search

deep-dance / core

GNU General Public License v3.0
2 stars 1 forks source link

tensorflow version and cuda compatability #15

Closed nikozoe closed 3 years ago

nikozoe commented 3 years ago

Tensorflow does not find my gpu with cuda 10.2 installed. Apparently it is not compatible and requieres cuda 10.1 instead (see here) I therefore followed the official tensorflow documentation here to get it to work.

Also in the requirements.txt of the "creator" part, an old tensorflow version is listed and therefore installed. With older tensorflow version one would have to install tensorflow-gpu to use the gpu. I installed the newer tensorflow 2.2.0 version which can handle both gpu and cpu. In a jupyter notebook you can check with

print("Num GPUs Available: ", len(tf.config.experimental.list_physical_devices('GPU')))

whether your gpu is recognized.

nikozoe commented 3 years ago

We now have a Dockerfile that takes care of this.