Redhat8 安装 cuda11.8 + cudnn 8.6 + trt8.5.3 保姆级教程

详细记录 redhat 8 系统， 从 0 -1 安装 Cuda, Cudnn , Tensorrt 全部过程。

Driver

1. 我安装的是 NVIDIA-Linux-x86_64-525.89.02.run
2. 安装步骤自行百度下，这个我大哥做系统时候给我安装完事了。
3. 安装驱动时候 一定选支持11.8的驱动，使用 nividia-smi 查询一下。
   

---

Cuda 11.8

1. 下载地址: https://developer.nvidia.com/cuda-11.1.0-download-archive?target_os=Linux&target_arch=x86_64&target_distro=RHEL&target_version=7&target_type=rpmlocal
2. install

   # 不选 driver 和 nvidia-fs
   sudo sh ./cuda_11.8.0_520.61.05_linux.run
   

3. 配环境变量

   sudo vim ~/.bashrc
   export PATH="/usr/local/cuda-11.8/bin:$PATH"
   export LD_LIBRARY_PATH="/usr/local/cuda-11.8/lib64:$LD_LIBRARY_PATH"
   source ~/.bashrc
   

4. 验证

   nvcc -V
   cd /usr/local/cuda-11.8/samples/1_Utilities/deviceQuery
   sudo make
   ./deviceQuery
   
   cd ../bandwidthTest
   sudo make
   ./bandwidthTest
      
   # 如果这两个测试的最后结果都是Result = PASS，说明CUDA安装成功.
   

5. 后悔药

   # uninstall cuda
   sudo /usr/local/cuda-11.8/bin/cuda-uninstaller
   # uninstall nvidia driver
   #sudo /usr/bin/nvidia-uninstall
   

---

Cudnn 8.6

1. 参考: https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#installcuda

2. 安装

   sudo rpm -i cudnn-local-repo-rhel8-8.6.0.163-1.0-1.x86_64.rpm
   # Refresh the repository metadata.
   sudo yum clean all
   # Install the runtime library.
   sudo yum install libcudnn8-8.6.0.163-1.cuda11.8 -y
   sudo yum install libcudnn8-devel-8.6.0.163-1.cuda11.8 -y
   sudo yum install libcudnn8-samples-8.6.0.163-1.cuda11.8 -y
   

3. 不出意外会出现如下错误

     Problem: conflicting requests
      - nothing provides libcublas.so.11()(64bit) needed by libcudnn8-8.6.0.163-1.cuda11.8.x86_64
      - nothing provides libcublasLt.so.11()(64bit) needed by libcudnn8-8.6.0.163-1.cuda11.8.x86_64
   

4. 下载先关依赖rpm ，如果还出现什么依赖库问题，去下载相应rpm，应该都能解决.
   1. https://rhel.pkgs.org/ , 搜索 cuda-toolkit-config-common-11.8.89-1.noarch.rpm cuda-toolkit-11-config-common-11.8.89-1.noarch.rpm cuda-toolkit-11-8-config-common-11.8.89-1.noarch.rpm libcublas-11-8-11.11.3.6-1.x86_64.rpm libcublas-devel-11-8-11.11.3.6-1.x86_64.rpm
   2. https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/ , 搜索 cuda-cudart-devel-11-8-11.8.89-1.x86_64.rpm

   # install
   sudo yum install cuda-toolkit-config-common-11.8.89-1.noarch.rpm -y 
   sudo yum install cuda-toolkit-11-config-common-11.8.89-1.noarch.rpm -y 
   sudo yum install cuda-toolkit-11-8-config-common-11.8.89-1.noarch.rpm -y 
   sudo yum install libcublas-11-8-11.11.3.6-1.x86_64.rpm -y 
   sudo yum install libcublas-devel-11-8-11.11.3.6-1.x86_64.rpm -y 
   

5. 验证

   cd /usr/src/cudnn_samples_v8/mnistCUDNN
   make clean && make
   ./mnistCUDNN
   Test passed!
   

6. 不出意外还会出现错误，fatal error: FreeImage.h: No such file or directory. 按照网上的办法都是安装 freeimage 和 freeimage-devel, 但我安装会出现 not match… 错误, 所以还是老套路，本地下载包安装。

   wget https://download-ib01.fedoraproject.org/pub/epel/8/Everything/x86_64/Packages/j/jxrlib-1.1-1.el8.x86_64.rpm
   sudo yum install jxrlib-1.1-1.el8.x86_64.rpm
   wget https://download-ib01.fedoraproject.org/pub/epel/8/Everything/x86_64/Packages/f/freeimage-3.17.0-5.el8.x86_64.rpm
   sudo yum install freeimage
   wget https://download-ib01.fedoraproject.org/pub/epel/8/Everything/x86_64/Packages/f/freeimage-devel-3.17.0-5.el8.x86_64.rpm
   sudo yum install freeimage-devel
   

7. 后悔药

   sudo yum erase "cudnn-local-repo*" -y
   sudo yum erase "libcudnn*"
   

---

Tensorrt

1. 安装 https://docs.nvidia.com/deeplearning/tensorrt/install-guide/index.html#installing

   os="rhelx"
   tag="cudax.x-trt8.x.x.x-ga-yyyymmdd"
   sudo rpm -Uvh nv-tensorrt-repo-${os}-${tag}-1-1.x86_64.rpm
   sudo yum clean expire-cache
   sudo yum install tensorrt
   

2. 按照官方教程安装，还是会提示缺少依赖，也不知道那里有问题。有大佬知道，麻烦指导下，好人一生平安啊。 最后还是老套路，下载相应依赖。

   wget https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-cccl-11-8-11.8.89-1.x86_64.rpm
   sudo yum install cuda-cccl-11-8-11.8.89-1.x86_64.rpm -y
   wget https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-nvcc-11-8-11.8.89-1.x86_64.rpm
   sudo yum install cuda-nvcc-11-8-11.8.89-1.x86_64.rpm -y
   wget https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-cudart-11-8-11.8.89-1.x86_64.rpm
   sudo yum install cuda-cudart-11-8-11.8.89-1.x86_64.rpm -y
   wget https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-cudart-devel-11-8-11.8.89-1.x86_64.rpm
   sudo yum install cuda-cudart-devel-11-8-11.8.89-1.x86_64.rpm -y
   sudo yum install tensorrt
   

3. 验证

   rpm -qa | grep tensorrt
   # You should see something similar to the following:
   tensorrt-8.5.3.x-1.cuda11.8.x86_64
   

4. 后悔药

   sudo yum erase "libnvinfer*" -y
   sudo yum erase "nv-tensorrt-repo*" -y
   sudo yum erase "nv-tensorrt-local-repo*" -y
   sudo yum erase graphsurgeon-tf onnx-graphsurgeon -y
   sudo yum erase uff-converter-tf -y
   sudo yum autoremove -y
   

---

END

1. 以上记录全部的安装过程，有需要的小伙伴可以参考一下。
2. 码字不易，欢迎三连 。

愿你看尽繁华，归来仍是少年。