[한국어 가이드]

Quick Set Up NVIDIA GPUs with GCP VM

• OS: Ubuntu 16.04
• GPU: NVIDIA K80, P100, V100
  • CUDA: 9.0.176
  • NCCL: 2.2.13
  • CUDNN: 7.1.4.18
• Python: 3.5.2
  • TensorFlow: 1.12.0
  • PyTorch: 1.0.0
  • Other ML libraries
• Programs:
  • Anaconda: 5.3.1, Python: 3.7.1 (is not default)
  • nvidia-docker: 2.0.0

How to use

• Run gcp_setup_ubuntu1604.sh
• Before you run this file, please check your OS Ubuntu 16.04.
• You can modify this code if you want to use other OS or CUDA version.

wget https://raw.githubusercontent.com/GzuPark/gcp-ubuntu-gpu/master/gcp_setup_ubuntu1604.sh
bash gcp_setup_ubuntu1604.sh

Etc

Anaconda

• If you want to change from default python interpreter to Anaconda:

sudo bash
ln -s -f /usr/bin/anaconda3/bin/python3 /usr/bin/python
ln -s -f /usr/bin/anaconda3/bin/pip /usr/bin/pip
ln -s -f /usr/bin/anaconda3/bin/jupyter /usr/bin/jupyter

Jupyter

• First, run this command:

jupyter notebook --allow-root

• And then, check your EXTERNAL_IP of a instance, enter like below on your web-browser (ex: 10.100.0.8)

10.100.0.8:8888

Issues

Please comment at the Issues tap.