Csmart-Digit
  • Introduction
    • Csmart-Digit Knowledge Base
    • Introduction and Use Cases
    • Mechanical Drawings
  • Installation
    • Software and Drivers Installation
    • GPU Installation
  • First Use
    • Device Interface
    • Powering On for the First Time
    • Preparing for Optimal Performance
  • System Architecture
    • Overview
    • System Diagram
    • Acquisition System
    • Server and API
    • AI Training System
  • Csmart Fundamentals
    • Data and Digitalization in Coffee
    • Introduction to Digital Images
    • Introduction to Neural Networks and CNNs
    • Creating a CNN Training Dataset
    • Grading Methods Definition
    • Csmart Methodology for Weight Estimation
    • AI Model Metrics
  • User Guide
    • Csmart-Digit Training
  • User Guide - PDF
    • Csmart-Digit Desktop
    • Csmart-Digit Web App
  • Release Info
    • Csmart-Digit Desktop
  • Downloadable Resources
    • Brochures
Powered by GitBook
On this page
  • Why Install CUDA and cuDNN Drivers?
  • Installation of CUDA Toolkit and cuDNN
  • Verify CUDA Installation
  1. Installation

GPU Installation

PreviousSoftware and Drivers InstallationNextDevice Interface

Last updated 3 months ago

NoteNote: Skip this section if the computer was provided by Csmart along with the machine.

Why Install CUDA and cuDNN Drivers?

The Csmart-Digit system uses NVIDIA GPUs to enhance speed and performance when working with AI models. To fully utilize the GPU, the system requires CUDA, a tool that accelerates complex tasks, and cuDNN, a library optimized for AI and deep learning operations. Installing these drivers ensures the system is properly configured for faster and more efficient performance during tasks such as training AI models and processing data.

While this step is not mandatory, as the software can run using only the CPU, it is essential to achieve the best performance from your Csmart-Digit system.

For compatibility with onnxruntime-gpu, the default CUDA version since version 1.19.0 is 12.x, which requires cuDNN 9.x.

Installation of CUDA Toolkit and cuDNN

  1. Install Visual Studio C++ Redistributable for x64 Download and install the VS C++ Redistributable package from Microsoft:

  2. Install cuDNN 9.5 Download cuDNN 9.5 from the official NVIDIA website:

  3. Install CUDA Toolkit 12 Download CUDA Toolkit 12 from the NVIDIA archive:

  4. Move .dll to the Correct Folder Copy all files from:

    • C:\Program Files\NVIDIA\CUDNN\v9.5\bin\12.6

    • C:\Program Files\NVIDIA\CUDNN\v9.5\include\12.6

    • C:\Program Files\NVIDIA\CUDNN\v9.5\lib\12.6\x64

    and paste/overwrite them to:

    • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\bin

    • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\include

    • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\lib\x64

  5. Add New Paths to System Environment Variables Include the following paths in your system’s Path environment variable:

    • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\bin

    • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\include

    • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\lib\x64

Verify CUDA Installation

To confirm that CUDA is installed correctly, click in the search bar and type cmd

In the command line, type these commands:

conda list | findstr cuda
nvcc --version
Visual Studio C++ Redistributable 2019 x64
cuDNN 9.5 Downloads
CUDA Toolkit 12 Downloads