0

While I have seen this question asked several years ago, they were incomplete and unanswered. One that was answered, does not apply.

I have a hybrid video setup with the Intel Skylake GT2 and NVidia GTX960M.

I don't like running my system with NVidia drivers because it creates excess heat and noise from the GPU fan (dual fan laptop). The Intel graphics are sufficient for me, even with the simple games I sometimes play.

With that said, I do some video editing and 3D animation rendering. I would like to have a program such as Blender use the NVidia GPU for processing, but I want Ubuntu to use the Intel GPU for its video.

I've tried installing nvidia-cuda-toolkit and nvidia-modprobe but this makes Ubuntu use the NVidia GPU. If I go back into into "Additional Driver" of the Software & Updates application, I can switch it back to Intel, but that removes all the drivers. With the two above packages installed, Blender has the ability to use the NVidia GPU for processing.

I'm aware of nvidia-prime, however, this option doesn't work on my system. I have tried every suggestion on AskUbuntu and every version of the NVidia driver to try to get Prime to work, so please don't suggest that. Choosing the Intel option with Prime messes everything up causing me to have to purge, reinstall and reconfigure much of the system just to get it to boot again. The last time I tried it, I had to reinstall my entire system. I do not want to go down this path again, and I won't try.

My output of lspci shows the devices:

00:02.0 VGA compatible controller: Intel Corporation Skylake Integrated Graphics (rev 06) 01:00.0 3D controller: NVIDIA Corporation GM107M [GeForce GTX 960M] (rev a2)

But I can't get Blender to use the GPU, and it only shows me the CPU option.

So how can I make CUDA available to Blender without Ubuntu using NVidia drivers to run? It must be possible, I just can't figure out how.

2
  • Configure xorg to use the skylake driver instead of nvidia, this would mean manually creating an xorg config to specify the device and screen instead of a dynamic xorg config. You will need the nvidia drivers to use cuda. On FreeBSD the nvidia provided libGL is linked into place, breaking that link should allow the non-nvidia libgl to be used.
    – sambler
    Oct 5, 2017 at 4:47
  • @sambler Thank you, but, doing this still makes CUDA unavailable to Blender. Not sure what I'm missing here...
    – Delorean
    Oct 5, 2017 at 15:43

0

You must log in to answer this question.

Browse other questions tagged .