how do I delete symbolic link to Cuda 12 so I can install multiple versions and swtich between them using a simple bash script?

Question

https://notesbyair.github.io/blog/cs/2020-05-26-installing-multiple-versions-of-cuda-cudnn/

I already have Cuda on my WSL2:

@LAPTOP-DNCQ5AAC:/mnt/d/ai/llm/ng-video-lecture$ nvidia-smi
Sun Aug 13 16:22:37 2023
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.116      Driver Version: 528.96       CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  On   | 00000000:01:00.0 Off |                  N/A |
| N/A   42C    P8     2W /  50W |      9MiB /  4096MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A       213      G   /Xwayland                       N/A      |
+-----------------------------------------------------------------------------+

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Mon_Apr__3_17:16:06_PDT_2023
Cuda compilation tools, release 12.1, V12.1.105
Build cuda_12.1.r12.1/compiler.32688072_0

I am pretty sure I said "yes" to creating a symbolic link during installation.

Now to follow that blog at the url I posted above, how do I get rid of my Cuda symbolic link?

I need to install Cuda version 11.8 or 11.7 because I want to run PyTorch C++ on Cuda and they don't have a Cuda 12 version.

I found a command

@LAPTOP-DNCQ5AAC:/mnt/d/ai/llm/ng-video-lecture$ sudo find / -type l | grep libcuda
/home/nobu/miniconda3/envs/rapids-23.04/lib/libcudart.so.11.0
/home/nobu/miniconda3/envs/rapids-23.04/lib/libcudart.so
/home/nobu/miniconda3/envs/ldm/lib/libcudart.so.11.0
/home/nobu/miniconda3/envs/ldm/lib/libcudart.so
/home/nobu/miniconda3/envs/tf/lib/libcudart.so.11.0
/home/nobu/miniconda3/envs/tf/lib/libcudart.so
/home/nobu/miniconda3/envs/paddleocr/lib/libcudart.so.10.2
/home/nobu/miniconda3/envs/paddleocr/lib/libcudart.so
/home/nobu/miniconda3/envs/vision_transformer/lib/libcudart.so.11.0
/home/nobu/miniconda3/envs/vision_transformer/lib/libcudart.so
/home/nobu/miniconda3/envs/pytorch/lib/libcudart.so.11.0
/home/nobu/miniconda3/envs/pytorch/lib/libcudart.so
/home/nobu/miniconda3/envs/open-mmlab/lib/libcudart.so.11.0
/home/nobu/miniconda3/envs/open-mmlab/lib/libcudart.so
/home/nobu/miniconda3/pkgs/cudatoolkit-11.1.1-hb139c0e_12/lib/libcudart.so.11.0
/home/nobu/miniconda3/pkgs/cudatoolkit-11.1.1-hb139c0e_12/lib/libcudart.so
/home/nobu/miniconda3/pkgs/cudatoolkit-11.3.1-h2bc3f7f_2/lib/libcudart.so.11.0
/home/nobu/miniconda3/pkgs/cudatoolkit-11.3.1-h2bc3f7f_2/lib/libcudart.so
/home/nobu/miniconda3/pkgs/cudatoolkit-10.2.89-hfd86e86_1/lib/libcudart.so.10.2
/home/nobu/miniconda3/pkgs/cudatoolkit-10.2.89-hfd86e86_1/lib/libcudart.so
/home/nobu/miniconda3/pkgs/cudatoolkit-11.2.2-hbe64b41_11/lib/libcudart.so.11.0
/home/nobu/miniconda3/pkgs/cudatoolkit-11.2.2-hbe64b41_11/lib/libcudart.so
/home/nobu/miniconda3/pkgs/cudatoolkit-11.8.0-h37601d7_11/lib/libcudart.so.11.0
/home/nobu/miniconda3/pkgs/cudatoolkit-11.8.0-h37601d7_11/lib/libcudart.so
find: File system loop detected; ‘/mnt/wslg/distro’ is part of the same file system loop as ‘/’.
find: ‘/mnt/d/$RECYCLE.BIN/S-1-5-21-2207593063-2945772590-794792826-1001’: Permission denied

What's happening at the end there?

Edit: Thanks to @chan for advice on using multiple cudas.

I read this to install 11.8

How do I install NVIDIA and CUDA drivers into Ubuntu?

so basically at the installation menu I can skip the drivers but right now when I select Driver Options in that menu, I get a blank bash:

Answer 1

It's been a while since I played with CUDA but in my workstaion, I have in .bashrc

export CUDA_HOME=${CUDA_HOME}:/usr/local/cuda
export CUDA_PATH=/usr/local/cuda

and

$ls /usr/local/cuda*
cuda/      cuda-11/   cuda-11.3/
$ ls /usr/local/cuda -l
lrwxrwxrwx 1 root root 22  Jun 10  2021 /usr/local/cuda -> /etc/alternatives/cuda
$ ls -l /etc/alternatives/cuda
lrwxrwxrwx 1 root root 20  Jun 10  2021 /etc/alternatives/cuda -> /usr/local/cuda-11.3

so you can install may CUDA versions under /usr/local and use update-alternative command to set the default version. For example after you install another version, you do update=alternatives -config cuda to set the version you want as the default from the listed versions. Please do man update-alternatives to see how to use it(use with sudo ). (I guess the install script already did update-alternative --install cuda for the new installed version, so it will appear in the update-alternatives --list command.)