1. Install the graphics card driver
Refer to another article: Tutorial on installing Nvidia graphics card driver in Ubuntu20.04_ytusdc's blog-CSDN blog
2. Install CUDA
NVIDIA official website (latest version): CUDA Toolkit 12.2 Update 1 Downloads | NVIDIA Developer
CUDA historical version download address: CUDA Toolkit Archive | NVIDIA Developer
Here is the official installation guide given by nvidia (you can refer to it if you encounter problems):
NVIDIA CUDA Installation Guide for Linux
2.1. Select runfile and execute the installation according to the code entered into the terminal as prompted by the official website:
2.2. Installation process option selection
gcc, if an error is reported, just install gcc and g++.
sudo apt-get install gcc
sudo apt-get install g++
2.3. Then continue and enter accept.
2.4. Then continue, select the first driver and press Enter to cancel, because we already have a driver.
2.5. Configure environment variables in .bashrc
Then just install it directly. After waiting for a while, the installation will be completed, and then configure the environment variables in .bashrc.
sudo gedit ~/.bashrc
Add: to the last line of the opened file:
export PATH=$PATH:/usr/local/cuda/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64
export LIBRARY_PATH=$LIBRARY_PATH:/usr/local/cuda/lib64
Ok, the cuda configuration is completed. The normal test is to use the script in the sample to test. It may be the latest version of cuda. If there is no sample, just assume that it is installed. Use nvcc -V to test whether there is cuda below.
Then
source ~/.bashrc
3. Install CUDNN
Download the corresponding version of cudnn.
The latest version download address: Log in | NVIDIA Developer
cudnn historical version: cuDNN Archive | NVIDIA Developer
Official Installation Guide: Installation Guide - NVIDIA Docs
Download the cudnn version corresponding to CUDA
3.1. Download the Deb version and install it.
After downloading, it is a Deb type file and needs to be decompressed.
sudo dpkg -i cudnn*.deb
3.2. There will be a prompt after decompression. Import the CUDA GPG Key according to the prompts.
Write according to instructions
sudo cp /var/cudnn-local-repo-ubuntu2004-8.5.0.96/cudnn-local-0579404E-keyring.gpg /usr/share/keyrings/
Notice! ! ! Most tutorials on the Internet, including the official website, will let you download 3 packages, but they are actually out. In the latest version, after decompression, these 3 packages are already in /var. You can see these 3 Deb packages in the cudnn package. , just need to decompress in sequence
3.3. Refresh the repository metadata. (required, otherwise the following commands cannot be used)
sudo apt-get update
3.4、Install the runtime library.
sudo apt-get install libcudnn8=8.x.x.x-1+cudaX.Y
3.5、Install the developer library.
sudo apt-get install libcudnn8-dev=8.x.x.x-1+cudaX.Y
3.6、Install the code samples.
sudo apt-get install libcudnn8-samples=8.x.x.x-1+cudaX.Y
above X.Y
and 8.x.x.x 换成 /var/cudnn-local-repo-ubuntu2004-8.5.0.96 中文件中的相应版本号
4. Test whether the installation is successful
Because the structure of the new version seems to have changed, just follow the steps below
cp -r /usr/src/cudnn_samples_v8/ $HOME
cd $HOME/cudnn_samples_v8/mnistCUDNN
make clean && make
./mnistCUDNN
If the following results appear, cudnn is fully installed.
Errors encountered during testing:
As shown in the picture above, it is because the corresponding library is indeed installed. Just install the corresponding library and retest.
sudo apt-get install libfreeimage3 libfreeimage-dev