ubuntu系统部署A40环境

1、安装驱动

Download The Official NVIDIA Drivers | NVIDIA

在上面的链接中选择自己的gpu显卡的类型，下载对应的驱动程序，

根据如下的官方安装方法进行安装

1. Introduction — NVIDIA Driver Installation Guide r570 documentation

#先删除原有的驱动
sudo apt-get remove --purge nvidia*
dpkg -i nvidia-driver-local-repo-$distro-$version_$arch_ext.deb
apt update
cp /var/nvidia-driver-local-repo-$distro-$version/nvidia-driver-*-keyring.gpg /usr/share/keyrings/
#可能不成功，但是没有关系
apt install nvidia-open
apt install cuda-drivers
reboot
#重启之后，输入nvidia-smi,有正确输出则安装成功

2、根据安装的驱动，选择对应的cuda，进行安装

CUDA Toolkit 12.2 Downloads | NVIDIA Developer

vim ~/.bashrc

export PATH=$PATH:/usr/local/cuda-12.2/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-12.2/lib64

参考链接：

手动安装Tesla驱动（Linux)_GPU云服务器(EGS)-阿里云帮助中心

ununtu20.04 cuda11.7和cudnn8.6安装 - Wangtn - 博客园

FAQ:如果安装了家用显卡，再安装计算卡驱动可能会出现问题

An NVIDIA kernel module 'nvidia-drm' appears to already be loaded in your kernel.

https://unix.stackexchange.com/questions/440840/how-to-unload-kernel-module-nvidia-drm

sudo systemctl isolate multi-user.target

lsmod | grep nvidia.drm

sudo modprobe -r nvidia-drm

sudo systemctl start graphical.target