moskomule / cuda-server2

Renewed cuda-server for my research

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

cuda-server2

Docker Automated build

Renewed cuda-server for my research. cuda-server2 has the latest PyTorch, ipython etc., as well as conda environment and linuxbrew, by default.

Note that cuda-server2 may not be fully secure, so use it with care in a secure environment with users whom you can trust.

Requirements

Check nvidia-docker page.

Make sure you have installed the NVIDIA driver and Docker 19.03 for your Linux distribution.

Usage

Setup

# chmod 777 run.sh
./run.sh CONTAINER_NAME CUDA_VERSION USER_OPTIONS

where

  • CONTAINER_NAME is any name you like
  • CUDA_VERSION is currently 92, 101 or 102
  • USER_OPTIONS includes
    • PORTS e.g., -p 30022:22. Do not forget to specify at least one port for port 22
    • VOLUMES e.g., -v $HOME/Download:/home/user/Download
    • GPU e.g., all (default), "'devices=0,1'". This option corresponds --gpus of nvidia-docker

Access

ssh -p XXXXX:22 CONTAINER_NAME@localhost

The default password is Docker!.

About

Renewed cuda-server for my research


Languages

Language:Dockerfile 81.9%Language:Shell 18.1%