neural-doodle-gpu

Here you can find a CUDA-enabled Docker container image for the project Neural Doodle.

This is a container image for rendering on a GPU. A variant of the container to render on a CPU can be found here.

Prerequisites

Hardware

• NVIDIA GPU > Fermi 2.1

Software

• Ubuntu 14.04
• Docker >= 1.9
• NVIDIA GPU drivers >= 352.39
• NVIDIA Docker >= 1.0.0 BETA 2

Optional

• CUDA Toolkit >= 7.5

Usage

To install NVIDIA CUDA with NVIDIA GPU drivers on Ubuntu 14.04

sudo ./install-cuda-drivers-ubuntu-14.04.sh

Reboot the machine after installation.

To install Docker Engine with an NVIDIA Docker wrapper and setup a Docker volume with a current version of NVIDIA GPU drivers (ensure that you have NVIDIA GPU drivers installed on the host)

sudo ./install-nvidia-docker-ubuntu-14.04.sh

Re-login after installation.

To setup a Docker volume with NVIDIA GPU drivers manually

sudo nvidia-docker volume setup

To get an image from Docker Hub

docker pull toksaitov/neural-doodle:gpu

To build the image

docker build --tag=toksaitov/neural-doodle:gpu .

To start the container with an STDIN open and an allocated pseudo-TTY, with a volume /data inside the container mounted under ./data on a host

nvidia-docker run --interactive --tty --volume="$(pwd)/data":/data/ toksaitov/neural-doodle:gpu <neural-doodle arguments>

To start with an interactive shell inside the container

nvidia-docker run --interactive --tty --volume="$(pwd)/data":/data/ --entrypoint=bash toksaitov/neural-doodle:gpu

Docker Hub

toksaitov/neural-doodle

Credits

Neural Doodle is developed by Alex J. Champandard.

The Docker image was created by Dmitrii Toksaitov.