1 Download and installation of cuda

1.1 Check the suitable cuda version

 The cuda supported on my computer is 11.6

1.2 cuda toolkit download

​​​​​kCUDA Toolkit Archive | NVIDIA Developer

Enter the above webpage and find the suitable cuda

1.3 cuda toolkit installation

Double-click the exe file to install

1.4 Configuration environment

 Open Settings -> Advanced System Settings -> Environment Variables 

        The ones in the red box are added automatically by the system, and the ones in the blue box are not automatically added by the system in some cases, and need to be added manually. Pay attention to your own path when adding.

NVCUDASAMPLES_ROOT  C:\ProgramData\NVIDIA Corporation\CUDA Samples\v11.6
 
NVCUDASAMPLES11_6_ROOT  C:\ProgramData\NVIDIA Corporation\CUDA Samples\v11.6

 1.4 Verification

 win+R, enter cmd, enter nvcc --version to view the version number, enter set cuda to view the set environment variables

2 cuANN download and installation 

2.1 cuDNN download

 The download address is as follows, you need to register
https://developer.nvidia.com/rdp/cudnn-download before downloading

 The following link has a version that suits you

cuDNN Archive | NVIDIA Developer

2.2 cuDNN configuration

Unzip cuDNN to drive D

 Copy the three folders to the cuda installation directory. The default installation path is

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6

2.3 Add environment variables

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\include
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\lib
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\libnvvp

2.4 Verification

       win+R cmd to enter the installation directory, then enter extras\demo_suite, execute .\bandwidthTest.exe and .\deviceQuery.exe, and get the following figure.

3 Install the GPU version of torch (no need to ignore torch)

When running the deep learning code, the following error occurs because the GPU version of torch was originally installed.

3.1 Install GPU version torch

Enter the official website Start Locally | PyTorch

Select each configuration and copy the red line link

Open torch and torchvision to choose the version that suits you

 Enter the path to execute pip

pip install torch-1.12.1+cu116-cp38-cp38-win_amd64.whl
 
pip install torchvision-0.13.1+cu116-cp38-cp38-win_amd64.whl

 cuda is available, and the code can be executed normally