ubuntu-for-robotics

News:

• October 25: Added Support for Tensorflow v1.11, CUDA v9.0 and cudnn v7.3

About:

This repository is for setting-up cuda-9/8, nvidia-396/387/384 driver, OpenCV-3.3, ROS Kinetic, Tensorflow-1.11/1.7/1.4/1.2.1, Pytorch-0.4; maintained for settting-up the step-zero Deep Learning framework (Tensorflow), especially for Robotics application.

Install fresh ubuntu 16.04 (x86-64) and follow: (Tested on Intel i7-6700K, 32GB DDR4 RAM, nvidia Titan X)

Order of Installation:

• Cuda
• OpenCV
• ROS
• Tensorflow
• CuDNN

First do:

sudo apt update
sudo apt upgrade

nvidia-384 setup:

sudo apt install nvidia-smi nvidia-384

Cuda 9-0 setup:

• Download Cuda for Ubuntu 16.04 x86_64: (deb-local) or simply do,

wget https://developer.nvidia.com/compute/cuda/9.0/Prod/local_installers/cuda-repo-ubuntu1604-9-0-local_9.0.176-1_amd64-deb

• Do:

sudo dpkg -i cuda-repo-ubuntu1604-9-0-local_9.0.176-1_amd64.deb 
sudo apt-get update
sudo apt-get install cuda

Cuda 8-0 setup:

• Download Cuda for Ubuntu 16.04 x86_64: (deb-local) https://developer.nvidia.com/cuda-80-ga2-download-archive or simply do, wget https://developer.nvidia.com/compute/cuda/8.0/Prod2/local_installers/cuda-repo-ubuntu1604-8-0-local-ga2_8.0.61-1_amd64-deb
• Do:

sudo dpkg -i cuda-repo-ubuntu1604-8-0-local-ga2_8.0.61-1_amd64.deb
sudo apt-get update
sudo apt-get install cuda

• Check for cuda version:

sudo apt install nvidia-cuda-toolkit
nvcc --version

• If you see Cuda compilation tools, release 7.5 instead of cuda-8.0, do: echo 'export PATH=/usr/local/cuda-8.0/bin${PATH:+:${PATH}}' >> ~/.bashrc
• Check nvcc --version again (on a fresh terminal window). It should show you cuda-8.0.

Install OpenCV-3.3:

bash install_opencv.sh

Install ROS-Kinetic-base:

bash install_ros.sh

Install Tensorflow:

bash install_tf.sh

Install CuDNN:

bash install_cudnn.sh

FAQ(s):

• Installing cuda for Quadro series GPU