Main references:
https://github.com/lbzhang/dl-setuphttp://ouxinyu.github.io/Blogs/20151108001.htmlhttp://www.cnblogs.com/ Sandykid/p/5397554.html these three are relatively new, integrated to complete the HP Z820 on the Caffe environment configuration Environment: Ubuntu 14.04 Desktop, CUDA 7.5Basics is mainly part of the preparation work
sudo apt-get updatesudo apt-get upgradesudoinstall build- Essentialsudo apt-get autoremove
Install Git
sudo Install git
Nvidia DriversDescription
- Now the graphics driver installation, CUDA installation does not need to close the graphical interface can also be achieved;
- Nor, as many bloggers have said, can not be updated before the Cuda, otherwise it may crash or something;
- The installation of Nvidia drivers can be implemented via additional drivers in the "Software and updates" that comes with Ubuntu;
- or in accordance with the method provided in the Web site to achieve driver installation;
- Another argument is now Cuda Integrated drive, in the case of turning off the graphical interface to install Cuda can automatically put the driver also installed, this may be through the online installation, incidentally installed dependency implementation, the offline installation package may be small;
- Recommended 352, not the latest 361.42, the newest is not necessarily the most suitable
To view the video card model:
grep -I. nvidia
Online Installation:
sudo add-apt-repository ppa:graphics-drivers/ppasudo apt-get updatesudo Install nvidia-352
or download the website from the installation package, install
Or, as I said, using Ubuntu's own software and choosing the right driver for the new option is easier; reboot the system:
sudo shutdown-r now
To view the driver version:
Cat /proc/driver/nvidia/version
CUDAAlso download the offline installation package, which uses CUDA version 7.5:
sudo dpkg-i cuda-repo-ubuntu1404*amd64.debsudo apt-get updatesudo Install Cuda
To add a path:
Echo ' export Path=/usr/local/cuda/bin: $PATH ' >> ~/bashrcecho'export ld_library_path=/usr/local/cuda/lib64: $LD _library_path' >> ~/~/.BASHRC
To check the Cuda version:
Nvcc-v
Restart the system:
sudo shutdown-r now
CuDNNThe latest version is
CuDNN V5, but after trying,
V5In the process of compiling a problem, recommend
CuDNN V4 for LinuxJump to the path where CUDNN is located, unzip the installation:
CD ~/downloads/tar xvf cudnn*tgzcd cudasudocp * /* * * libcudnn*/usr/local/cuda/lib64/sudochmod a+r/usr/local/cuda/lib64/libcudnn*
Here is actually copying the relevant files to the Cuda directory, it is strange why Cuda does not integrate CUDNN, it may have been self-brought, did not verify;
Link CUDNN library file, I wonder if it must:
sudo Ln -sf/usr/local/lib/libcudnn.so. 4.0. 7 /usr/local/lib/libcudnn.so. 4 sudo Ln -sf/usr/local/lib/libcudnn.so. 4 /usr/local/lib/libcudnn.so
Check:
Nvidia-smi
OpenblasCUDNN and Openblas only install one to complete the Caffe compilation, optional operation
mkdir ~/~/gitgit clone https://github.com/xianyi/openblas.gitCD Openblasmake1)]sudomakeinstall
To add a path:
Echo ' export ld_library_path=/usr/local/lib: $LD _library_path ' >> ~/.BASHRC
Common Tools
- For related python dependencies, you can refer to the given URL attempt, or you can install Anaconda2 one step directly, just pay attention to adding additional environment variables:
- Of course also can not install Anaconda2 direct installation;
- Run for a minute and check to make sure the dependencies are there;
sudo Install -y libfreetype6-devlibpng12-Install -u matplotlib ipython[all] Jupyter Pandas Scikit-image
There are also some caffe dependency packages:
sudo Install libprotobuf-dev libleveldb-dev libsnappy-dev libopencv-dev libhdf5-serial-dev protobuf-Compiler sudoinstall --no-install-recommends libboost-all-devsudo Install Libgflags-dev Libgoogle-glog-dev Liblmdb-dev
OpenCV
Recommended installation, often need to use: Download the installation package, switch to the File save folder, install dependencies:
sudo SH Ubuntu/dependencies. SH
Enter Ubuntu\3.0\ install OPENCV 3.0.0:
sudo sh opencv3_0_0. SH
If a successful installation should not have any error prompts:
Use a useful command to view the version:
pkg-config --modversion OpenCV
CaffeTo download the installation package:
CD ~/gitgit clone https://github.com/bvlc/caffe.gitCD CaffeCP Makefile.config.example Makefile.config
The main is to create a new git folder, download the Caffe installation package into the GIT path:
The last command is to copy a configuration file, the source file is left as a backup, we want to modify this configuration file, open and modify;
sudo gedit/caffe/makefile.configuse_cudnn:1include_dirs:= $ (python_include)/ usr/local/include/usr/lib/x86_64-linux-gnu/hdf5/serial/includelibrary_dirs:= $ (python_lib)/usr/local/ lib/usr/lib/usr/lib/x86_64-linux-gnu/hdf5/serial# enabled by default openblaspython_lib:=/usr/local/Lib # PYTHN Library Matlab_dir:=/usr/local/matlab/=3 # OpenCV
Just list the modified section, if you use the Anaconda configuration file has related changes, only need to uncomment it, and modify the path for their own anaconda installation path;
Compile Caffe:
make all –j16 make test –j16 make Runtest-j16
Adding "-16" is faster for compiling with multi-core;
Compile pythn and Matlab for Caffe
make Pycaffe-j16 make Matcaffe-j16
The test is described at the end of the second URL:)
A handy Ubuntu+caffe configuration note