Let's build models using NVIDIA GPU

First part in the series of Learning Edge Machine Learning

Jenish Rudani
Last updated on Apr 15, 2022 1 min read Tutorial, Guide
Image credit: Unsplash

Installation

Windows 10

Preparing Python Environment

First step is to install Anaconda to maintain different versions of Python for various projects that you might be working on.

Now, after installing Anaconda, add Conda to 'Environment Path Variable' by typing following command in 'Anaconda Powershell Prompt'

conda init powershell

Creating new environment, and installing tensorflow-gpu

conda create --name test python=3.10
conda activate test
python -m pip install tensorflow-gpu

Installing NVIDIA Cuda-Toolkit, cuDNN, Microsoft Visual C++ Compiler

Check compatibility matrix from Tensorflow Official Website and make note of compatible Python, CUDA, cuDNN version.

  1. Download correct version of Cuda Toolkit from Cuda-Toolkit Archive and follow standard installation instruction.
  2. Now, download the correct version of NVIDIA CUDA® Deep Neural Network library (cuDNN) from cuDNN Archive and extract the zip file.
  3. The NVIDIA CUDA compiler nvcc uses a Visual C/C++ compiler behind the scenes, so we need to install compatible C/C++ compiler from Visual C++ Archive.

Integrating Cuda-Toolkit and cuDNN

Copy bin, include, lib directory from cuDNN directory, and paste it in C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2

Setting up Environmental Variables

Add following new Path Variables to User Variable

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\libnvvp
  1. Tensorflow-GPU Compatibility Matrix
  2. Cuda Toolkit Archive
  3. cuDNN Archive
  4. CUDA / Microsoft Visual C++ compatibility
tutorial ml gpu