers with their upgraded Nvidia GPUs had been affected by the crash to login issue and that all I could do was hope for a fix either from Apple or Nvidia. Of course, they continued offering their upgrade service despite knowing that their systems were prone to instability. This is a company dedicated to providing high-spec custom Mac workstations to the creative industry at a relatively high price, you'd hope their tech support was up to standard. Apparently not. It's just really disappointing.

I've pulled the GT120 now, let's see how it goes. Thanks again everyone.

 

[bsbeamer]

I'm using an authentic 5,1 as my personal machine. Upgraded from a single to dual 3.46 CPU, added 128GB RAM, swapped to 7+ internal SSDs, and usually use a USB3 PCIe card with the system. Currently have an NVIDIA GTX 1080 Founders Edition 8GB card installed that was not flashed for boot screens (standard PC style off the shelf card). Have a Sapphire Pulse RX 580 8GB card on standby for Mojave (also standard PC style off the shelf card). I only use one GPU in this system at a time (currently the NVIDIA GTX 1080).

Currently on 10.13.6 build 17G65 with NVIDIA Web Drivers 387.10.10.15.15.108 (which has been pulled) and CUDA 396.148. I do not run into any of the issues you describe above. Unless at a client's office, I'm on this machine almost daily and it's often left rendering video overnight without issue.

I've had crashes starting several applications at times that resolve after trashing preferences (more common than Adobe wants to admit). I've had crashes at random times when overtaxing the machine with 3D work, occasionally when using multiple stacked layer effects (talking 40+) and forget to manage GPU memory properly with checking/unchecking GPU acceleration box, or after installing an update and a plugin is "bad" that needs to either be updated or reinstalled. Had one 10.13 point update that created issues with APFS on my system drive. Aside from that, this machine has been a dependable beast. The type of thing where as long as you backup your data often, you can sleep well at night.

Not once have I had a crash at login and never a crash related to startup with NVIDIA Web Drivers and this GPU.

 

[vel0city]

Interesting to know, thank you @bsbeamer. Sounds like you're thrashing that machine hard, and it gives me hope that I can resolve this annoying crash to login issue.

 

[bookemdano]

You can resolve it. All you have to do is remove the GT120. That's been the resolution for each of the 5-6 cases we've seen in these forums here in the past few months. It stands to reason why bsbeamer isn't experiencing this issue--the key sentence in his post was

I only use one GPU in this system at a time

Maybe someday NVIDIA will fix their web drivers, but until then the solution is to remove one card. I can pretty much guarantee (based on all previous experience with this bug) that will fix it.

 

[vel0city]

Thank you @bookemdano - I've removed the GT120, will update this thread with any news.

Should I upgrade to 10.13.X?

 

[bsbeamer]

Maybe someday NVIDIA will fix their web drivers, but until then the solution is to remove one card. I can pretty much guarantee (based on all previous experience with this bug) that will fix it.

Just to note...

Have a semi-current client who is using two of the same NVIDIA GPUs (believe 980's) in their Mac Pro 4,1>5,1 without issue as long as both serve a monitor. Some software can take advantage of this (most cannot). They are recognizable in the NVIDIA Web Driver preference pane.

Previous clients had external PCIe expander cases using multiple identical NVIDIA GPUs for CUDA and they did work decently at the time (believe 10.12). The key is identical GPUs. I'm sure this can be bent with same series and same model identifier, but I have no personal experience with that.

Personally find one GPU in a 5,1 is good enough for 95% of users. Unless SLI can get fully enabled on the Mac side, it's doubtful I'll be exploring that much further on my 5,1 in the future.

Should I upgrade to 10.13.X?

You'll get the most current NVIDIA Web Driver and CUDA driver if you do and a bunch of other "improvements" with the update. I find less issues with these newer drivers than the ones in 10.12.X versions. Check your software and if it's compatible, should be no reason to stick with 10.12 vs 10.13. You also can update the MacPro firmware to protect against Spectre/Meltdown and some related stuff.

The 10.14 Mojave beta firmware would enable faster PCI link speed, which may have a minor speed improvement for your work. Expect that firmware (or an updated version) to be included in the Mojave release next week. However, it is not (yet) confirmed NVIDIA will have Mojave compatible Web Drivers.

You would likely have the best experience if you swapped the GTX 1080Ti out of your machine and used just the GT120 to do updates. After everything is updated/installed, shut down, swap the GTX 1080Ti back in (remove GT120).

Personal suggestion: clone your system drive and all data to an external hard drive BEFORE proceeding with any updates. Carbon Copy Cloner works great for this.

 

[bookemdano]

Thank you @bookemdano - I've removed the GT120, will update this thread with any news.

Should I upgrade to 10.13.X?

You can if you want to. High Sierra is quite stable with 10.13.6, and unless you are trying to completely avoid APFS or have software that isn't compatible I see little reason not to upgrade at this point.

