服务器重启之后NVIDIA出现问题原因汇总

问题一: nvidia-smi报错：NVIDIA-SMI has failed because it couldn‘t communicate with the NVIDIA driver 原因及避坑解决方案

        场景描述: 由于训练服务器卡顿, 服务器重启后, 再次跑模型的时候, 发现cuda不可用, 于是输入“nvidia-smi”才发现了一个错误，如下：

NVIDIA-SMI has failed because it couldn’t communicate with the NVIDIA driver

        这是由于重启服务器，linux内核升级导致的，由于linux内核升级，之前的Nvidia驱动就不匹配连接了，但是此时Nvidia驱动还在，可以通过命令 nvcc -V 找到答案

        但是输入nvcc -V 命令的时候, 发现没有此命令, 说明没有安装, 然后安装nvidia-cuda-toolkit, 安装命令为: sudo apt install nvidia-cuda-toolkit

安装完成之后, 使用nvcc -V 命令, 展示如下:

上网搜索各种方案之后, 解法方法如下:  

第一步: 安装dkms:

sudo apt-get install dkms

第二步: 查看本机连接不上的驱动版本

ls -l /usr/src/

可以看到有个一nvidia的文件, 这里是nvidia-470.94. 如果没有这类文件, 请先下载对应的文件.下载🔗: xxxxx

第三步: 安装适合的驱动:

sudo /data/disk-2T/xxxx/softwares/cuda/NVIDIA-Linux-x86_64-470.94.run

这个安装的路径, 写自己下载470.94.run所在的文件路径.适当自己更换一下路径即可.

或者是命令

sudo dkms install -m nvidia -v 470.94

这条命令 -v 后面需要填写本机的nvidia驱动版本，根据第二步得到! 如果这个安装过程中出现问题, 请看下面的问题三!

到了这里, 如果安装成功, 那么就恭喜了, 哈哈哈, 使用nvidia-smi如果可以展示下图即为成功!.

 

如果安装失败了, 就是下面的问题二了~~~~

问题二:  Nvidia 显卡 Failed to initialize NVML Driver/library version mismatch 错误解决方案

问题复现: 

$ nvidia-smi 
 
-->
Failed to initialize NVML: Driver/library version mismatch

问题分析: 

        NVIDIA内核驱动版本与系统驱动不一致

定位问题: 

1. 查看显卡驱动所使用的内核版本, 命令如下:

cat /proc/driver/nvidia/version

结果如下: 

NVRM version: NVIDIA UNIX x86_64 Kernel Module  470.94  Mon Dec  6 22:42:02 UTC 2021
GCC version:  gcc version 7.5.0 (Ubuntu 7.5.0-3ubuntu1~18.04)

         内核版本Kerner Module为470.93 系统内核18.04

2. 查看系统驱动日志, 命令如下:

cat /var/log/dpkg.log | grep nvidia

结果如下:

