NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the lat

在Azure 申请了一台 K80的虚拟机，redhat7 系统。在安装cuda时，可以成功，但在执行nvidia-smi 时，总是报错：

NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.

以前经常在centos 上安装cuda ,已经非常熟悉，这次按照原来的讨论安装为什么总报错，排查后，发现：Azure给的系统为最小系统，缺少许多必要的基础环境包，所以必须自己手动安装，总结如下：

1、安装GCC，

yum install gcc*

2、装dkms

wget http://dl.fedoraproject.org/pub/epel/7/x86_64/e/epel-release-7-9.noarch.rpm

 rpm -ivh epel-release-7-9.noarch.rpm

yum install --enablerepo=epel dkms

3、安装kernel 开发包等

yum install kernel*

4、安装cuda

http://blog.csdn.net/u013000139/article/details/53992523

5、挂载驱动：

modprobe nvidia

6、然后可以安装其他包了，anaconda,tensorfolw 等