This configuration is only a test configuration, it is estimated that with this type of graphics card to do parallel computing almost no, but the configuration method, a lot of it is worth borrowing, and ultimately want to apply in GEANT4 parallel computing.
All right, ladies and gentlemen, I'm starting to play.
First, if you have failed to install many times, then make sure that you have uninstalled the existing graphics drivers and Cuda
1. If you have already installed Cuda-related, and unsuccessful (do not see this success), please see below related, if not installed, please skip 1.1 uninstall the relevant files
In the table of contents (for example, cuda7.0 version):
/usr/local/cuda-7.0/bin
Cuda comes with the uninstall tool uninstallCudatoolkit_7.0.pl, using the command:
shell sudo ./uninstall_cuda_toolkit_7.0.pl
Uninstall the original NVIDIA driver:
shell sudo apt-get remove nvidia*
Uninstalling Nvidia in this way can cause the graphical interface to become thicker and larger (you know, hehe). Another way to unload is to be gentle:
shell sudo apt-get remove --purge nvidia-*
sudo apt-get install ubuntu-desktop
sudo rm /etc/X11/xorg.conf echo ‘nouveau‘ | sudo tee -a /etc/modules
1.2 If it's not as troublesome, re-install the system, save time and effort
How to install, please self-search ha, do not repeat.
2. Install driver 2.1 Download graphics driver
Http://www.nvidia.com/Download/index.aspx?lang=en-us
My test card is relatively low, the gods do not spray, a look, lying trough, but also with the 200M series, too rubbish. Cough, um, I'm doing a record of the configuration process. (Smiling face), after a picture is downloaded, it should be a file like this Nvidia-linux-x86_64-340.104.run
-340.104 version number, generally use XXX to express, your xxx and my, other people may not be the same, so do not side.
2.2 Turn off Ubuntu's graphical interface
On the command line, enter
sudo service lightdm stop
2.3 Add executable permissions to XXX
chmod +x NVIDIA-Linux-x86_64-340.104.run
Note XXX may not be the same yo, do not copy directly
2.4 Execution
./NVIDIA-Linux-x86_64-340.104.run
When installing, do not install 32bit Library, only this one chooses the right that no, is 32 bit of that Kochhar, do not choose, the rest chooses left.
2.5 After installing the driver, test
nvidia-smi
As shown, it proves that you have installed the driver successfully:
Do you find that there is a not supported, no side, the master told you no matter.
2.6 Update Grub
sudo echo ‘GRUB_CMDLINE_LINUX="nomodeset"‘ >> /etc/default/grub
sudo update-grub sudo nvidia-xconfig -a --cool-bits=28 --allow-empty-initial-configuration
2.7 reboot
Open graphical interface sudo service lightdm start
not much to say, do not want to see the beauty of mosaic can not open
3 Installation cuda3.1 Of course, it was downloaded first, which is also used to say
Come on, the website is here, I am under the cuda6.5, test success, interested or the graphics card, you can choose a high version. Https://developer.nvidia.com/cuda-downloads Choose the right version oh, don't choose too high, unless you buy a new tens of thousands of new graphics card.
3.2 Installation
My Cuda version name is such a cuda6.5.14linuxaarch64native.run so you follow your own version of the download to compare.
./cuda_6.5.14_linux_aarch64_native.run
Skip the front to see, you can press the space, you can also CTRL + C, according to the prompt, all agree on the good.
3.3 Adding environment variables
sudo usermod -a -G video $USER
echo "" >> ~/.bashrc
echo "export PATH=/usr/local/cuda-6.5/bin:$PATH" >> ~/.bashrc
echo "export LD_LIBRARY_PATH=/usr/local/cuda-6.5/lib64:$LD_LIBRARY_PATH" >> ~/.bashrc
3.4 and then restart
sudo shutdown -r now
3.5 gcc downgrade (quite important)
The GCC version of 16,04 may be too high, not suitable for cuda6.5 or nearby versions, so lower the GCC version level
3.5.1, of course, is to download gcc.
The gcc--4.7 version is downloaded here.
sudo apt-get install gcc-4.7 g++-4.7
3.5.2 Changing to Cuda's path
cd /usr/local/cuda/bin
The path may be different, also do not copy said, find the Cuda and cuda-6.5 place is good
3.5.3 in the bin file, create a new GCC folder and connect the relevant GCC
sudo su
mkdir gcc
cd gcc
ln -s /usr/bin/gcc-4.4 gcc
ln -s /usr/bin/cpp-4.4 cpp
ln -s /usr/bin/g++-4.4 g++
exit
Note Yo, enter exit, child Ah, the current directory ha, don't get the wrong directory, should still be in the/usr/local/cuda/bin directory
3.5.4 increasing the variation path of NVCC
sudo su
echo "compiler-bindir = /opt/cuda/bin/gcc" >> nvcc.profile
exit
3.5.5
and reboot.
sudo reboot
4 Here's the end, let's start the test.
There is a devicequery example under the catalogue/usr/local/cuda/samples/1_utilities/devicequery, just test it, and you'll succeed.
Of course, you can also test other examples, such as this path under/usr/local/cuda/samples/3_imaging/boxfilter
In this example there is a woman, a will be very beautiful, a will with a dozen mosaic as vague, probably can learn, understand, haha
All right, Belle, it's time to do something, compile the GPU version of GEANT4.
(Stretch, good sleepy) tomorrow in the mutation it, yuck, wrong, this morning in the compilation bar, sleep ~ ~
Ubuntu16.04+nvidia Gt240m+cuda 6.5 configuration