CUDA error: Illegal address in cuCtxSyncronize() #43780

New Issue

Giuseppe De Marco · 2015-02-22T18:39:46+01:00

Giuseppe De Marco commented

2015-02-22 18:39:46 +01:00

System Information
Debian jessie 8 and Ubuntu 14.04 with a GTX690

Blender Version
2.73a hash bbf09d9

Linux maker 3.16.0-4-amd64 #1 SMP Debian 3.16.7-ckt4-3 (2015-02-03) x86_64 GNU/Linux

same fault on Ubuntu 14.04, same blender version.

I'm running the rendering, avoiding this fault, decreasing materials dinamics ( noise, voronoi, too many nodes with high scales and details ) and increasing world streight ( global illumination ).

I'm using CuDA on a GTX690, using all its kepler (x2) core.
The same video card is used for Xorg.

Is there any experience avoiding this faults, decoupling cuda card from xorg ?
Example, I could use an intel for Xorg and let the Cuda works on a dedicated card.

If needed I can upload the faulty blend for tests.

**System Information** Debian jessie 8 and Ubuntu 14.04 with a GTX690 **Blender Version** 2.73a hash bbf09d9 Linux maker 3.16.0-4-amd64 #1 SMP Debian 3.16.7-ckt4-3 (2015-02-03) x86_64 GNU/Linux same fault on Ubuntu 14.04, same blender version. I'm running the rendering, avoiding this fault, decreasing materials dinamics ( noise, voronoi, too many nodes with high scales and details ) and increasing world streight ( global illumination ). I'm using CuDA on a GTX690, using all its kepler (x2) core. The same video card is used for Xorg. Is there any experience avoiding this faults, decoupling cuda card from xorg ? Example, I could use an intel for Xorg and let the Cuda works on a dedicated card. If needed I can upload the faulty blend for tests.

Giuseppe De Marco commented

2015-02-22 18:39:46 +01:00

Changed status to: 'Open'

Giuseppe De Marco commented

2015-02-22 18:39:46 +01:00

Added subscriber: @demarcog83

Martijn Berger commented

2015-02-22 19:04:08 +01:00

Added subscriber: @MartijnBerger

Martijn Berger commented

2015-02-22 19:04:08 +01:00

The same video card is used for Xorg. <--- most likely cause is that your are running out of vram.

run nvidia_smi and see how close you get to running out

The same video card is used for Xorg. <--- most likely cause is that your are running out of vram. run nvidia_smi and see how close you get to running out

Sergey Sharybin commented

2015-02-23 10:46:44 +01:00

Added subscriber: @Sergey

Sergey Sharybin commented

2015-02-23 10:46:44 +01:00

Two possibilities here:

Either your scene is too complex to be rendered o nthe display GPU (because CUDA sets a time limit on how long you can run the kernel)
Or you're running out of VRAM.

In any case, having full logs from the terminal would help. And as was mentioned above, check how much physical VRAM is being allocated (it'll be more than blender reports because of the fragmentation and memory used by the kernel itself).

Two possibilities here: - Either your scene is too complex to be rendered o nthe display GPU (because CUDA sets a time limit on how long you can run the kernel) - Or you're running out of VRAM. In any case, having full logs from the terminal would help. And as was mentioned above, check how much physical VRAM is being allocated (it'll be more than blender reports because of the fragmentation and memory used by the kernel itself).

Giuseppe De Marco commented

2015-02-24 15:46:33 +01:00

Yes, you're right
I moved monitor plug and xorg to the intel card, then I configured mesa glx with

update-alternatives --config glx

this permits me to leave nvidia glx support inside the system without purge it and breaking drivers and cuda dependencies.
Now I'm rendering my animation without any problems, GPU video card is now completely dedicated to rendering/HPC and all is calm on western side.

thank you guys

Yes, you're right I moved monitor plug and xorg to the intel card, then I configured mesa glx with ``` update-alternatives --config glx ``` this permits me to leave nvidia glx support inside the system without purge it and breaking drivers and cuda dependencies. Now I'm rendering my animation without any problems, GPU video card is now completely dedicated to rendering/HPC and all is calm on western side. thank you guys

Bastien Montagne commented

2015-03-05 11:00:23 +01:00

Changed status from 'Open' to: 'Archived'

Bastien Montagne closed this issue

2015-03-05 11:00:23 +01:00

Bastien Montagne self-assigned this 2015-03-05 11:00:23 +01:00

Bastien Montagne commented

2015-03-05 11:00:23 +01:00

Time to close then. :)

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

CUDA error: Illegal address in cuCtxSyncronize() #43780