I'm trying to render on a Tesla K80, but it doesn't work.

blender --debug-cycles 

outputs

Ignoring device Tesla K80, not officially supported yet

Release notes of 2.77 say:

Add support of GK210 GPUs (NVidia Tesla K80). (ff0dcc5) 

https://wiki.blender.org/index.php/Dev:Ref/Release_Notes/2.77/Cycles

But I had to use a custom build of Blender (very recent), so I'm not sure if the problem is of the custom build or if Teslas aren't supported at all yet. Couldn't try further cause my shift ended.

Any information on the current state of "Tesla & Cycles" I will mark as correct answer.

EDIT I've search in the Blender source for the string "Ignoring device Tesla ..." and I think I found it in opencl_util.cpp although I'm not 100% sure. If that's true it would mean that Blender uses OpenCL instead of CUDA. Not sure why it would, CUDA drivers and toolkit where succesfully installed (confirmed) on the target machine. Any insight on how to force CUDA instead of OpenCL?

EDIT2: It doesn't work either with official Blender builds. Same problem. Now some linux specialists had a look at the issue but couldn't figure it out yet.

  • What's custom in your custom build? It is indeed possible that the build you are using hasn't got the required CUDA kernels compiled. – aliasguru Mar 25 '17 at 11:21
  • It's the new Pricipled Shader only, but additional branches like denoising have been requested already. So the offical version of Blender definitely should work with Tesla K80? – Lee.D Mar 25 '17 at 11:25
  • I'd assume so, though I don't have access to one. Did you run a test with latest Blender 2.78c yet? If that works, it's a matter of providing a build with the same cuda toolkit version. – aliasguru Mar 25 '17 at 11:48
  • I don't have access to the GRID from home either, will have to test it again on Monday. First thing I will be testing is an offical version, I see now I should have started with that. Thanks for the advice about cuda toolkit, I'll read into that. – Lee.D Mar 25 '17 at 12:10
  • the cuda toolkit is only relevant for the one who is compiling the custom blender build. The user does not need it. If the one compiling it doesn't have the latest toolkit installed, some cuda kernels won't be generated, and thus certain devices requiring it won't be supported. – aliasguru Mar 25 '17 at 12:37

Your Answer

 
discard

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.