Solution to installation failure of cuda-8.0 driver on centos7

Solution to installation failure of cuda-8.0 driver on centos7

Once upon a time, the NVIDIA Titan X card is inserted on the centos7 machine, according to the official website tutorial (https://developer.nvidia.com/cuda-downloads) one-click installation of cuda-8.0 everything went smoothly

Installation Instructions:

1. 'Sudo rpm-icuda-repo-rhel7-8-0-local-8.0.44-1.x86_64.rpm'
2. 'Sudo yum clean all'
3. 'Sudo yum install cuda'

Not long ago, when another centos7 installed cuda following the above steps, it was suddenly found that the machine was not plugged in to the GPU.

Today, this machine is inserted with four Titan X blocks to reinstall cuda. Reinstallation is uninstalled cyclically.

# Yum install cuda does not report an error, cuda-install-samples.sh or something is okay, when the real gpu is called

$/Usr/local/cuda/extras/demo_suite/deviceQuery

Modprobe: FATAL: Module nvidia-uvm not found.

$ Ls/proc/driver | grep nvdia

$ Modinfo nvdia

Modinfo: ERROR: Module nvidia not found.

You know, under normal circumstances, the output of modinfo nvdia should be similar.

Filename:/lib/modules/3.10.0-327.36.2.el7.x86 _ 64/extra/nvidia. ko
Alias: char-major-195 -*
Version: 367.48
Supported: external
License: NVIDIA

Follow the tutorials on the official website to start debug...

1.1 view the video card status

$ Lspci | grep-I nvidia visible four graphics cards, normal

1.2 check that the installation package is correct

$ Md5sum cuda-repo-rhel7-8-0-local-8.0.44-1.x86_64-rpm

24fea3b7f2e5f7e3f155cd73bc008108 cuda-repo-rhel7-8-0-local-8.0.44-1.x86_64-rpm

Compared with the checksum (http://developer.download.nvidia.com/compute/cuda/8.0/Prod/docs/sidebar/md5sum.txt) on the official website, no error.

1.3 Check System Dependencies

$ Yum info dkms

$ Yum info libvdpau

$ Yum info kernel-devel

1.4 install the nvdia module for the kernel

The dkms module can be detected by modinfo only after three steps: added, build, and install.

$ Dkms status

Nvidia, 367.48: added

Obviously, the nvidia module was only added during installation, and the installed module has not yet been generated. The reason is unknown.

# Dkms build-m nvdia-v 367.48

The error "kernel headers not found" is reported, probably because/lib/modules/3.10.0-327. el7.x86 _ 64/build/include cannot be found.

If we

$ Cd/var/lib/dkms/nvidia/367.48/source/

$ Make

/Lib/modules/3.10.0-327. el7.x86 _ 64/build: No file or directory error will be reported.

Add a link. If/lib/modules/3.10.0-327. el7.x86 _ 64/build is an existing file, delete it.

$ Ln-s/usr/src/kernels/3.10.0-327.36.3.el7.x86 _ 64 // lib/modules/3.10.0-327. el7.x86 _ 64/build

# Dkms build-m nvdia-v 367.48

# Dkms install-m nvdia-v 367.48

You may need to restart

$ Modinfo nvdia

$ Modinfo nvdia-uvm

Problem Solving ~

Run the cuda quick start guide on the official website and nvidia cuda getting started guidefor linux.

Too long, misleading value is greater than reference value (okay .. Still useful for reference)

Cuda-8.0 no/usr/bin/nvidia-uninstall

It is normal that nvdia is not called in/dev.

Reference

Http://www.cnblogs.com/wwang/archive/2011/06/21/2085571.html

Http://blog.csdn.net/qq_21111579/article/details/49337191