How building linux with Cuda

I build blender for linux, without Cuda perfectly. Now my next step is to add the Cuda support to Blender. I install the toolkit of Cuda version 8 and according to the wiki, in linux you do not have to activate any option in the cmake. Doing it this way does not activate the option to use Cuda. If I activate the option of cycles_cuda_binaries (that according to the wiki it is only for windows), besides that to throw a lot of time compiling, in the end it gives me error.

Can you tell me what you need to do to support Cuda in Blender in the Linux version?

The instructions in the wiki are supposed to be complete. If you encountered an error following them, please paste the error and attach the contents of your CMakeCache.txt.

I’ve looked at the wiki again, but I see that the wiki has been completely redesigned. There is no reference to how to compile Blender with Cuda support in linux, as in the previous version if there was some information.

add CMakeCache.txt


What is the error you are getting?

this are error:

CMakeFiles/Makefile2:1414: fallo en las instrucciones para el objetivo ‘intern/cycles/kernel/CMakeFiles/cycles_kernel_cuda.dir/all’
make[1]: *** [intern/cycles/kernel/CMakeFiles/cycles_kernel_cuda.dir/all] Error 2
Makefile:160: fallo en las instrucciones para el objetivo ‘all’
make: *** [all] Error 2

The actual error is earlier in the logs, perhaps you can attach the complete build log.



Any idea why it fails to compile?

These are not the correct logs, what we need is the log that is printed in the console when building. You can either just copy it manually from the terminal, or run a command like make &> build_log.txt to write it to a file.

I have gone back to compile blender with cuda, it keeps giving me error and this is the log that you have asked me.

Is there a possibility to compile blender with cuda, for a certain card?

It seems the CUDA compiler crashes, maybe it’s running out of memory.

To compile only for a specific card, find the compute capability for your card here.

Then modify CYCLES_CUDA_BINARIES_ARCH in your CMakeCache.txt to only build for that compute capability.

ok thank you very much.