I have first installed Cuda Toolkit 11.3. Sadly Pytorch does not support it yet, so now I want to install cuda toolkit 11.1
I have installed 11.3 via .deb, on Ubuntu 20.04
For uninstallation I have followed
https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#removing-cuda-tk-and-driver
For installing 11.1 I followed https://developer.nvidia.com/cuda-11.1.0-download-archive?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=2004&target_type=deblocal However, apt still only gives me the opportunity to install 11.3.
Please give me advice how I can get fully rid of 11.3 and install 11.1
Thanks!
Hi, there.
Your steps are correct except when installing, you need to explicitly give the version number.
check this for further details software installation - How to install specific version of some package? - Ask Ubuntu
I looked at the documentation and compared with the guide I wrote for Cuda 9.2 here: Installing CUDA Toolkit 9.2 on Ubuntu 16.04: Fresh Install, Install by Removing Older Version, Install and Retain Old Version | Amikelive | Technology Blog
They are still pretty similar. If you want to downgrade, you need to check if the deb packages for 11.3 are still installed. The command dkpg -l | grep cuda should be handy to help you figure out which packages that haven’t been removed.