No description, website, or topics provided.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
LICENSE
README.md

README.md

nvidia-gpu-cloud-with-titannvidia-gpu-cloud-with-titan

This is my notes after 6-7 installs.

Once the server is set up the promise of a 5 minute install holds true, but you still need to know the basics of using Docker.

Ubunutu Setup:

  • Following with:

  • http://docs.nvidia.com/ngc/ngc-titan-setup-guide/index.html

  • Burn 16.04.3 Desktop AMD64 Aug 2017 ISO to DVD

  • or create a flash USB: "Startup Disk Creator" in Ubuntu Desktop

  • (They really should state the exact ISO to pull.)

  • http://releases.ubuntu.com/16.04/

  • Don’t check the boxes for “download updates or install third party software”

  • Force UEFI Installation:click “Go Back”

  • “Erase Disk and install Ubuntu”: Install Now

  • On first start up make sure 4.10 kernel is running by typing:

  • uname -a

  • you might have to edit grub to get 4.10 kernel to boot:

  •  sudo gedit /etc/default/grub
    
  •   sudo update-grub
    
  • Follow the directions at:

  • http://docs.nvidia.com/ngc/ngc-titan-setup-guide/index.html

  • Make script files by copying out of Firefox(Chrome copies wrong)

  • I keeping having to run second script more than once. If so:

  • Remove reboot at end to make sure the second script works

  • You might have to run manually:

  • sudo rm /etc/apt/sources.list.d/cuda.list

  • sudo apt-key add /var/nvidia-driver-local-repo-387.34/7fa2af80.pub

  • Then run the second script again, then reboot.

  • Run the 3rd and forth scripts.


Software to install on local machines:

Using Nvidia GPU Cloud with Titan Usage:

  • sudo mkdir /workspace
  • cd /workspace
  • sudo mkdir jobs
  • cd jobs
  • login(you need to paste in your key):
  • docker login -u '$oauthtoken' --password-stdin nvcr.io <<< 'Qt4M2VkL---------Your API KEY -------3ZGZZmNIw'
  • docker pull nvcr.io/nvidia/digits:18.04
  • nvidia-docker run -it --rm --net=host --name digits -v /data:/data -v /digits-jobs:/workspace/jobs nvcr.io/nvidia/digits:18.04
  • In a new terminal open a shell inside the workspace:
  • docker exec -it digits /bin/bash
  • The base directory for the frameworks(caffe/tensorflow/digits/etc) inside the container:
  • cd /opt/ | ls
  • To show the binaries the container has:
  • cd/usr/local/bin

Other setup :

  • Git clone required git repos
  • Move common shell scripts to: /~ and /workspace/jobs/

Docker tasks for myself

  • Write a script to start a container on system startup

Links for myself:

Useful Docker commands

  • start the docker Daemon: sudo dockerd
  • List containers:
  • sudo docker ps
  • Shell into container:
  • sudo docker exec -it digits /bin/bash
  • Use the ID from “docker ps” or name the container when you run it.