I’m new to Pytorch. I’m trying to build a new system which will include a Turing GPU to run Pytorch. I also want to install a low-profile GPU for display purpose because my CPU doesn’t support onboard graphic. I’m aiming to have a Kepler-based GT 710. I think I can install the latest Nvidia driver and CUDA 10 on such a system. However can I also successfully compile Pytorch from source? Have this concern because I read Pytorch compiling doesn’t support 3.x architecture.
Actually I wanna ask how the compilation works? Does it recognize the Turing card and will compile smoothly? If so, I’m curious how many CUDA_visible_device will be there. Or does it fail to compile because of Kepler card?
Thanks a lot.