Prepare machine for ml with cuda stack
cuda: get nvidia drivers for the highest version CUDA we want to install 1 - pick latest version from this list 2 - go through selection and you will be given the installation instructions for deb or run files ex deb file link: https://developer.download.nvidia.com/compute/cuda/11.7.0/local_installers/cuda-repo-ubuntu2004-11-7-local_11.7.0-515.43.04-1_amd64.deb the number sequnce between CUDA version and system architecture denotes driver version needed: in this example 515.43.04
SOURCES: https://blog.kovalevskyi.com/multiple-version-of-cuda-libraries-on-the-same-machine-b9502d50ae77
BREAKING
if we remove the cuda tool kit that comes with the system then the applications taht are linked against shared object files from that tool kit version can break example: ffmpeg’s hardware acceleration can depend on so files from cuda tool kit
so get the version of system cuda tool kit and install using above method
for runfiles wget FILE
example : sudo sh cuda_11.6.2_510.47.03_linux.run —silent —toolkit —toolkitpath=/usr/local/cuda-11.6
cuda stack CUDA CuDNN CuBLAS NCCL