GPU Instance

ubuntu based.

• tensorflow

  • https://www.tensorflow.org/install/install_linux
  • pip install --upgrade tensorflow-gpu Could not find any downloads that satisfy the requirement tensorflow-gpu.
  • I was able to install it by specifying the URL as described.
  • 1.1.0

• scikit-learn

  • sudo pip install scikit-learn

• keras

  • sudo pip install keras

• I don’t want to lose intermediate results when a spot instance dies.

  • Only one instance is mounted at a time.
    • If you keep the EBS mounted, that’s not going away.
  • If you want to mount it on multiple instances
    • EFS (but it does not work in the Tokyo Region)
    • Set up your own NFS server → easy if you enter a VPC
    • monit+rsync
    • Watch for termination notices: spot/termination-time
    • http://qiita.com/iogi/items/62f84e0ebfdc7afd9539
    • Dropbox
      • CLI does not go in well → give up or stand up X and VNC

New Amazon EC2 P2 instances are now available in the Tokyo Region ｜ Developers.IO

Running Chainer on an EC2 GPU instance with 5 lines of Chainer and 4 lines of Chainer - If you make it the Lord everywhere, it will be true everywhere. http://sla.hatenablog.com/entry/chainer_on_ec2 Quickly use Chainer with AWS https://orientalrobotics.blogspot.jp/2015/07/awschainer.html Chainer on EC2 spot instance environment with AWS Lambda - Qiita http://qiita.com/pyr_revs/items/bf483b96cca6b01a9110 How to stop an EC2 instance when a script that takes a long time to process finishes - Masteries http://papix.hatenablog.com/entry/2016/06/02/131754

This page is auto-translated from /nishio/GPUインスタンス using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.