Hardware and software setup of my GPU box

Notes on the GPU box built on 2017.

Hardware

Tips:

• RAM should be 2x the size of GPU memory
• Motherboard and case need to be the same type (e.g., Mini ITX)

My part list on pcpartpicker.com:

• CPU: Intel — Core i5–6500 3.2GHz Quad-Core Processor
• Motherboard: MSI — B250I GAMING PRO AC Mini ITX LGA1151 Motherboard
• Memory: Corsair — Vengeance LPX 16GB (2 x 8GB) DDR4–3000 Memory
• Storage: Seagate — Barracuda 3TB 3.5" 7200RPM Internal Hard Drive
• Video Card: EVGA — GeForce GTX 1060 6GB 6GB GAMING Video Card ($309.00 @ Adorama)
• Case: Thermaltake — Core V1 Mini ITX Tower Case
• Power Supply: Corsair — CSM 550W 80+ Gold Certified Semi-Modular ATX Power Supply

References:

• http://timdettmers.com/2017/04/09/which-gpu-for-deep-learning/
• https://youtu.be/cRjPVN3oo4s?t=1981
• https://blog.slavv.com/the-1700-great-deep-learning-box-assembly-setup-and-benchmarks-148c5ebe6415
• http://forums.fast.ai/t/making-your-own-server/174
• https://medium.com/towards-data-science/building-your-own-deep-learning-box-47b918aea1eb
• https://www.oreilly.com/learning/build-a-super-fast-deep-learning-machine-for-under-1000
• https://pcpartpicker.com/
• https://superuser.com/questions/1186150/gpu-memory-bandwidth-vs-speed
• http://graphific.github.io/posts/building-a-deep-learning-dream-machine/
• https://medium.com/@timcamber/deep-learning-pc-build-5cffa71ad97
• https://medium.com/towards-data-science/build-a-deep-learning-rig-for-800-4434e21a424f

Software

OS

Create a bootable Ubuntu USB stick following this reference: https://tutorials.ubuntu.com/tutorial/tutorial-create-a-usb-stick-on-macos

GPU Driver

Install cuda driver before installing graphic card:

wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-repo-ubuntu1604_9.0.176-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1604_9.0.176-1_amd64.deb
sudo apt-key adv --fetch-keys http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/7fa2af80.pub
sudo apt-get update
sudo apt-get install cuda

Add to ~/.bash_profile:

export PATH=/usr/local/cuda-9.0/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
export LIBRARY_PATH=/usr/local/cuda-9.0/lib64${LIBRARY_PATH:+:${LIBRARY_PATH}}

References:

• https://medium.com/towards-data-science/building-your-own-deep-learning-box-47b918aea1eb
• https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1604&target_type=debnetwork
• https://blog.slavv.com/the-1700-great-deep-learning-box-assembly-setup-and-benchmarks-148c5ebe6415
• https://github.com/fastai/courses/blob/master/setup/install-gpu.sh

Setup ssh

Find out default gateway using route -n. For example 192.168.1.1. Go to this IP in your browser, enter username and password (e.g., admin, admin).

Find out inet addr using ifconfig. For example 192.168.1.114, enable Port 22 for this IP. Verify that the port is open using http://www.portchecktool.com/.

Get a host name from https://www.noip.com/.

sudo apt-get install openssh-server
sudo service ssh status

You can now ssh to the box using ssh user_dl_box@hostname and the password of the box, and can ssh into its running Jupyter notebook:

ssh -NL 8887:localhost:8887 user_dl_box@hostname

References:

• https://askubuntu.com/questions/140236/can-i-access-my-home-pc-from-the-office-with-ssh
• https://www.pcworld.com/article/244314/how_to_forward_ports_on_your_router.html
• http://www.noip.com/support/knowledgebase/installing-the-linux-dynamic-update-client-on-ubuntu/

Setup Docker

See here for some tips on setting up docker for deep learning: https://github.com/yang-zhang/docker-setup