es-dev-stack

An on-premises, bare-metal solution for deploying GPU-powered applications in containers

Blog Post with deployment details:

http://www.emergingstack.com/2016/01/10/Nvidia-GPU-plus-CoreOS-plus-Docker-plus-TensorFlow.html

Prerequisites

• CoreOS-compatible dedicated machine with vanilla CoreOS installed
• Current-generation Nvidia GPU (tested with TitanX)

To Build

Nvidia Drivers Installation Image

$ cd es-dev-stack/corenvidiadrivers

$ docker build -t cuda .

GPU-enabled TensorFlow Image

$ cd es-dev-stack/tflowgpu

$ docker build -t tflowgpu .

To Run

Stage 1 - Install Nvidia Drivers & Register GPU Devices (One-Time)

# docker run -it --privileged cuda

# ./mkdevs.sh

Stage 2 - TensorFlow Docker Container with mapped GPU devices

$ docker run --device /dev/nvidia0:/dev/nvidia0 --device /dev/nvidia1:/dev/nvidia1 --device /dev/nvidiactl:/dev/nvidiactl --device /dev/nvidia-uvm:/dev/nvidia-uvm -it -p 8888:8888 --privileged tflowgpu

To Test

• Open your web browser to http://{host IP}:8888 and launch the CNN.ipynb notebook
• Execute all steps to confirm
• To validate GPU is utilized, watch the statistics produced from the Nvidia-SMI tool;

$ docker exec -it {container ID} /bin/bash

From within the running container:

$ watch nvidia-smi

Credits:

This solution takes inspiration from a few community sources. Thanks to;

Nvidia driver setup via Docker - Joshua Kolden [email protected]

ConvNet demo notebook - Edward Banner [email protected]