Only search projects.blender.org
 Log In
New Account
Home My Page Projects Blender 2.x BF release
Summary Activity Tracker SCM Files

Blender 2.6 Bug Tracker: Browse

[#32752] Cuda on OSX Mac Book Pro Retina GT 650M

Date:
2012-10-04 10:45		 Priority:
3
State:
Closed
Submitted by:
detlef mueller (dettty) 		Assigned to:
Nobody (None)
Category:
OS related: OSX		 Status:
Closed
Relates to:
Duplicates:
Patches:
 
Summary:
Cuda on OSX Mac Book Pro Retina GT 650M
Detailed description
After Discussing already that the blender does not see the graphic card as computing device, I did more tests. I was able to run Adobe CS6 mercury engine, which is also cuda based, without problems. Also Cuda-Z Application does recognize the graphics card and delivers results out oh their performance tests.
I have installed the latest Cuda driver: 5.0.36
see also Blender 2.6 Bug Tracker item #32534

Followup

Message
  • Date: 2012-10-04 11:29
  • Sender: Thomas Dinges
  • OSX does not come with precompiled CUDA Kernels, you have to install the CUDA Toolkit 4.2 first.
  • Date: 2012-10-04 13:58
  • Sender: detlef mueller
  • What does that mean? To run Cycles with CUDA in OSX I have to install the Toolkit? I thought the toolkit is necessary for developers.
  • Date: 2012-10-04 14:20
  • Sender: detlef mueller
  • o.k. its finding now the cuda graphics card. When I set the Device to GPU compute the following error message pops up:

    Compiling CUDA kernel ...
    gcc: No such file or directory
    CUDA kernel compilation failed, see console for details.

    both in Experimental and Supported feature set.
    The screen stays with a dark checker board
  • Date: 2012-10-04 14:29
  • Sender: Thomas Dinges
  • You need the GCC compiler as well for your operating system.
  • Date: 2012-10-04 14:58
  • Sender: detlef mueller
  • works now., but its weird if you only want to use a software, that you have to install all of this tools ;-(
  • Date: 2012-10-04 15:30
  • Sender: Thomas Dinges
  • Our Mac maintainer says its problematic to bundle the kernels.

    Closing the report.
  • Date: 2012-10-04 16:03
  • Sender: jens verwiebe
  • It was my decision to ...
    1. not to give advantage to Nv/cuda vs. AMD/openCL users
    2. not to deliver something i cannot test myself

    Easiest solution: get cuda kernels from graficall builds and copy to the /scripts/cache
    To be always on the newest cuda revision ( may vary with OSX used ): get Cuda-Toolkit + cmdline-tools from Apple and compile kernels yourself.

    Jens
 

Attached Files:

No Files Currently Attached

Changes:

Field Old Value Date By
status_idOpen2012-10-04 15:16dingto
close_dateNone2012-10-04 15:16dingto
StatusNew2012-10-04 15:16dingto