Hello. Need instructions and assistance on installing NVIDIA GPU docker containers on our recently purchased AMD threadripper. It’s behind a firewall with no internet connection. I assume lambda stack is pre-installed on the machine, so need to transfer files/containers via USB stick… Lambdalabs website indicate that this is possible.
-
Lambda stack is installed.
-
You also need to install other packages for docker:
Tutorial: NVIDIA NGC Tutorial: Run a PyTorch Docker Container using nvidia-container-toolkit on Ubuntu
sudo apt-get install -y docker.io nvidia-container-toolkit
sudo systemctl daemon-reload
sudo systemctl restart docker- So doing the mirror repository on-site is the likely solution so it can also keep up to date.
(or for some sites using a proxy server to allow specific sites). See below. - To just download/copy a package:
$ mkdir download
$ cd download
$ apt download docker.io nvidia-container-toolkit
$ ls
docker.io_20.10.7-0ubuntu5~20.04.2_amd64.deb
nvidia-container-toolkit_1.5.1+dfsg-0lambda0.20.04.1_amd64.deb
Then copy those or dependencies over that you need.
You can see the dependencies with:
$ apt-cache show docker.io
Depends: adduser, containerd (>= 1.2.6-0ubuntu1~), iptables,
debconf (>= 0.5) | debconf-2.0, libc6 (>= 2.8), libdevmapper1.02.1 (>= 2:1.02.97),
libseccomp2 (>= 2.1.0), libsystemd0 (>= 209~)
$ apt-cache show nvidia-container-toolkit
Depends: libnvidia-container-tools (>= 1.4.0)
- So doing the mirror repository on-site is the likely solution so it can also keep up to date.
-
For moving a docker image:
docker pull image:tag
docker save image:tag -o file.tarThen on the new machine (airgapped)
docker load -i file.tarSave those to a device that you can move to the airgapped location. In the airgapped you can use those on the local machine or on a local registry server. Setting up a local docker registry server: https://docs.docker.com/registry/deploying/
For setting up a mirrored repository:
I was able to setup a mirrored repository from the Lambda stack:
$ sudo cp /etc/apt/mirror.list /etc/apt/mirror.list-bak
$ sudo vi /etc/apt/mirror.list
$ diff /etc/apt/mirror.list /etc/apt/mirror.list-bak
12d11
< set base_path /var/www/html/ubuntu
17d15
< deb http://archive.lambdalabs.com/ubuntu focal main
… acommented out all other repositories(Ubuntu)…
$ sudo mkdir -p /var/www/html/ubuntu/var
$ sudo cp /var/spool/apt-mirror/var/postmirror.sh /var/www/html/ubuntu/var
$ sudo apt-mirror
$ find /var/www/html -name ‘*.deb’ | wc -l
311
$ du -sh /var/www/html/ubuntu
5.6G /var/www/html/ubuntu