Here are the basic step of installing the CUDA-8.0 driver and toolkit on Ubuntu 16.04

My Device: DELL XPS 15 9550 OS: Windows 10 Home + Ubuntu 16.04 Graphics: GTX 960M + Intel HD something CUDA Version: 8.0

P.S: when your computer has dual graphics card installed like me and you are using non-NVIDIA card for display, DO NOT INSTALL openGL Library when you install CUDA

First Thing First

PLEASE PLEASE read the official CUDA installation guide Check every details including the system requirements and installation status

GPU requirement Check

1.CUDA-capable check Use command: $ lspci | grep -i nvidia This command will list your GPU version, compare it with the CUDA support list in NVIDIA CUDA official site

2.Check Whether your Linux version support CUDA

3.Check gcc installation using $gcc –-version

4.Check kernel header and pacjage development Use command: $uname –r $ sudo apt-get install linux-headers-$(uname -r) to install the correct kernel header

Choose instalation methods

There are several methods to install CUDA, I really recommand using runfile Download the runfile at Here

runfile installation

1. Disable nouveau run command: $ lsmod | grep nouveau if something show up, then nouveau is currently loading To disable it: Creat file named blacklist-nouveau.conf under /etc/modprobe.d, adding following content

blacklist nouveau options nouveau modeset=0

Then run: $ sudo update-initramfs –u Finally use command $ lsmod | grep nouveau again, if no output then you can proceed

1. Text Mode Installation Reboot your computer, when you enter login, use alt+ctrl+F1 to enter text mode, then log in.

run $ sudo service lightdm stop to close GUI interface

run sudo init 3

go to your runfile location, run $ sudo sh cuda_<your_version>_linux.run --override

Follow the option, be sure to choose no when asking whether to install OpenGL if you have dual graphics card

After Installation, restart lightdm again by running $ sudo service lightdm start

Reboot

Env Setting: $ sudo gedit /etc/profile, add following content at end of the profile:

export PATH=/usr/local/cuda-8.0/bin:$PATH export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64

The installation part has finished

Post Check

run $ cat /proc/driver/nvidia/version to print NVIDIA Driver version

run $ nvcc –V to print CUDA version

Try to compile CUDA Example change location to ~/NVIDIA_CUDA-8.0_Samples run $ make

Run CUDA program change location to ~/NVIDIA_CUDA-8.0_Samples/bin run $ sudo ./deviceQuery to print GPU info

It is possible you get error like “cuda driver version is insufficient for cuda runtime”, this is possibally because your CUDA version and Driver version are not compatiable, use the following command: run sudo apt-get purge nvidia* run sudo apt-get install nvidia-current run sudo reboot

• Previous
  LeetCode 刷题笔记