You are not logged in.
Apparently tensorflow needs cudnn6 to use the gpu. I have cuda 8 and cudnn5 and 7 but not 6. I get this error whenever I try to import tensorflow after installing tensorflow-gpu: ImportError: libcudnn.so.6: cannot open shared object file: No such file or directory
I have cuda-8.0.61-2 installed from the repositories.
So I got cudnn6 from here: https://aur.archlinux.org/packages/cudnn6/
And tried make -si but I get: error: target not found: cuda-8.0
Anyone having the same problem? How could I fix this?
Last edited by odrec (2017-12-04 17:50:35)
Offline
The only tensorflow-gpu i can find is libtensorflow-gpu from aur .
Is that what you are trying to use and if so, why ?
You have tried running tensorflow from repos ?
Last edited by Lone_Wolf (2017-11-27 14:28:51)
Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.
(A works at time B) && (time C > time B ) ≠ (A works at time C)
Offline
The only tensorflow-gpu i can find is libtensorflow-gpu from aur .
Is that what you are trying to use and if so, why ?
You have tried running tensorflow from repos ?
Sorry I should have clarified. tensorflow-gpu is a python package so it can be installed with pip install. It relies on Cuda though so I needed the Cuda packages installed. I found a solution in the end. I installed Cuda from the repositories which is Cuda 8 and then I got cudnn6 from Aur: https://aur.archlinux.org/packages/cudnn6/
Like it says on the comments, the pkgbuild for cuda 8 seems to be corrupted so one user provided a new pkgbuild here: https://pastebin.com/vpTRmkAz. Now everything works!
Offline