Either way though, you should download and run the High Sierra installer from the Mac App Store just to get the MP51.0089.B00 firmware (Mojave has an even newer firmware you can download instead if you want to). This newer firmware patches the CPU microcode to mitigate the terrible Spectre & Meltdown vulnerabilities discovered last year.

The firmware flash is the first step after launching the installer (read the instructions provided on the screen). After the firmware flashes successfully, it will boot you back into Sierra and (assuming the flash was successful) proceed on to let you install High Sierra. If you don't wish to proceed, you can cancel at that point and stay on Sierra. But at least then you would have the new firmware so you're protected against those vulnerabilities.

Edit: Yes, bsbeamer is correct that if you want to do the High Sierra firmware update you would need to keep your GT120 installed. If you'd rather just install the Mojave firmware update (even if you don't want to use Mojave) you can do that with just your 1080 installed. The public Mojave installer will be released next Monday (24th of Sept.).

 

[vel0city]

I have a couple of questions before I do the above please:

I'm always a bit wary of firmware updates - is this a foolproof, risk-free process?

At which point should I install the Nvidia web divers please? Should I install them when I'm booted into High Sierra with the GT120, shut down, swap the cards out, then select the drivers when booted with the 1080 ti installed?

Thanks again.
[doublepost=1537437303][/doublepost]One other thing - I have my secondary display (23" ACD) connected to the GT120. Is it necessary to uninstall the 1080 Ti during the firmware/software updates, or could I just work from the 23" ACD? Thanks guys.

 

[bookemdano]

Have you decided which firmware update to apply? The instructions are different if you're doing the High Sierra MP51.0089.B00 firmware or the Mojave 138.0.0.0.0 firmware.

If you're doing MP51.0089.B00 then there's no need to swap anything (you still have both cards installed, right?). Just go to the Mac App Store, search for High Sierra (or find it on the main page), download it, and run it. The first thing it will do is prompt you to update the firmware. The screen provides specific instructions, but basically you shut down the Mac and hold down the power button until you hear a long beep tone and the power light flashes (your CD tray will also open and then close). After you let go of the button, wait a bit and the Mac will eventually restart itself back into Sierra. Then, go into System Profiler and verify that it says "Boot ROM Version: MP51.0089.B00".

Then you can decide if you actually want to install High Sierra itself. If so, I would shut down and remove the 1080, leave in the GT120 connected to one monitor for the install itself. Once it's installed and running, then install the appropriate web drivers, shut down, take out the GT120 and install the 1080.

If you're doing the 138.0.0.0.0 update from Mojave then you would need to remove the GT120 before running the Mojave installer. Otherwise though the process is the same as above. After you get back to Sierra and check System Profiler, your Boot ROM Version should show 138.0.0.0.0.

At that point you shouldn't proceed to install Mojave though, because web drivers for Mojave are not yet released (and we're still not 100% sure NVIDIA will release them). You could proceed to install High Sierra, but for that you should put the GT120 back in.

Flashing firmware is pretty safe--it's a procedure Apple intends for you to do, so that should make you feel better. Maybe it's not 100% safe, but few things in life are.

 

[Squuiid]

I have a couple of questions before I do the above please:

I'm always a bit wary of firmware updates - is this a foolproof, risk-free process?

At which point should I install the Nvidia web divers please? Should I install them when I'm booted into High Sierra with the GT120, shut down, swap the cards out, then select the drivers when booted with the 1080 ti installed?

Thanks again.
[doublepost=1537437303][/doublepost]One other thing - I have my secondary display (23" ACD) connected to the GT120. Is it necessary to uninstall the 1080 Ti during the firmware/software updates, or could I just work from the 23" ACD? Thanks guys.

Wait until Monday and then download macOS Mojave installer from the store. Double click the installer and it will prompt you to install the new firmware. It will NOT force you to install Mojave, don't worry.
No need to swap out the 1080Ti, the firmware update will install fine. And you really don't have much to worry from this firmware update. I and many, many forum users have been running it for quite a while now and it offers significant benefits. Once upgraded your firmware should be 138.0.0.0.0 or above.

Hopefully you have already removed your GT 120 from you Mac and the kernel panics have stopped.

 

[vel0city]

Thank you both very much for the detailed advice. Much appreciated.

I'm wondering if the newer Nvidia drivers will improve performance in Octane for C4D? I've mostly switched to Redshift now which is pretty much rock solid stable, but whenever I go back to Octane it crashes C4D at least once an hour.

 

[bsbeamer]

I'm wondering if the newer Nvidia drivers will improve performance in Octane for C4D? I've mostly switched to Redshift now which is pretty much rock solid stable, but whenever I go back to Octane it crashes C4D at least once an hour.

What version & release of C4D are you using?
Still on macOS 10.12?
What version of the NVIDIA Web Drivers and CUDA was causing this issue?