There are different ways for installing the mentioned library.
Both the Deb packages and normal tar files exist. My question is, which one should I be choosing over the other? Why would I want to use the deb packages or vice versa?
I read somewhere that using the deb packages in this very specific case is not recommended, as these packages, contain all the dependencies, and therefore if you install it, they may very well overwrite the newer libraries, and result in unstable OS, or some other applications. On the other hand, deb packages allow for easy installations and removal, and their updates seem to be a easy. So whats the rule of thumb in this situation (dealing with libraries such as cuda/cudnn/etc)

1 Answer 1


In my experience, I found that using the packaged .deb is the easiest way forward . The zipped versions are used for building from source. This is usually used when you have specific changes to make to the build process (definitely not required for the average user). Moreover, the build process takes time and is resource heavy.

In addition, some packages are available on the repository such as nvidia-cuda-toolkit. This makes for an easy install.

  • Thanks a lot. it seems nvidia-cuda-toolkit is not particularly up-to-date? right? cause the cuda 10.2 is out but it only has cuda9.1 if I'm not mistaken ! So we cant rely on the deb packages to be the most up-to-date versions?
    – Hossein
    Dec 11, 2019 at 14:34
  • That is correct. The repository versions may not be up to date, but they are guaranteed to be stable if you are on an LTS release of Ubuntu. For example, while your repository contains cuda9.1, my Kubuntu 19.10 installation contains toolkit version 10.1.168-1build1. Dec 11, 2019 at 17:41

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .