AI模型运维——NVIDIA驱动、cuda、cudnn、nccl安装

目前大部分使用GPU的AI模型，都使用的英伟达这套。

需要注意的是，驱动、cuda、cudnn版本需要一一对应，高低版本互不兼容。

驱动和cuda对应关系：https://docs.nvidia.com/deploy/cuda-compatibility/index.html

驱动下载：https://www.nvidia.cn/Download/index.aspx?lang=cn

CUDA下载：https://developer.nvidia.com/cuda-downloads

一、NVIDIA驱动安装

看下是否有nvidia-smi命令，如果没用就需要安装驱动

# 卸载驱动，不卸载直接装应该也行
yum remove xorg-x11-drv-nvidia* nvidia-kmod
 
# 安装
rpm -ivh nvidia-diag-driver-local-repo-rhel7-384.183-1.0-1.x86_64.rpm
yum install cuda-drivers

二、cuda安装

cuda

rpm -ivh cuda-repo-rhel7-9-0-local-9.0.176-1.x86_64.rpm
rpm -ivh cuda-repo-rhel7-9-0-local-cublas-performance-update-1.0-1.x86_64.rpm
rpm -ivh cuda-repo-rhel7-9-0-local-cublas-performance-update-2-1.0-1.x86_64.rpm
rpm -ivh cuda-repo-rhel7-9-0-local-cublas-performance-update-3-1.0-1.x86_64.rpm
rpm -ivh cuda-repo-rhel7-9-0-176-local-patch-4-1.0-1.x86_64.rpm
 
yum install cuda
cat /usr/local/cuda/version.txt

cudnn

tar -xzf cudnn-9.0-linux-x64-v7.4.1.5.tgz
cp cuda/include/cudnn.h /usr/local/cuda/include
cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2

环境变量 .bashrc

export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
export CUDA_HOME=/usr/local/cuda

三、nccl安装

rpm -ivh nccl-repo-rhel7-2.4.8-ga-cuda9.0-1-1.x86_64.rpm
# yum update
yum install libnccl libnccl-devel libnccl-static