Page MenuHome

NVidia OpenCL support
AbandonedPublic

Authored by Martin Sand (sandstorm) on Aug 21 2016, 9:23 AM.

Details

Summary

Added command line parameters to start Blender with NVidia OpenCL support

Diff Detail

Event Timeline

Martin Sand (sandstorm) retitled this revision from to NVidia OpenCL support.
Martin Sand (sandstorm) updated this object.

Remove lines 154 and 155. You only need one blank line after a code block.

Martin Sand (sandstorm) edited edge metadata.

Removed lines 154 and 155

Brecht Van Lommel (brecht) requested changes to this revision.EditedAug 21 2016, 2:50 PM

We should not add this to the documentation, these are environment variables for developers testing, and we don't officially support users running Blender with them. For example for new NVidia 10xx cards the better solution is to use the latest builds from builder.blender.org or wait for the 2.78 release.

This revision now requires changes to proceed.Aug 21 2016, 2:50 PM

Good point, I was going to request the markup changes and then ask Sergey to take a look like you just did. What changes are you requesting?

If we need this error in the manual we should explain why it can happen:

  • User has a new graphics card not yet officially supported by the Blender version they are running.
  • User has installed the CUDA toolkit (not something they normally need to do).
  • The toolkit does not yet support their graphics card. If Blender detects this it will try to dynamically compile a CUDA kernel and fail.

The first suggestion should be to check if the latest Blender version (official or buildbot) supports their graphics card. The second should be to see if they can install a newer CUDA developer toolkit. In the manual I would not suggest to use OpenCL at all.

We might make NVidia OpenCL officially supported at some point, but right now it can still break at any time.

Thanks for the feedback. So would you put the adapted text into the error section of this page or on another page like troubleshooting/gpu?

As an end-user I would not take it into the support, but there should be an option to use the card if CUDA is not there.
I had this situation with my GTX 660 (I think OpenCL was not possible at that time) and now with the GTX 1070.

Martin Sand (sandstorm) edited edge metadata.

Ok, I updated it accordingly. Missed it that I already added it to the error section.
For the moment I as a user would be happy to have OpenCL command line parameters there.
May I ask why it is not recommended to use OpenCL when having an NVidia grpahics card?
It is the only option to make use of it as long as the CUDA toolkit is not rolled out.

If we officially supported OpenCL on NVidia it would be available as an option in the UI. By the time the next generation NVidia card comes out this may well be the case, in the meantime we do not want to advise users setting up environment variables that may break Blender, even if they happen to work in some cases.

I suggest to change the text to this:

This error may happen if you have a new NVidia graphics card that is not yet supported the Blender version and CUDA toolkit you have installed. In this case Blender may try to dynamically build a kernel for your graphics card and fail.

In this case you can:
# Check if the latest Blender version (official or buildbot) supports your graphics card. 
# If you build Blender yourself, try to download and install a newer CUDA developer toolkit.

Normally users do not need to install the CUDA toolkit as Blender comes with precompiled kernels.
Martin Sand (sandstorm) edited edge metadata.

Updated as requested

Aaron Carlisle (Blendify) added inline comments.
manual/render/cycles/gpu_rendering.rst
148

For buildbot call "experimental" and add link to the download page.

Martin Sand (sandstorm) marked an inline comment as done.

Replace Buildbot with experimental and added download link

Aaron Carlisle (Blendify) abandoned this revision.