[Debian 12] Do I need to install CUDA separately?

[Dr. Banana]

I can fold with my gpu, just interested to know why I can do so despite not installing CUDA. I installed the proprietary driver, but on debian there are separate instructions for CUDA:

Code: Select all

apt install nvidia-cuda-dev nvidia-cuda-toolkit

Do I need these? I think I read somewhere on this forum that fah downloads its own version of the cuda toolkit, is that still true?

[bikeaddict]

It falls back to using OpenCL if the CUDA libraries can't be loaded. CUDA has significantly better performance than OpenCL, so it's better to use that unless you have some kind of philosophical objection to something proprietary.

[pcwolf]

Seems to me that the -dev and -toolkit are more for *programming* CUDA than running.

Manjaro has a package "cuda" which did improve my folding over just the plain NVidia 550 driver. You can check the folding information log where right after downloading the work project you can see the Platform sort down through Open-Cl to CUDA and shortly thereafter which one is selected for folding.

[toTOW]

The proprietary driver has all the necessary libraries to run FAH with CUDA.