dockerguide.md

Home | DevBox | Athena | Supercomputing@SUTD

Guide to using nvidia-docker

Note that if you choose to use docker commands instead of nvidia-docker, your work might not be running on the GPU.

To check your cuda version,

username@server:~$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2016 NVIDIA Corporation
Built on Sun_Sep__4_22:14:01_CDT_2016
Cuda compilation tools, release 8.0, V8.0.44

To check what images you have,

username@server:~$ nvidia-docker images

To run a docker image,

username@server:~$ nvidia-docker run -it --name *NAMEFORYOURDOCKER* *IMAGEID*

Note: In the naming of your docker, you might like to prefix the name with your initials e.g. my_pyconda3 so you can find your containers much easily,

username@server:~$ nvidia-docker ps -a | grep my
username@server:~$ nvidia-docker ps -a | grep my
616ae2c5df7b        pycuda                            "/bin/bash"              5 minutes ago       Exited (0) 7 seconds ago                        my_pycuda

To get some basic stuff, like pip and curl,

username@server:~$ sudo apt-get install python-pip python-dev build-essential
username@server:~$ sudo pip install --upgrade pip

---

Anaconda3

To use Anaconda3, there is a image named pyconda3. You can run a container with this image,

username@server:~$ nvidia-docker run -it --name my_pyconda3 pyconda3

In the command above, my_pyconda3 is the name that is assigned to your container(please use another name when you try) and pyconda3 is the name of the image you are using for your container. You will then get something like below,

root@d8bf528b7a96:/#

which means you are in your new container. You can check which Python3 distribution you are currently on now,

root@d8bf528b7a96:~# which python
/usr/bin/python

which is not the Anaconda3 distribution. Anaconda3 is located at /root/anaconda3 and to activate it, simply

root@d8bf528b7a96:/# cd /root/anaconda3
root@d8bf528b7a96:~/anaconda3# source bin/activate ~/anaconda3/
(/root/anaconda3/) root@d8bf528b7a96:~/anaconda3#

You are using Anaconda3 now, which can be easily checked with

(/root/anaconda3/) root@d8bf528b7a96:~/anaconda3# which python
/root/anaconda3/bin/python

To create specialized conda environments, refer to the conda documentation

To exit the Anaconda3 environment, simply

(/root/anaconda3/) root@d8bf528b7a96:~/anaconda3# source deactivate
root@d8bf528b7a96:~/anaconda3#

To exit you container without exiting the shell so your processes in your container are still running, use the escape sequence Ctrl + p, Ctrl + q
will help you turn interactive mode into daemon mode

To exit your container,

root@d8bf528b7a96:~/anaconda3# exit
username@server:~$

If you want to re-enter the container,

nvidia-docker restart my_pyconda3
nvidia-docker attach my_pyconda3

restart is to get the container to start running and attach is to enter the container.

If you are done with your container, meaning you do not need it anymore,

username@server:~$ nvidia-docker stop my_pyconda3
username@server:~$ nvidia-docker rm my_pyconda3

where stop will stop the container and rm will permanently delete your container. Only do that your your containers only!

Listing containers

To see the list of containers that are still running,

username@server:~$ nvidia-docker ps

To see the list of containers inhibiting the server,

username@server:~$ nvidia-docker ps --all

To see the list of containers inhibiting the server with a particular keyword,

username@server:~$ nvidia-docker ps -a | grep *keyword*

List of images

Refer to the list of images here.