It has been a while since I wrote on this platform, as if I had almost forgotten about it. But here I am again, and I have a bunch of long written posts that I will be rolling out at shorter intervals, and not just installation stuff, I promise So as I am pursuing a master’s degree in Computer Science, I had to write a term paper (ugh I hate them!) about some modern use of parallel programming. That’s when I chanced upon CUDA for deep neural networks (or cuDNN). Now this may not sound too modern, but it still is quite unknown for beginners like me (I am always a beginner, believe me).CuDNN is actually a popular library that can be used with many modern deep learning frameworks like Tensorflow, Pytorch and Caffe to utilize the GPU on your computer. Thus this post assumes that your computer already has a NVIDIA CUDA-enabled GPU._NOTE: If you have an AMD GPU, this post won’t work. You can instead try this out._You will also need Visual Studio Express Community Version 2017 or higher for CUDA to work properly on your computer.

Begin with CUDA

I had installed the 10.1 version for stability, but 10.2 is the current version offered. This is also necessary to check as we’ll need to check its compatibility with the version of TensorFlow that we install.

A step in Nvidia CUDA Toolkit installer for Windows 10

It performs all the system checks and also checks for requisites. So if you forgot to install Visual Studio earlier, this installer will remind you again! Also, stick with all the default options in the installer. We do not want to customize anything.Sometimes, an Nvidia GeForce Experience window opens up and will prompt you to sign up to Nvidia. You can skip this step if you want for now. I went ahead and signed up as signing in is required for downloading cuDNN.

#cuda #cudnn #tensorflow #anaconda