ubuntu16.04 + Cuda10.0 + Cudnn7.4 + tfgpu1.15

Cuda and Cudnn install

Version

ubuntu 16.04 x86_64
TF: 1.15.0-gpu
Cuda:9.0
cudnn 7.5
tf 1.15 + cuda 10.2 + cudnn 7.6 may also be fine.

0. Nvidia Driver install(Not Necessary)

Press " ctrl+alt+f1 " to enter command mode.

* Forbidden nouveau

$blacklist nouveau
$options nouveau modeset=0
$update-initramfs -u
$reboot
$lsmod | grep nouveau

* shut down x server.

$su root // Cause I want to install for all users.
$/etc/init.d/lightdm stop // Stop Server
$sudo /etc/init.d/lightdm status // Only for check.

• if (you want to restart) {sudo /etc/init.d/lightdm restart}

* uninstall nvidia driver, cuda,

$ sudo apt-get --purge remove nvidia*
$ sudo apt autoremove
To remove CUDA Toolkit:
$ sudo apt-get --purge remove “cublas” “cuda*”
To remove NVIDIA Drivers:
$ sudo apt-get --purge remove “nvidia”
$ etc/init.d/lightm status (Make sure it’s started)
$ reboot
$ su root

1 Download and install Cuda 10.0

1.0 install dependency

root@$ apt-get install freeglut3-dev build-essential libx11-dev libxmu-dev libxi-dev libgl1-mesa-glx libglu1-mesa libglu1-mesa-dev"
– for kernel problem only.
root@$ apt install dkms

1.1

wget root@$ : https://developer.nvidia.com/compute/cuda/10.0/Prod/local_installers/cuda_10.0.130_410.48_linux
root@$ :sh cuda_10.0.130_410.48_linux --no-opengl-files
#open cl will cause more prolems.
root@$ : sh cuda_10.0.130_410.48_linux

---

1.2 if you get this

summery
Driver: installed
Toolkit: installed
Samples: Installed in /root, but missing recommend libraries
To uninstall CUDA Toolkit, run the uninstall script in /usr/local/cuda-9.0/bin
To uninstall the NVIDIA Driver, run nvidia-unistall
Please see CUDA_Installation_Guide_linux.pdf in /usr/local/cuda-9.0/doc/pdf for detailed information on setting up CUDA.
log file is /tmp/cuda_install_31840.log

Please make sure that

• PATH includes /uer/local/cuda-10.0
• LD_LIBRARY_PATH includes /usr/local/cuda-10.0/lib64 or add /user/loca/cuda-9.0/lib64 to /etc/ld.so.conf and run ldconfig as root.

---

2 install cudnn

wget https://developer.nvidia.com/compute/machine-learning/cudnn/secure/v7.4.2/prod/10.0_20181213/cudnn-10.0-linux-x64-v7.4.2.24.tgz
tar -xzvf cudnn-10.0-linux-x64-v7.4.2.24.tgz
cp cuda/include/cudnn.h /usr/local/cuda/include
cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
chmod a+r /usr/local/cuda

1.3

3.Install Tensorflow_gpu

root@$ conda install tensorflow_gpu==1.15

4.Check NVIDIA-smi, Cuda, cuDNN

• driver
  root@$ nvidia-smi
  If error.
  $ apt install nvidia-cuda-toolkit
  $ nvidia-smi
• cuda version
  $ echo “CUDA Version”
  $ cat /usr/local/cuda/version.txt
  $ cat /proc/driver/nvidia/version
• cudnn version.
  $ echo “cudnn Version”
  $ cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2
• NVCC
  $ echo “NVCC INFO”
  $ which nvcc
  $ nvcc --version

vi ~/.bashrc

• #if no results
  $ export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib.64:$LD_LIBRARY_PATH
• #For primer targets. If here is no lib, then serarch cuda/lib
  export LD_LIBRARY_PATH=/usr/local/cuda/lib:$LD_LIBRARY_PATH
  $ export PATH=$PATH:/usr/local/cuda/bin

5.Test samples(Path may change)

• Check cuda.
  root@$ cd /usr/local/cuda-10.0/samples/1_Utilities/deviceQuery/
  root@$ make
  root@$ ./deviceQuery
• if print cuda information. Passed！