You are not logged in.
Pages: 1
I've been trying to use TensorFlow, but I've been encountering an error on trying to run a program utilizing it:
ImportError: libcudart.so.9.0: cannot open shared object file: No such file or directory
Anyway, after doing some Google-fu I found this issue on GitHub. Supposedly, the issue is that I have CUDA 9.1 installed, but I need 9.0 installed. The cuda package is currently at 9.1, and there is no cuda-9 package in the AUR. Of course, NVIDIA didn't provide a CUDA 9.0 package for Arch. Symlinking didn't work either, just began to list off other libraries that needed symlinking and then told me I was using the wrong version.
Any ideas on how I would go along installing CUDA 9.0 on Arch?
Offline
Offline
Didn't think about that, thanks.
Offline
@fdservices unless every package was rolled back to a data cuda was in the repositories, would that not be a partial update?
@SilverMight tensorflow is in the community repository and should be linked against cuda 9.1 could you not use that?
Offline
@loqs yes, and partial updates are not recommended - so beware.
@SilverMight did you install tensorflow using pacman or the AUR? If not then tht my well be your problem. Pacman will keep track of dependencies, if you install in any other way then you are on your own ![]()
Andrew
Offline
I have tensorflow-cuda however tensorflow is being used in a python virtualenv. The version being used does not work with cuda 9.1, only cuda 9.0
Offline
You can not rebuild the version of tensorflow in virtualenv to use cuda 9.1?
Offline
Pages: 1