A notable point in this post is the installation of TensorFlow 2.9.1 (TF) with CUDA 11.3, which is not officially supported!
<aside>
⚠️ Since the version of what is to be mentioned in this post is really important, keep in mind that what I write are for the time I am writing this post!
</aside>
From this to that
I want to create a Dockerfile based on a machine with the following specifications,
- My computer: Dell XPS 7590 (Intel i7 9750H/2.6GHz, GeForce GTX 1650 Mobile, RAM 32GB, SSD 1TB).
- OS: Pop!_OS 20.04 LTS (a distribution based upon Ubuntu 20.04).
- Nvidia driver (on the physical machine): 510.73.05
- CUDA (on the physical machine): 10.1
- Docker engine: 20.10.17
- Python: 3.9.7
In this Dockerfile we can create a container that supports,
- TensorFlow: 2.9.1.
- PyTorch: 1.12.1+cu113
- CUDA: 11.3
- cuDNN: 8
- Python: 3.8.10
- OS: Ubuntu 20.04.5 LTS (Focal Fossa)
- Zsh & oh-my-zsh are already installed.
- Jupyter notebook is installed and automatically runs as an entrypoint.
- OpenSSH support (for accessing the container via SSH)