cuDNN installation process
1. Download cuDNN
Because the CUDA version downloaded in the previous blog post "Target Detection Step 3" is 11.1, then we need to find the cuDNN version corresponding to CUDA11.1 version. The address is: cuDNN Archive | NVIDIA Developer
2. Install cuDNN
The downloaded cuDNN file is a compressed package. After decompression, the following folder will appear:
Next, you need to copy the files in the three unzipped folders to the folders corresponding to CUDA, and open the CUDA installation directory:
So far, cuDNN has been successfully installed inside CUDA. Because we installed cuDNN, it is necessary to test CUDA again according to the test method mentioned in the previous blog post.
3. Check whether CUDA is affected by cuDNN
A. Check the CUDA version
win+R, enter cmd. Enter "nvcc -V" or "nvcc --version" in it, and if the following screenshot appears, it proves that CUDA is installed successfully:
B. Run the CUDA test file
Open the CUDA installation directory and find the two files as shown in the screenshot below:
The Result at the end of the above two screenshots are all in the pass state, so the cuDNN we just installed will not affect the original CUDA.
Summary: The use of graphics cards, GPUs, CUDA, and cuDNN is a process of gradually increasing demand. When I was young, I liked to play large-scale online games. The original integrated graphics card could not meet the game requirements, so I had to have a good graphics card and a suitable graphics card driver. After growing up, learning computer vision requires the use of GPUs for complex calculations and reasoning. In this case, we know about GPUs. With the deepening of learning, when the simple operation of the original GPU could not meet our network performance in deep learning, CUDA appeared in front of us. Finally, we started to learn the deep neural network, and CUDA's support for GPU could not meet our needs. We introduced cuDNN, which focuses on performance.
Reposted from: Target Detection Step 4: Introduction of graphics card, GPU, CUDA, cuDNN and how to install cuDNN under Windows 10?