Tensorflow does not detect GPU - lambdalabs

I try to get tensorflow GPU up and running in a virtual environment (venv):

I use lambdalabs OS is Ubuntu 20.04.3 LTS.

I have following python script: checkGPY.py:

import tensorflow as tf
if tf.test.gpu_device_name(): print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))
else: print("Please install GPU version of TF")

Outside the venv it works fine. I obtain Default GPU Device: /device:GPU:0. If a train a small neural network (NN) and watch nvidia-smi I see that the GPU memory increases during training. So the GPU resources are used for NN training.

However if I run it is inside a venv (I installed tensorflow version: 2.6.0 inside the venv.)

(venv) x@y $ python checkGPU.py

I obtain: Please install GPU version of TF

I obtain also following: Could not load dynamic library 'libcudnn.so.8'; dlerror: libcudnn.so.8: cannot open shared object file: No such file or directory

So I understand that the dynamic library libcudnn.so.8 cannot be accessed from inside the venv.

How can I resolve this?

1 Answer

To solve this I follow the instructions written on Tensorflow in venv using lambdalabs