What Do We Need to Know to Install TensorFlow with GPU Correctly?

Using a GPU to train our Deep Learning models is almost essential these days, because training on a GPU is so much “faster” than training on a CPU. In the following link you can find an explanation of why a GPU is faster than a CPU for training Deep Learning models: Python vs Numpy vs GPU (TensorFlow): Execution Times

If this is the first time you are facing a TensorFlow installation and you want to use the GPU, you need to know what to install on your PC (NVidia drivers, CUDA, cuDNN, Python and TensorFlow-gpu) and with which versions, so that TensorFlow can use the GPU. In the following video we explain it in detail:

In short, you need to answer the following questions (LINKS):

1. Which operating system do I use?
2. Which GPU do I have?
3. Which driver version can I install?
   • Driver download: Descarga los últimos controladores oficiales de NVIDIA (NVIDIA)
4. Which CUDA versions can I install depending on the version of the drivers I already have installed?
   • CUDA vs Drivers compatibility: CUDA Compatibility (NVIDIA Docs)
   • CUDA download: CUDA Toolkit 13.3 Update 1 Downloads (NVIDIA Developer)
5. Which TensorFlow versions (with Python) can I install depending on the CUDA version? And which cuDNN version should I install?
   • Python vs TensorFlow vs CUDA vs cuDNN compatibility: Build from source on Windows (TensorFlow)
   • cuDNN download: cuDNN Archive (NVIDIA Developer) (you must sign up to be able to download cuDNN)