GPU devices not detected in tf-nightly in latest releases (Google Colaboratory)

Hello,

I installed tf-nightly in a Google Colaboratory notebook and noticed that the model training time is significantly longer than when I was training it at the beginning of August. Then I ran

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

and no GPU devices were detected, even though the accelerator was set to ‘GPU’.

But after uninstalling the latest release and installing tf-nightly-2.4.0-dev20200815 the GPU was successfully detected.
Trying out different releases, seems like it doesn’t work as of 2.4.0-dev20200819 release.

1 possible answer(s) on “GPU devices not detected in tf-nightly in latest releases (Google Colaboratory)

  1. 2.4.0-dev20200819 is the first release with CUDA 11, so it is very likely that that’s the root issue. Colab needs to be adjusted to use CUDA 11 instead of CUDA 10.1.