Multiple CUDA installation i.e, 10, 10.1 and 10.2 with a stable Nvidia driver [DISCUSSION]

What will be the best way to manage multiple CUDA, cuDNN and TensorRT versions under the same Nvidia driver?

Diving into the field of R&D requires me to switch between multiple CUDA stacks for different projects, like CUDA 10.1 for DeepStream, etc. Is there an official Nvidia document that allows me to do that?

Can I use Anaconda for this?

one possible approach: containers

Take a look at NGC

What you have asked is one of the most important problem statements NGC is trying to solve.