1. Install cuda
- Enter nvidia-smi on the command line to view the driver information.
Download the corresponding version of the installation program CUDA Toolkit Archive | NVIDIA Developer from the official website.
Verify after the installation is complete:
Verify: Enter nvcc --version to check
2. Install cudnn
Download Cudann: cuDNN Download | NVIDIA Developer (you need to register an account first)
Unzip the compressed package to the CUDA installation path (D:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1)
Verification:
Enter d:\Program Files\ NVIDIA GPU Computing Toolkit\CUDA\v11.1\extras\demo_suite
runs bandwidthTest.exe
and the output result is Result = PASS, the installation
is successful
3. Install the GPU version of Pytorch
Enter the pytroch official website: https://pytorch.org/get-started/previous-versions/
to find the corresponding version above, and directly copy the corresponding command to download
Using the wheel command is obviously faster.