X4000 graphics driver crashes (AMD Radeon HD 7970, AMD FirePro D300, D500, D770, AMD Radeon R9 M380 27" iMac))

Steps causing the bug to occur

Compositions using the Shade with Unlit Image node seem to trigger a bug in the X4000 graphics driver, leading to hangs, GPU crashes, and system crashes.

How did the result differ from what you expected?

The crash reports look something like this:

Graphics hardware encountered an error and was reset: 0x0000000e

or

Graphics hardware encountered an error and was reset: 0x0000002e

and

Thread 8 Crashed:: Dispatch queue: com.apple.root.default-priority
0   libsystem_kernel.dylib        	0x00007fff8dc43866 __pthread_kill + 10
1   libsystem_pthread.dylib       	0x00007fff93af535c pthread_kill + 92
2   libsystem_c.dylib             	0x00007fff8f049b1a abort + 125
3   libGPUSupportMercury.dylib    	0x00007fff8be071ca gpusKillClient + 111
4   libGPUSupportMercury.dylib    	0x00007fff8be0865f gpusSubmitDataBuffers + 484
5   com.apple.AMDRadeonX4000GLDriver	0x00001234803a54d6 0x123480000000 + 3822806
6   GLEngine                      	0x00007fff93c38115 gleUnbindTextureObject + 59
7   GLEngine                      	0x00007fff93c07aea gleUnbindDeleteHashNameAndObject + 69
8   GLEngine                      	0x00007fff93b4ec87 glDeleteRenderbuffersEXT_Exec + 548
9   alex-icosphere-5              	0x000000010c21f45c VuoSceneRenderer_renderToImage + 2832
10  alex-icosphere-5              	0x000000010c1d0fcf vuo_scene_render_image__nodeInstanceEvent + 319
…

or

Thread 8 Crashed:: Dispatch queue: com.apple.root.default-priority
0   libsystem_kernel.dylib        	0x00007fff8dc43866 __pthread_kill + 10
1   libsystem_pthread.dylib       	0x00007fff93af535c pthread_kill + 92
2   libsystem_c.dylib             	0x00007fff8f049b1a abort + 125
3   libGPUSupportMercury.dylib    	0x00007fff8be071ca gpusKillClient + 111
4   libGPUSupportMercury.dylib    	0x00007fff8be0865f gpusSubmitDataBuffers + 484
5   com.apple.AMDRadeonX4000GLDriver	0x00001234803a54d6 0x123480000000 + 3822806
6   alex-icosphere-5              	0x0000000101841506 VuoSceneRenderer_draw + 1382
7   alex-icosphere-5              	0x000000010184470f VuoSceneRenderer_renderToImage + 1695
8   alex-icosphere-5              	0x00000001017fd42f vuo_scene_render_image__nodeInstanceEvent + 319
…

Have you found a workaround?

No

Other notes

  • Vuo version: 1.2.0
  • macOS version: OS X 10.11

alex-sphere.vuo (8.73 KB)

Fragcolor_Uniform_Crash.txt (51.1 KB)

Fragcolor_Crash.txt (50.1 KB)

We haven’t been able to reproduce this crash using the hardware we have here. We’ll see if we can obtain hardware that exhibits this bug, then hopefully find a workaround.

Hey @smokris, I was able to check this bug on a current Late 2015 AMD iMac. We get exactly the same crash. Attached is the system details. I used the Alex-Sphere.vuo with Vuo 1.2 on a kiosk machine at an Apple Retailer. The R9 M380 didn’t reset- so I had to hard reboot the machine- and in the process all of the logs were wiped (as is the case in Kiosk mode. Machine was 10.11, 5K Retina, 27inch iMac.

!

@alexmitchellmus hahaha ;) Testing vuo bugs on late high-end Apple retailer kiosk machines ! Love the idea :)

Remember me when I once installed my “iOS screensaver for OS X” in an Apple Store ! People were like “why is this iMac 27 having an iOS lock screen ?” :)

In case that could help to solve the bugs I’ve tested the Composition on an AMD iMac 21" on 10.10 and endured no crash.

Hope it helps.

Specs joined.

iMac.png

@alexmitchellmus and @bodysoulspirit, thanks for checking. On Tuesday hopefully all the parts will arrive, so I can test this myself.

@bodysoulspirit, it makes sense that the 2011 iMac (AMD Radeon HD 6750M) doesn’t experience this X4000 driver bug, since the 6750M uses the X3000 driver.

@smokris thanks for the update! Looks like you will have an upgraded MacPro! The 5770 is getting old anyway! ;-)

Just saw this, is it still an issue with the latest build? I have a 2013 MacPro with an AMD Firepro D500 at home I can test the comp with if needed.

@cwilms-loyalist, thanks for offering to test. The problem does still happen in Vuo 1.2.2, as far as I know. We were able to obtain an AMD Radeon HD 7970 and reproduce it there. The solution will require some fairly deep changes to Vuo’s graphics system (how OpenGL is used in multithreading). Sometime before the 1.3 release, we may put out an alpha release with those changes, and would appreciate if you could test it on your computer.

We’re continuing to make progress on working around this GPU driver bug. We had suspected that certain changes to the graphics system (eliminating the use of OpenGL shared contexts) would fix the crashes/hangs, and we confirmed that they do.

We’ve started using computers with AMD GPUs more frequently, and have also observed similar system-wide hangs when using Final Cut Pro X on those computers (without any Vuo compositions running). And other people have reported issues using Photoshop’s GPU acceleration with AMD GPUs. So this seems to be a fairly widespread AMD driver issue :^/

Thanks for the update Jaymie!

We just released Vuo 1.2.6-alpha2, which should fix this issue. We’d appreciate if you could test and let us know if it works for you.

AMAZING!!! I’ve been testing the same project that used to completely lock up for a few hours now and not one freeze!!! I’ve even added some of the complexity of the composition back in and it’s running beautifully!

Fantastic Job VUO team! Thank-you SO much for making this Alpha2 available before release. You just put my VUO project back on track for our early June event and QC is back on the sidelines where it should be. I’m so excited!  

1 Like

Fixed in 1.2.6.