ubuntu20.04 + cuda10.0 + cudnn7.6.4

1、装显卡驱动

1.1检测显卡型号和对应推荐安装驱动

lu@computer:~/project$ ubuntu-drivers devices
WARNING:root:_pkg_get_support nvidia-driver-390: package has invalid Support Legacyheader, cannot determine support level
== /sys/devices/pci0000:00/0000:00:1c.4/0000:04:00.0 ==
modalias : pci:v000010DEd00001347sv00001043sd00001A6Dbc03sc02i00
vendor   : NVIDIA Corporation
model    : GM108M [GeForce 940M]
driver   : nvidia-driver-465 - third-party non-free
driver   : nvidia-driver-418-server - distro non-free
driver   : nvidia-driver-450-server - distro non-free
driver   : nvidia-driver-390 - distro non-free
driver   : nvidia-driver-460 - third-party non-free recommended
driver   : nvidia-driver-460-server - distro non-free
driver   : xserver-xorg-video-nouveau - distro free builtin

lu@computer:~/project$ 


1.2如果有安装过旧的驱动,需要先删除旧的驱动:

sudo apt-get purge nvidia*

1.3禁用自带的驱动,通过如下命令创建一个文件:

sudo vim /etc/modprobe.d/blacklist-nouveau.conf

在文件内添加如下内容:

blacklist nouveau
options nouveau modeset=0

更新一下:

sudo update-initramfs -u

1.4安装nvidia驱动

sudo add-apt-repository ppa:graphics-drivers/ppa

sudo apt-get update

sudo apt-get install nvidia-driver-418-server 

sudo apt-get install mesa-common-dev

sudo apt-get install freeglut3-dev

注意:

1.具体驱动名称是“nvidia-driver-xxx”还是“nvidia-xxx”,需要根据上面“ubuntu-drivers devices”输出结果,由上图可见在ubuntu20.04 上是“nvidia-driver-xxx”

2.cuda10.0对应的驱动是410以上的版本,故这里安装418版本

 

1.5确保BIOS中Secure Boot设置为disabled,重启检查驱动是否安装成功

nvidia-smi

注意:

1.也可以执行下面指令自动安装所推荐的驱动,但这种方式没法选择驱动版本

sudo ubuntu-drivers autoinstall

 

2、安装CUDA

链接:https://developer.nvidia.com/cuda-toolkit-archive

2.1CUDA推荐下载.run可以根据提示安装

执行如下命令:

sudo bash cuda_10.0.130_410.48_linux.run

压住回车键,直到服务条款显示到100%。接着按下面的步骤选择:

accept

n(不要安装driver)

y

y

y

 

2.2安装完成后,设置环境变量

打开主目录下的 .bashrc文件添加如下路径,例如我的.bashrc文件在/home/lu/下。

    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-10.0/lib64
    export PATH=$PATH:/usr/local/cuda-10.0/bin
    export CUDA_HOME=$CUDA_HOME:/usr/local/cuda-10.0

终端运行:source ~/.bashrc

检查:nvcc --version

 

3.cudnn的安装

3.1下载安装文件

按需求下载cudnn的安装文件:https://developer.nvidia.com/rdp/cudnn-archive

3.2安装cudnn

解压下载的文件,可以看到cuda文件夹,在当前目录打开终端,执行如下命令:

   sudo cp cuda/include/cudnn* /usr/local/cuda/include/
     
   sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64/
     
   sudo chmod a+r /usr/local/cuda/include/cudnn*
     
   sudo chmod a+r /usr/local/cuda/lib64/libcudnn*

3.3查看cudnn版本

在终端输入

cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2


 

你可能感兴趣的:(人工智能,cuda,cudnn)