[Blender] How to use NVIDIA GPU in Blender Flatpak version (Linux/GNU) ?
So I found the answer how to use NVIDIA GPU in Blender with this answer https://askubuntu.com/questions/1316566/how-do-you-offload-a-flatpak-application-to-run-on-the-nvidia-graphics-card-when
In Short version: Run in Terminal the following command
flatpak override --user --device=dri --env=__NV_PRIME_RENDER_OFFLOAD=1 --env=__VK_LAYER_NV_optimus=NVIDIA_only --env=__GLX_VENDOR_LIBRARY_NAME=nvidia org.blender.Blender
So, after i did that and used the command watch -n 1 nvidia-smi after I started Blender, I noticed that is using Nvidia, however when I went to renderings and wanted to choose Cycles with GPU Compute → the GPU Compute was greyed out
What did I miss?
Edit: I found the related Issue https://github.com/flathub/org.blender.Blender/issues/8
Has someone succeeded in using Cuda in Blender Flathub Version?
Comments
I have been using Pop!_Os from couple of years now which is based on Ubuntu. In my experience nvidia works very well with popos (you need to download the nvidia version here).
Cuda is available in blender for me. I have also used pytorch for some small experiments and cuda works well
I am using MX Linux (Debian based) and they also have a tool to install the nvidia drivers.
However, there could be a fault with the nvidia install script.
Did you use Blender as flatpak? Or was it from the distro repo etc
I usually download the latest .tar.xz file from blender.org and then use the blender executable from its un-compressed folder. I have also installed blender from Pop!_OS shop. cuda has worked well in both the cases.
I was using snap's Blender in Ubuntu and had a ton of problems. In particular, I couldn't pip install packages to the Blender internal Python. Changed to using the tarball from blender.org and problems gone. CUDA is available in cycles rendering devices.
I checked the Debian nvidia site and did install the cuda and the optix library, and then I did download the tarball and cuda did function. Afterwards, I opened the flathub version and there it now functions. I don't know why it does now, however I am happy :)