APT Documentation: Local GPU Back End

We use Conda to install dependencies on Windows OS and Docker containers to provide dependencies on Linux.

Contents

• Docker setup - Linux
• Conda setup - Windows

Docker setup - Linux

• CUDA version >= 11.4
• Docker
• nvidia-docker

• Install CUDA version >= 11.4. This can be done by downloading and installing either the
  • CUDA Toolkit or
  • CUDA driver only.
• Install Docker (>=19.03) following the instructions on Docker's website. The free Community Edition is fine.
• Set up Docker so that it does not require sudo privileges as described here: Manage Docker as a non-root user.
• Install nvidia-docker.
• Download the latest pre-built APT docker container by running the following in a Linux terminal:

  docker pull bransonlabapt/apt_docker:latest

• Test. Here are some test commands you can try to check that everything is working as expected:
  • Test your Docker installation:

    docker run hello-world

  • Test your nvidia-docker installation:

     docker run --gpus all nvidia/cuda:9.0-base nvidia-smi 

To use this back end within APT, select Local (Docker) under the Track > GPU/Backend Configuration menu. You can test that everything hooks up between Docker and APT by selecting menu item Track > GPU/Backend Configuration > Test backend configuration.

Conda setup - Windows

• Anaconda Python 3
• CUDA version >= 11.4
• See <APT>\install\apt_conda_environment.yml for dependencies installed via conda.

• Install Anaconda Python 3 version for Windows. Instructions available here. Importantly, in "Advanced Options" check "Add Anaconda3 to my PATH environment variable".
• Install CUDA version >= 418.39. This can be done by downloading and installing either the
  • CUDA Toolkit or
  • CUDA driver only.
• The following is done at an Anaconda prompt (Start Menu > Anaconda3 > Anaconda Prompt):
  • Update Conda:

    conda update conda
    conda upgrade --all

  • Create the APT environment:

    conda env create --file \install\apt_conda_environment.yml

    where <APT> is the path to your APT folder.

To use this back end within APT, select Local (Conda) under the Track > GPU/Backend Configuration menu. You should test that everything hooks up between Conda and APT by selecting menu item Track > GPU/Backend Configuration > Test backend configuration.