- Step1. Install Ubuntu 14.04
- Step2. sudo apt-get update && sudo apt-get install build-essential
- Step3. Install Nvidia driver(http://askubuntu.com/questions/451221/how-do-i-install-the-nvidia-driver-for-a-geforce-gt-630)
- If you want to use docker, skip to Install Tensorflow Via docker
- Step4. Install CUDA
- Step5. Edit ~/.bashrc file and add following two lines
- export PATH=/usr/local/cuda/bin:$PATH
- export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
- Step6. Install Cudnn
Using nvidia-docker, DO not need to install, CUDA and Cudnn
- Step1. Install docker: https://docs.docker.com/engine/installation/
- Step2. Install nvidia-docker: https://github.com/NVIDIA/nvidia-docker
- Step3. Install tensorflow docker: https://www.tensorflow.org/versions/r0.10/get_started/os_setup.html#docker-installation
- More Tensorflow docker tags: https://hub.docker.com/r/tensorflow/tensorflow/tags/
- Step7. apt-cache policy libcudnn5 (To verify installed Cudnn)
- http://pyrasis.com/Docker/Docker-HOWTO (Koean)
- http://www.slideshare.net/pyrasis/docker-fordummies-44424016 (Korean)
- http://m.blog.naver.com/alice_k106/220340499760 (Korean)
- sudo nvidia-docker commit ID dockername:Tags : After modifying something in a docker image, making a new docker image's tag
- sudo nvidia-docker run docker_imagename:Tags /bin/bash: run nvidia-docker
- sudo docker ps -a: See all containers ID
- Remove container: sudo docker rm ContainerID
- sudo docker images: See all Image list
- Remove Images: sudo docker rmi Imagename:Tags
- Push Image to dockerhub: sudo docker push Imagename:Tags (For first time login required,like github)
- Run docker
- Write: groupadd -r groupname -g 1000 && useradd -u 1000 -r -g username -d <home/dir> -s /sbin/nologin -c "Docker image user" username && chown -R useradd:groupname <home/dir>
- 1000:userid