Months ago … I setup Docker with nVidia support on pokey old computer with a PNY K620 Quadro GPU.

(These notes are somwhat old - sorry)

Host Setup

I used Ubuntu Server 18.04.1 for amd64 as the host OS. It was a fresh installation so this should be possible for anyone to follow. Everything else was downloaded as part of these instructions.

Software Setup

Pre-Install

The system reqiores some setup to build support for the functionality discussed. Since this was a totally “fresh” system, so I merely had to install GCC and the Linux headers;

$ sudo apt-get install gcc
$ sudo apt-get install linux-headers-$(uname -r)

CUDA Install

My notes state that I used “version 10” of the CUDA SDK …but … the link I saved reffers to 10.1u1? Regardless, I setup the .deb file as I was instructed;

$ sudo dpkg -i cuda-repo-ubuntu1804_10.0.130-1_amd64.deb
$ sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/7fa2af80.pub
$ sudo apt-get update
$ sudo apt-get install cuda

This now needs setting up some environment variables setup.

But how do we do that?

Linux seems to change what the right way is for this three times a decade. According to this, editing the .profile is in vouge so I added these lines to the end of the file.

export PATH=/usr/local/cuda-10.0/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

There were some instructions relating to POWER9 but I was unable to get anything to happen with them. After a reboot, nvidia-smi showed me some kerbunk that made me believe that the GPU was/is operating correctly.

Docker

For $reasons you need Docker’s .deb package. I found some setup instructions here which led to the following steps.

$ sudo apt update
$ sudo apt install apt-transport-https ca-certificates curl software-properties-common
$ curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo apt-key add -
$ sudo add-apt-repository "deb [arch=amd64] https://download.docker.com/linux/ubuntu bionic stable"
$ sudo apt update
$ apt-cache policy docker-ce
$ sudo apt install docker-ce
$ sudo systemctl status docker
$ sudo usermod -aG docker ${USER}

nVidia-Docker

Nothing complex left, I followed the instructions;

$ curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
$ distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
$ curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
$ sudo apt-get update
$ sudo apt-get install -y nvidia-docker2
$ sudo pkill -SIGHUP dockerd

Verification

I ran the suggested command; looks like it’s running the nvidia-smi inside a throwaway container based on the nvidia/cuda:9.0-base image.

$ docker run --runtime=nvidia --rm nvidia/cuda:9.0-base nvidia-smi

It dumps the wall of text below.

psxpl3@glaze:~$ docker run --runtime=nvidia --rm nvidia/cuda:9.0-base nvidia-smi
Unable to find image 'nvidia/cuda:9.0-base' locally
9.0-base: Pulling from nvidia/cuda
7b722c1070cd: Pull complete
5fbf74db61f1: Pull complete
ed41cb72e5c9: Pull complete
7ea47a67709e: Pull complete
35400734fa04: Pull complete
195acf8a5739: Pull complete
127028f911f6: Pull complete
Digest: sha256:157d05b8a9f3a26dce71c9e824d3fab769d77326f471d0143a236c37d278450d
Status: Downloaded newer image for nvidia/cuda:9.0-base
Thu Feb 14 16:34:44 2019
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 410.79       Driver Version: 410.79       CUDA Version: 10.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  Quadro K620         Off  | 00000000:03:00.0 Off |                  N/A |
| 35%   44C    P0     2W /  30W |      0MiB /  2001MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

… I still think we should be using some sort of OpenCL/GLES3/Vulkan rather than CUDA … #fiteme