How to install TensorFlow-GPU in ubuntu and Linux

Installing the TensorFlow-GPU version seems very hard for the beginner which is because the version of python and Cuda needs to be supported by TensorFlow-GPU. So, in this blog, I will show you the simple way to install it in Linux OS.

Before starting you should know that your system must full the Hardware requirement which is:

NVIDIA GPU drivers — CUDA 11.2 requires 450.80.02 or higher.

Now install Cuda using the command below:

sudo apt install cuda

The above command will install the latest version of the Cuda in your system.

Now, click the below link which will redirect you to the TensorFlow-GPU webpage:

https://www.tensorflow.org/install/source#gpu

On the webpage you will see below table :

Image 1 (Image from TensorFlow.org)

The above image shows the requirement to install TensorFlow-GPU where column 1 shows the version of TensorFlow and its respective row shows the requirement that our system must have before installing TensorFlow-GPU.

For example:

TensorFlow 2.7.0( Watch First row of image 1) require:

1. Python Version: 3.7 to 3.9
2. Compiler: GCC 7.3.1
3. Build tools: Bazel 3.7.2
4. cuDNN: 8.1
5. Cuda: 11.2

For demonstration, I will be installing the first row in the image 1 table which is TensorFlow-GPU 2.7.0 version.

At first, check the python version and Cuda version using the below command in the terminal:

python3 -V
nvidia-smi #To check the version of cuda

Image 2

Above image 2, show that we have python 3.8.10 and Cuda 11.5 which is higher than the requirement shown in image 1 but still it works.

Now, use the below command to install TensorFlow-GPU 2.7.0:

pip3 install 'tensorflow-gpu==2.7.0'

Below image 3 shows the installation process of TensorFlow-GPU.

Image 3

Now use the below code to conform the installation of the TensorFlow-GPU:

import tensorflow as 
tensorflow_version = tf.__version__
print(tensorflow_version) 

If you run the program than you will get the TensorFlow-GPU version we just install.

Bye, and fell free to commet if you have any query.