I’m new to Pytorch (and Anaconda). I’ve read related topics in the forum but am still confused about the compatibility. My ubuntu has already got CUDA 10.1 + CuDNN installed with Nvidia 418.56 driver. I’ve also had CUDA environment and library path setup. Then I installed Pytorch through Anaconda which comes with CUDA toolkit 10.0. I’m wondering:
- Am I correct that Pytorch only calls CUDA tools purely within the Anaconda library path, without looking for CUDA tools in the OS environment?
- For long term, is it safe to keep two versions of CUDA like this? I’d like to keep the OS-based CUDA for some graphic works, as its feature is a bit more up-to-date. But I don’t want to mess things up.
Any suggestion? Thanks.