I would like to use Qubes for data science work. I understand that OpenGL is not supported for security concerns. I'm wondering whether I could use a OpenCL on a GPU in Qubes for running deep learning algorithms. I'm leaning towards getting an AMD Vega GPU when it comes out. I guess what I am wondering is whether the GPU could safely be in a separate container for churning through machine/deep learning problems and whether there is any reason why that wouldn't work.
Does that make sense? Would this be supported? Is it a security concern?