Scripts for using GPU with NVIDIA Docker on AWS

The scripts in this repository are used to run Jupyter notebook with NVIDIA Docker on AWS.

These scripts are based on a post (in Japanese).

1. Launch EC2 instance

• ubuntu/images/hvm-ssd/ubuntu-trusty-14.04-amd64-server-20161213
• g2.2xlarge (also work on g2.8xlarge but I did not test)

2. Disable nouveau

This step is required for installing NVIDIA driver.

$ ssh -i "public_key.pem" [email protected] 'bash -s' < disable_nouveau.sh

When this script is finised, the remote machine will restart.

3. Install NVIDIA driver

$ ssh -i "public_key.pem" [email protected] 'bash -s' < install_nvidia_driver.sh

4. Install Docker

$ ssh -i "public_key.pem" [email protected] 'bash -s' < install_docker.sh

5. Install NVDIA Docker

$ ssh -i "public_key.pem" [email protected] 'bash -s' < install_nvidia_docker.sh

6. Build Docker image

$ scp -r -i "public_key.pem" docker [email protected]:~/
$ ssh -i "public_key.pem" [email protected] 'cd docker; chmod 755 *.sh'
$ ssh -i "public_key.pem" [email protected] 'bash -s' < build_docker_image.sh

7. Run notebook server

$ ssh -i "public_key.pem" [email protected]

# On remote
$ cd docker
$ ./run_docker_container.sh