2022-06-24 11:45:00 install libnvidia-compute-515:amd64 <none> 515.48.07-0ubuntu0.18.04.1
2022-06-24 11:45:00 status half-installed libnvidia-compute-515:amd64 515.48.07-0ubuntu0.18.04.1
2022-06-24 11:45:04 status unpacked libnvidia-compute-515:amd64 515.48.07-0ubuntu0.18.04.1
2022-06-24 11:45:04 status unpacked libnvidia-compute-515:amd64 515.48.07-0ubuntu0.18.04.1
2022-06-24 11:45:18 install nvidia-cuda-dev:amd64 <none> 9.1.85-3ubuntu1
2022-06-24 11:45:18 status half-installed nvidia-cuda-dev:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:34 status unpacked nvidia-cuda-dev:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:34 status unpacked nvidia-cuda-dev:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:34 install nvidia-cuda-doc:all <none> 9.1.85-3ubuntu1
2022-06-24 11:45:34 status half-installed nvidia-cuda-doc:all 9.1.85-3ubuntu1
2022-06-24 11:45:38 status unpacked nvidia-cuda-doc:all 9.1.85-3ubuntu1
2022-06-24 11:45:38 status unpacked nvidia-cuda-doc:all 9.1.85-3ubuntu1
2022-06-24 11:45:38 install nvidia-cuda-gdb:amd64 <none> 9.1.85-3ubuntu1
2022-06-24 11:45:38 status half-installed nvidia-cuda-gdb:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:38 status unpacked nvidia-cuda-gdb:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:38 status unpacked nvidia-cuda-gdb:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:38 install nvidia-profiler:amd64 <none> 9.1.85-3ubuntu1
2022-06-24 11:45:38 status half-installed nvidia-profiler:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:39 status unpacked nvidia-profiler:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:39 status unpacked nvidia-profiler:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:39 install nvidia-opencl-dev:amd64 <none> 9.1.85-3ubuntu1
2022-06-24 11:45:39 status half-installed nvidia-opencl-dev:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:39 status unpacked nvidia-opencl-dev:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:39 status unpacked nvidia-opencl-dev:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:39 install nvidia-cuda-toolkit:amd64 <none> 9.1.85-3ubuntu1
2022-06-24 11:45:39 status half-installed nvidia-cuda-toolkit:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:40 status unpacked nvidia-cuda-toolkit:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:40 status unpacked nvidia-cuda-toolkit:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:40 install nvidia-visual-profiler:amd64 <none> 9.1.85-3ubuntu1
2022-06-24 11:45:40 status half-installed nvidia-visual-profiler:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:46 status unpacked nvidia-visual-profiler:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:46 status unpacked nvidia-visual-profiler:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:46 configure nvidia-cuda-doc:all 9.1.85-3ubuntu1 <none>
2022-06-24 11:45:46 status unpacked nvidia-cuda-doc:all 9.1.85-3ubuntu1
2022-06-24 11:45:46 status half-configured nvidia-cuda-doc:all 9.1.85-3ubuntu1
2022-06-24 11:45:46 status installed nvidia-cuda-doc:all 9.1.85-3ubuntu1
2022-06-24 11:45:47 configure libnvidia-compute-515:amd64 515.48.07-0ubuntu0.18.04.1 <none>
2022-06-24 11:45:47 status unpacked libnvidia-compute-515:amd64 515.48.07-0ubuntu0.18.04.1
2022-06-24 11:45:47 status unpacked libnvidia-compute-515:amd64 515.48.07-0ubuntu0.18.04.1
2022-06-24 11:45:47 status half-configured libnvidia-compute-515:amd64 515.48.07-0ubuntu0.18.04.1
2022-06-24 11:45:47 status installed libnvidia-compute-515:amd64 515.48.07-0ubuntu0.18.04.1
2022-06-24 11:45:47 configure nvidia-opencl-dev:amd64 9.1.85-3ubuntu1 <none>
2022-06-24 11:45:47 status unpacked nvidia-opencl-dev:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:47 status half-configured nvidia-opencl-dev:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:47 status installed nvidia-opencl-dev:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:47 configure nvidia-cuda-gdb:amd64 9.1.85-3ubuntu1 <none>
2022-06-24 11:45:47 status unpacked nvidia-cuda-gdb:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:47 status half-configured nvidia-cuda-gdb:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:47 status installed nvidia-cuda-gdb:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:47 configure nvidia-profiler:amd64 9.1.85-3ubuntu1 <none>
2022-06-24 11:45:47 status unpacked nvidia-profiler:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:47 status half-configured nvidia-profiler:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:47 status installed nvidia-profiler:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:47 configure nvidia-visual-profiler:amd64 9.1.85-3ubuntu1 <none>
2022-06-24 11:45:47 status unpacked nvidia-visual-profiler:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:47 status half-configured nvidia-visual-profiler:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:48 status installed nvidia-visual-profiler:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:48 configure nvidia-cuda-dev:amd64 9.1.85-3ubuntu1 <none>
2022-06-24 11:45:48 status unpacked nvidia-cuda-dev:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:48 status half-configured nvidia-cuda-dev:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:48 status installed nvidia-cuda-dev:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:48 configure nvidia-cuda-toolkit:amd64 9.1.85-3ubuntu1 <none>
2022-06-24 11:45:48 status unpacked nvidia-cuda-toolkit:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:48 status unpacked nvidia-cuda-toolkit:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:48 status half-configured nvidia-cuda-toolkit:amd64 9.1.85-3ubuntu1
2022-06-24 11:45:48 status installed nvidia-cuda-toolkit:amd64 9.1.85-3ubuntu1

        可以看到曾经安装过系统内核18.04的515.48.07的驱动

3. 查看驱动程序, 命令如下: 

sudo dpkg --list | grep nvidia-*

结果如下: 

ii  libnvidia-compute-460-server:amd64   515.48.07-0ubuntu0.18.04.1                 amd64        NVIDIA libcompute package
ii  libnvidia-container-tools            1.0.5-1                                    amd64        NVIDIA container runtime library (command-line tools)
ii  libnvidia-container1:amd64           1.0.5-1                                    amd64        NVIDIA container runtime library
ii  nvidia-container-runtime             3.1.4-1                                    amd64        NVIDIA container runtime
ii  nvidia-container-toolkit             1.0.5-1                                    amd64        NVIDIA container runtime hook
ii  nvidia-cuda-dev                      9.1.85-3ubuntu1                            amd64        NVIDIA CUDA development files
ii  nvidia-cuda-doc                      9.1.85-3ubuntu1                            all          NVIDIA CUDA and OpenCL documentation

• 可以看到系统安装了ubuntu 内核18.04 下的 nvidia 515 驱动
• 实际系统内核版本与驱动需求的版本不一致是问题产生的根源

解决方案:  

• 卸载现有驱动，重新安装

卸载驱动: 

sudo /usr/bin/nvidia-uninstall
sudo apt-get --purge remove nvidia-*
sudo apt-get purge nvidia*
sudo apt-get purge libnvidia*

输入下面命令不在有任何内容

sudo dpkg --list | grep nvidia-*

重新安装: 

sudo /data/disk-2T/xxxx/softwares/cuda/NVIDIA-Linux-x86_64-470.94.run

安装完成之后, 使用nvidia-smi查看结果.

问题三: gcc版本不匹配造成的安装失败

此时，如果你的gcc（尽量大于7.3版本）版本过低，那么上述命令sudo dkms install -m nvidia -v 470.103.01失败的原因就找到了，查看现有的gcc版本

gcc --version

gcc在/usr/bin目录下, 输入命令查看所有的gcc:

ls /usr/bin/gcc*
ls /usr/bin/g++*

结果如下: 

 将查到的版本加入gcc候选中，最后的数字是优先级，如下:

sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-7 20 --slave /usr/bin/g++ g++ /usr/bin/g++-7
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-9 10 --slave /usr/bin/g++ g++ /usr/bin/g++-9

完成上面的操作之后，我们就可以通过下面的指令来选择不同的gcc和g++的版本了

sudo update-alternatives --config gcc

结果如下: 

这里我选择的是gcc-7选0或是1都可以.成功!~~~🙃

重启电脑, 输入nvidia-smi, 链接成功

以上为总结的三个问题, 哈哈哈, 能帮助大家解决问题的, 就点赞支持一下吧!🤞✌️