最近在使用飞桨OCR，有几个特殊的符号需要进行识别，手上只有两台机器，一台1080TI单卡(windows 11)，一台1080Ti双卡(linux 22.04)，习惯性追新到飞桨最高支持的cuda11.7，其实1080Ti到cuda10就够用了，后面的新版本差没有明显的性能提升。
windows上无脑安装，linux上安装比较麻烦，记录下安装过程。
cuda、cudnn对nvidia驱动以及内核有依赖关系，cuda 11.7最低驱动版本是450.80，详细请看https://docs.nvidia.com/deeplearning/cudnn/support-matrix/index.html#cudnn-versions-linux

注意：使用离线方式进行安装，要注册Nvidia的开发者账号才能下载相应的安装包。

• 清理之前残留的nvidia驱动

  sudo apt autoremove -y nvidia* --purge
  sudo rm /etc/apt/sources.list.d/cuda*
  sudo apt-get autoremove && sudo apt-get autoclean
  sudo rm -rf /usr/local/cuda*
  

• 更新显卡驱动

  ubuntu-drivers devices
  sudo ubuntu-drivers autoinstall
  sudo apt install -y nvidia-driver-525
  sudo reboot
  

  重启后使用nvidia-smi检测驱动安装是否正确

• 安装 cuda 11.7.1: https://developer.nvidia.com/cuda-toolkit-archive https://developer.nvidia.com/cuda-11-7-1-download-archive

  wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-ubuntu2204.pin
  sudo mv cuda-ubuntu2204.pin /etc/apt/preferences.d/cuda-repository-pin-600
  wget https://developer.download.nvidia.com/compute/cuda/11.7.1/local_installers/cuda-repo-ubuntu2204-11-7-local_11.7.1-515.65.01-1_amd64.deb
  sudo dpkg -i cuda-repo-ubuntu2204-11-7-local_11.7.1-515.65.01-1_amd64.deb
  sudo cp /var/cuda-repo-ubuntu2204-11-7-local/cuda-*-keyring.gpg /usr/share/keyrings/
  sudo apt update
  sudo apt -y install cuda-11-7
  sudo reboot
  

  使用nvcc --version检测cuda版本，如果找不到ncvv需要把/usr/local/cuda-11.7/bin添加到环境变量。

• 安装 cudnn 8.9.3 for cuda 11: https://developer.nvidia.com/rdp/cudnn-download

  wget https://developer.nvidia.com/downloads/compute/cudnn/secure/8.9.3/local_installers/11.x/cudnn-local-repo-ubuntu2204-8.9.3.28_1.0-1_amd64.deb/
  sudo dpkg -i cudnn-local-repo-ubuntu2204-8.9.3.28_1.0-1_amd64.deb
  sudo cp /var/cudnn-local-repo-ubuntu2204-8.9.3.28/cudnn-local-7F7A158C-keyring.gpg /usr/share/keyrings/
  sudo apt update
  sudo apt -y install libcudnn8=8.9.3.28-1+cuda11.8 libcudnn8-dev=8.9.3.28-1+cuda11.8
  

• 安装 nccl 2.18.3 for cuda 11: https://developer.nvidia.com/nccl/nccl-download

  wget https://developer.nvidia.com/downloads/compute/machine-learning/nccl/secure/2.18.3/agnostic/x64/nccl_2.18.3-1+cuda11.0_x86_64.txz/
  tar xvf nccl_2.18.3-1+cuda11.0_x86_64.txz
  sudo mv nccl_2.18.3-1+cuda11.0_x86_64 /usr/local/nccl_2.18.3
  

• 安装 tensorRT 8.6.1 for cuda 11: https://developer.nvidia.com/nvidia-tensorrt-8x-download

  wget https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/secure/8.6.1/local_repos/nv-tensorrt-local-repo-ubuntu2204-8.6.1-cuda-11.8_1.0-1_amd64.deb
  sudo dpkg -i nv-tensorrt-local-repo-ubuntu2204-8.6.1-cuda-11.8_1.0-1_amd64.deb
  sudo cp $ ls /var/nv-tensorrt-local-repo-ubuntu2204-8.6.1-cuda-11.8/nv-tensorrt-local-0628887B-keyring.gpg /usr/share/keyrings/
  sudo apt update
  sudo apt -y install tensorrt=8.6.1.6-1+cuda11.8
  

• 添加路径到环境变量或者.bashrc

  export PATH=/usr/local/cuda-11.7/bin:~/.local/bin:$PATH
  export LD_LIBRARY_PATH=/usr/local/cuda-11.7/lib64:/usr/local/nccl_2.18.3/lib:$LD_LIBRARY_PATH