OS: Ubuntu 16.04(which has installed cuda 9.0 and cudnn 7.5)
First, I create a virtual environment using the following command: conda create -n pytorch_env python=3.6 conda
then I activate the pytorch_env and install pytorch using the following command: conda install pytorch torchvision cudatoolkit=10.0 -c pytorch
We all know that the conda will install pytorch gpu version and the cuda 10 and cudnn in the virtual environment pytorch_env. What I to know is whether the cuda 10 in pytorch_env will impact the cuda 9.0 which has been installed in my system ?
It shouldn’t impact any system-wide CUDA installs.
In fact I’m using a lot of different conda environments with different CUDA (and PyTorch) versions besides a system CUDA install for building PyTorch from source. So far I didn’t get into trouble.
OK，got it，thank you very much.
One more question,I install graphic card driver through the ppa:graphics-drivers/dev,
which driver version is better, long lived version 410.78 or short lived version 415.27 ?
Is the graphic driver in the ppa:graphics-drivers/dev reliable？or Should I install the graphic driver through the NVIDIA official website?