-
Download directly from the Nvidia website [Nvi] according to the GPU processor.http://www.nvidia.de/Download/index.aspx
-
Then go to boot menu and disable the secured boot then in terminal
$ chmod +x Nvidia driver $ sudo ./nvidia driver
$ sudo apt-get purge nvidia*
$ sudo add-apt-repository ppa:graphics-drivers/ppa
$ sudo apt-get update
$ sudo apt-get install nvidia-driver-440
After this reboot the pc then open terminal and type nvidia-smi
a display in the terminal with driver version ill be displayed
Create a file at
sudo nano /etc/modprobe.d/blacklist-nouveau.conf
copy this and paste in it:
blacklist nouveau options nouveau modeset=0
$ sudo update-initramfs -u
$ sudo reboot
https://developer.nvidia.com/cuda-toolkit-archive
$ sudo dpkg -i cuda-repo-<distro>_<version>_<architecture>.deb
$ sudo apt-key add /var/cuda-repo-<version>/7fa2af80.pub
$ sudo apt-get update
$ sudo apt-get install cuda
1.Type nano ~/.bashrc in terminal
2.paste the required path the below
#cuda
export PATH=/usr/local/cuda-10.2/bin:/usr/local/cuda-10.1/NsightCompute-2019.1${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-10.1/lib\${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
3.After pasting type source ~/.bashrc
Copy the samples from cuda(usr/local/cuda)
paste in home directory
go to device query
make clean
make
./devicequery
result:Pass
After this type nvcc --version in terminal you will get the cuda version if this displayes success
The order of versions has to be installed:
$ sudo dpkg -i libcudnn7 7.6.2.11-1+cuda9.0 amd64.deb
$ sudo dpkg -i libcudnn7-dev 7.6.2.11-1+cuda9.0 amd64.deb
$ sudo dpkg -i libcudnn7-doc 7.6.2.11-1+cuda9.0 amd64.deb
$ cp -r /usr/src/cudnn_samples_v7/ $HOME
$ cd $HOME/cudnn_samples_v7/mnistCUDNN
$ make clean && make
$ ./mnistCUDNN
$sudo pip install tensorflow(CPU support)
$sudo pip install tensorflow-gpu(GPU support)
https://github.com/arunodhayan/HaarCascade-Trained-to-detect-my-Watch-/blob/master/Install_opencv.md