Using Docker¶
Docker is a very useful tool to manage different projects in different CUDA environments.
Note
Before you use docker, you should at least know the basics of command-line interface (CLI). Please check out our linux_tutorial if you want to learn more about CLI.
Docker Containers¶
|
|
Docker Containers (Apps) |
Virtual Machines |
Code for port mapping of workstation & container¶
Description: create container with port mapping to workstation
nvidia-docker run -it -v /data/[user_name]:/workspace/[user_name] --shm-size=128gb
-v /etc/timezone:/etc/timezone:ro
-v /etc/localtime:/etc/localtime:ro -p 100XX-100XX:100XX-100XX
--name [container_name] pytorch/pytorch:1.9.0-cuda11.1-cudnn8-devel
Docker Images¶
Description: check the list of stored image
docker images
Note
For the docker image, you can check via Docker Hub
Useful code in docker¶
Check GPU status¶
nvidia-smi
List out active container¶
docker ps
Enter container¶
docker exec -it [container name] bash
Note
If your container is in non-active status, you need to start container before you can enter the container. You can use ‘docker ps -a’ to check your container status.
docker start [container_name]
Exit container¶
exit
Open jupyter notebook in workstation¶
jupyter notebook --ip=0.0.0.0 --port=[100XX] --allow-root --no-browser
Note
Browser to jupyter with port number: http://localhost:100XX/
Monitor GPU running progress¶
Description: can customize the updated timeframe for GPU. For example, request updated in every 0.2s.
watch -d -n 0.2 nvidia-smi