Tensorflow Installation in Windows 10 with GPU support

Tensorflow is an open source library from Google used for developing and training deep learning neural networks.

I know that! How do you install it? 😖😖

It is highly recommended that you create a virtual environment before proceeding to the installation process to avoid any conflicts with your existing packages and projects. If you have anaconda/minniconda installed, you can create a virtual environment by

Replace “virtualenv” with the environment name of your choice.

The version numbers of the software components are very strongly tied together. The following are the software requirements for tensorflow installation

It is important that you use the exact version numbers specified above (I’ll keep the tutorial updated with the latest version numbers).

It’s better to continue with the Express installation option unless you’re feeling adventurous.

You have to create an account and fill in a short survey before proceeding to the download page. Since we are using CUDA 10.1, you have to download the cuDNN version 7.6.5. Unzip the downloaded zip file and copy the contents of to the following location

After copying the files, your folder structure should look something like this

Most probably that folder doesn’t exist and you might have to create it. The reason for this will be stated later in this article.

After all of the downloading and installing is completed, we have to modify the environment variables. You have to add couple of folders to your existing path variable. You can do that from command prompt or using the traditional method from system properties dialogue box. It doesn’t matter which method you choose, but the following folders must be added to the path variable. If you changed the install location then change the paths accordingly.

Before adding them to path make sure that all of those directories actually exist. After all of the above folders are added to the path, you can proceed to installing tensorflow. You can install tensorflow with the following command (Make sure to activate the environment you created previously before installing tensorflow).

The above command works only for tensorflow versions above 1.15. If you want to install older versions, the CPU and GPU versions are different.

It’s time to check if the tensorflow is actually using the GPU. Open command prompt, gather some courage and run the following commands

If you can see the name of your GPU and some additional details of the GPU, then tensorflow is using your GPU. Congratulations!!

The whole process shouldn’t take more than 20 minutes to complete if you’re on a decent internet connection. Now that you’ve setup tensorflow to use your GPU, go ahead and train your first neural network to identify cats and dogs!!

Thank you for reading!!

Related posts:

Math and writing have come a long way since I was in middle school back in the early 80s, but one thing hasn't changed: the importance of practicing the fundamentals.

Someone wants to adopt you, they said. Funny, I thought. We’re cleaning up the streets, they said. I opened my mouth to speak and ended up here. I haven’t been out for two months. There are sixteen…

At certain stages in this great journey we call life we sit down and take a retrospect, going over, hand in hand with memory, the crooked old road and carefully treading in the same footsteps…