Before you start

This is intended to be a demo setup to run hardware accelerated content on AWS P2 instances https://aws.amazon.com/ec2/instance-types/#g2

Requirements

These scripts must be run on a P2 EC2 box running Ubuntu 16.04 (HVM).

What is included?

Among the other packages these scripts will install

nvidia drivers v367.57 http://www.nvidia.com/download/driverResults.aspx/108586/en-us
VirtualGL v2.5.2 http://www.virtualgl.org/
TurboVNC http://www.turbovnc.org/
docker https://www.docker.com/
nvidia-docker https://github.com/NVIDIA/nvidia-docker
nvidia-docker-compose https://github.com/eywalker/nvidia-docker-compose
a barebone Mate desktop https://mate-desktop.org/

Installation

SSH into your EC2 box, once in just run the following

git clone https://github.com/plumbee/nvidia-hw-accelerated-box.git
cd nvidia-hw-accelerated-box
sudo su
source setup.sh

Keep in mind that:

The installation will reboot the machine once done.
At the startup the VNC server will start on DISPLAY :1.

VNC access

Password

VNC password is dynamically set to the first 8 characters of your EC2 instance id.

Display

VNC starts on DISPLAY :1, VirtualGL content will render offscreen and the resulting image will be copied to DISPLAY :1.

Connection

VNC receives connections from port 5900 + Display. Given that DISPLAY is :1 the port will be 5901.

Running HW accelerated content via VirtualGL

VirtualGL provides a 'vglrun' executable which enables (m)any GUI application to access the GPU when needed. The command below will run glxgears with hardware acceleration

vglrun glxgears

Running HW accelerated GUI applications using nvidia-docker

Yes, it's possible, check this out https://github.com/plumbee/nvidia-virtualgl

kukiel/nvidia-hw-accelerated-box