TensorFlow GPU setup with Jupyter Notebook (for Windows)

Jupyter Notebook is one of the most popular IDEs for data science. If you have installed Anaconda Navigator and installed Python 3.x, you can now use this IDE. But nowadays it’s not enough to just use a product. We also need to know how to use it more efficiently.

Jupyter Notebook runs using the CPU in its default mode. If you usually use this IDE to train your deep learning models, you need more processing power. For this, you should run your Jupyter using GPU instead of CPU.

But how do we do it?

Assuming that you have Anaconda Navigator, Python 3.x and the “Desktop development with C++” kit installed in Visual Studio, we proceed step by step.

STEP 1: CUDA toolkit installation

First, we need to install the CUDA toolkit, but here we need to pay attention that the versions of the tools we install are compatible with each other. NVIDIA supports TensorFlow GPU usage for Windows users up to CUDA v11.2. When you enter the site, you may be confused with the v11.8 version, but if you install a version higher than 11.2, you will not be able to connect your Jupyter with the GPU. For this reason, you should check the list published by TensorFlow to ensure that the versions are compatible.