TensorFlow™ is an open source software library for high performance numerical computation. Its flexible architecture allows easy deployment of computation across a variety of platforms (CPUs, GPUs, TPUs), and from desktops to clusters of servers to mobile and edge devices. Originally developed by researchers and engineers from the Google Brain team within Google’s AI organization, it comes with strong support for machine learning and deep learning and the flexible numerical computation core is used across many other scientific domains. Source: https://www.tensorflow.org/
The first step to installing TensorFlow™ is determined which version you want to use. Because TensorFlow™ is capable of using the processing power of your GPU, the supported operating systems each have the ability to install TensorFlow with GPU support. TensorFlow-GPU uses the Nvidia CUDA platform to harvest all the parallel computing power of the GPU. Using parallel threads on the GPU increases the execution speed of your TensorFlow code, however all depends on the computing power of your GPU. In order have this GPU-Power, TensorFlow requires a minimal “Compute Capability” of 3.5 (3.0 when installing from source). To check if your GPU is up to this task, you can visit Nvidia CUDA.
An easy mistake to make, when installing TensorFlow-GPU, is the mismatch between Python, CUDA and TensorFlow version. Taylor Denouden has written a good article on Installing Tensorflow GPU on Ubuntu 18.04 LTS. This article takes you step-by-step through the installation process of the CUDA Toolkit and Tensorflow-GPU. And highlights some points where extra attention is needed.
Last tip, as mentioned before, pay attention to the correct versions required. The latest version is not always the best one version to use.