You are not logged in.
Hi All,
Last month I installed TensorFlow and in order to get it to play nice with my GPU, I had to manually install cudnn in my CUDA installation. This was necessary to meet install requirements for Tensorflow that worked on a slightly older version of cudnn than current.
This involved copying 3 files from a cudnn install package into my file tree inside /opt/cuda.
It worked as expected.
However, after my last update, with pamac I received a window with the notice that said something to the effect that my last install could not get file information for post transaction hooks for file libcudnn.so.6 .
I stupidly closed the window and having reviewed my pamac, journalctl and pacman logs, can find nothing amiss.
It doesn't seem too much of a big deal and I know where to copy any cudnn files to fix up CUDA to play with Tensorflow however, I am uneasy about copying random files here and there in my setup and I wondered, what is the proper way to do this? It sound like basically, there is somewhere where these files should be registered as installed and they aren't and that is the problem.
Happy to do some reading but the topics I could see on the Arch site seemed to relate to using the AUR. Appologies if it's obvious but even just a steer to the right page in the Arch Wiki would be appreciated.
Simon
Last edited by simon_sjw (2017-10-27 06:22:46)
Offline
$ pacman -Ss cudnn
community/cudnn 7.0.2-1
NVIDIA CUDA Deep Neural Network library
community/cudnn6 6.0.21-2
NVIDIA CUDA Deep Neural Network library (version 6)
$
The (recently added) cudnn6 PKGBUILD should help to clarify what's needed .
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)
Online
Thanks Lone_Wolf did the trick.
Offline