{{tag>Drivers Matériel Privateur}}

= Install drivers graphique propriétaire Nvidia CUDA

**Procédure d'installation de [[https://www.gnu.org/proprietary/proprietary.fr.html|logicielle privateur]]**

Voir :
* https://wiki.debian.org/fr/NvidiaGraphicsDrivers
* http://developer.download.nvidia.com/compute/cuda/6_5/rel/docs/CUDA_Getting_Started_Linux.pdf
* [[https://gist.github.com/wangruohui/bc7b9f424e3d5deb0c0b8bba990b1bc5|Intel for display, NVIDIA for computing]]

Voir aussi :
* [[https://www.easy-tensorflow.com|Easy TensorFlow]]
* [[https://developer.nvidia.com/cudnn|NVIDIA CUDA® Deep Neural Network library (cuDNN)]]
* [[https://devblogs.nvidia.com/parallelforall/increase-performance-gpu-boost-k80-autoboost/|GPU Boost]]

Pour CUDA voir OpenCL


== glxinfo

<code bash>
glxinfo
</code>

Les lignes intéressantes du glxinfo sont:
direct rendering: Yes
OpenGL vendor string: nouveau

== Install Drivers + CUDA8 sur Debian Jessie (Debian 8)

Install des dépendances
<code bash>
apt-get update
apt-get install build-essential binutils g++ mpich2 libxmu-dev libxi-dev linux-headers-$(uname -r)
</code>

Install de Cuda
<code bash>
cp *.run /root/
cd /root/
./cuda_8.0.44_linux.run
</code>

<code bash>
export PATH=$PATH:/usr/local/cuda/bin/
echo "export PATH=$PATH:/usr/local/cuda/bin" >> ~/.bashrc

#export LD_LIBRARY_PATH=/usr/local/cuda/lib64/:$LD_LIBRARY_PATH
echo "/usr/local/cuda/lib64" > /etc/ld.so.conf.d/cuda.conf
ldconfig
</code>

Redémarrage, car le drivers Libre est encore chargé
<code bash>
lsmod |grep nouveau
reboot
</code>

Le driver Libre **nouveau** n'est plus chargé
<code bash>
lsmod |grep nouveau
</code>

Grâce à ''/etc/modprobe.d/nvidia-installer-disable-nouveau.conf''

''/etc/modprobe.d/nvidia-installer-disable-nouveau.conf''
<code ->
# generated by nvidia-installer
blacklist nouveau
options nouveau modeset=0
</code>

<code bash>
sudo update-initramfs -u
</code>

Install des derniers drivers
<code bash>
systemctl stop gdm
./NVIDIA-Linux-x86_64-375.20.run
</code>

On redémarre
<code bash>
reboot
</code>


== Install de Cuda Docker

l faut désinstaller tous les paquets Debian cuda et nvidia :

<code bash>
#apt-get purge $(dpkg -l |egrep 'cuda|nvidia' |grep ^ii |awk '{print $2}')
apt purge 'nvidia-*'
</code>

Install de CUDA
<code bash>
./cuda_9.2.88_396.26_linux.run
./cuda_9.2.88.1_linux.run # Patch nvidia
</code>

<code bash>
echo 'export PATH=$PATH:/usr/local/cuda/bin' |sudo tee /etc/profile.d/cuda.sh
echo "/usr/local/cuda/lib64" |sudo tee /etc/ld.so.conf.d/cuda.conf
sudo ldconfig
</code>

''/etc/apt/sources.list.d/nvidia-docker.list''
<code ->
deb https://nvidia.github.io/libnvidia-container/debian9/$(ARCH) /
deb https://nvidia.github.io/nvidia-container-runtime/debian9/$(ARCH) /
deb https://nvidia.github.io/nvidia-docker/debian9/$(ARCH) /
</code>

Installer nvidia-docker2
<code bash>
sudo apt-get update
sudo apt-get install nvidia-docker2
</code>

<code bash>
nvidia-docker run -it -p 8888:8888 tensorflow/tensorflow:latest-gpu
</code>


== Vérif

On vérif que le driver en chargé
<code bash>
lsmod |grep nvidia
</code>


<code bash>
nvidia-smi && echo OK
nvidia-smi topo -m
nvcc --version
</code>

<code bash>
sudo apt-get install clinfo
clinfo
</code>

<code bash>
sudo apt-get install pyrit
pyrit list_cores
</code>

<code bash>
cat /proc/driver/nvidia/version
</code>

<code ->
NVRM version: NVIDIA UNIX x86_64 Kernel Module  375.20  Tue Nov 15 16:49:10 PST 2016
GCC version:  gcc version 4.9.2 (Debian 4.9.2-10)
</code>

En root
<code bash>
cd /usr/local/cuda/samples/1_Utilities/deviceQuery
make
./deviceQuery
</code>

<code bash>
cd /usr/local/cuda/samples/1_Utilities/bandwidthTest
make
/usr/local/cuda/samples/bin/x86_64/linux/release/bandwidthTest
</code>


== cudnn

=== Install

Voir :
* https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html
* https://www.easy-tensorflow.com/tf-tutorials/install/cuda-cudnn


Unzip the cuDNN package.
<code bash>
tar -xzvf cudnn-10.2-linux-x64-v7.6.5.32.tgz
</code>

Copy the following files into the CUDA Toolkit directory, and change the file permissions.
<code bash>
sudo cp cuda/include/cudnn.h /usr/local/cuda/include
sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
</code>

Needed ?
<code bash>
sudo apt-get install libcupti-dev 
</code>

ou 

<code bash>
# the runtime library
sudo dpkg -i libcudnn7_7.0.5.15–1+cuda9.0_amd64.deb

# the developer library
sudo dpkg -i libcudnn7-dev_7.0.5.15–1+cuda9.0_amd64.deb

# the code samples
sudo dpkg -i libcudnn7-doc_7.0.5.15–1+cuda9.0_amd64.deb
</code>


==== Autres


''~/.bashrc''
<code bash>
# put the following line in the end or your .bashrc file
export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/extras/CUPTI/lib64"
</code>

<code bash>
sudo ldconfig
</code>


=== Test 

Voir : http://www.robot.t.u-tokyo.ac.jp/~miyagusuku/deeplearning,/install/2018/10/01/dlbox_install.html



<code bash>
cp -r /usr/src/cudnn_samples_v7/ ~
cd ~/cudnn_samples_v7/mnistCUDNN
make clean && make
./mnistCUDNN
</code>


== Autres

<code bash>
cd /usr/local/cuda/samples/5_Simulations/nbody
make
GLPATH=/usr/lib make
GLPATH=/usr/lib/x86_64-linux-gnu make
./nbody -benchmark -numbodies=256000 -device=0

apt-get install gcc g++ gcc-4.8 g++-4.8 gcc-4.9 g++-4.9 libxi6 libxi-dev libglu1-mesa libglu1-mesa-dev libxmu6 linux-headers-amd64 linux-source freeglut3-dev
</code>
Voir https://andrewbolster.info/2016/04/fixcuda-on-debian-jessie