Using Deep Learning AMI with Beanstalk may not be able to use some Beanstalk features like health monitoring, sqsd with worker env. In my case, I built a custom AMI from Beanstalk docker AMI and deployed the application using docker-compose with GPU enabled. Details are as follows:

1. check Beanstalk docker AMI in Beanstalk default config
2. start an EC2 instance with Beanstalk docker AMI on GPU instance type (ex: g4dn.xlarge)
3. Install Nvidia driver (required S3 access) https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/install-nvidia-driver.html

sudo yum install -y gcc kernel-devel-$(uname -r) aws s3 cp --recursive s3://nvidia-gaming/linux/latest/ . chmod +x NVIDIA-Linux-x86_64*.run sudo ./NVIDIA-Linux-x86_64*.run nvidia-smi 

4. Install NVIDIA Container Toolkit (DONOT INSTALL DOCKER - docker will be installed by Beanstalk) https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#installing-on-amazon-linux

distribution=$(. /etc/os-release;echo $ID$VERSION_ID) \ && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.repo | sudo tee /etc/yum.repos.d/nvidia-docker.repo sudo yum clean expire-cache sudo yum install nvidia-docker2 

5. Stop instance and create a AMI from that instance
6. Set your AMI ID to Beanstalk config
7. Deploy docker container with docker-compose.yaml (https://docs.docker.com/compose/gpu-support/)

services: test: image: nvidia/cuda:10.2-base command: nvidia-smi deploy: resources: reservations: devices: - driver: nvidia count: all capabilities: [gpu] 

Note: For beanstalk worker (https://docs.aws.amazon.com/elasticbeanstalk/latest/dg/create_deploy_docker.container.console.html)

if you manage your Docker environment with Docker Compose, Elastic Beanstalk assumes that you run a proxy server as a container. Therefore it defaults to None for the Proxy server setting, and Elastic Beanstalk does not provide an NGINX configuration

• Nginx still be installed on EC2 instances by EB (just disabled). If you want to use nginx proxy and generating logs for enhanced health reporting, you can use .ebextensions configuration files (https://docs.aws.amazon.com/elasticbeanstalk/latest/dg/customize-containers-ec2.html) for create config file and start service (sudo systemctl start nginx)

This answer was posted by Piotr Walczyszyn (under the CC BY-SA 3.0 license) in the original revision of the question Running GPU accelerated apps in a Docker container on AWS Elastic Beanstalk. Reposted here to conform to SO's Q&A format.

This is how I managed to run GPU accelerated apps in a Docker container on AWS Elastic Beanstalk. It took me almost two days of trial and error runs, so I hope others will not waste time having this post here.

These are the settings of my Elastic Beanstalk environment:

1. Set AMI ID to one those with the NVIDIA drivers and Docker preconfigured. I used Deep Learning AMI (Amazon Linux) Version 25.3 - ami-068d6d02d8775ec52
2. Set Instance type to one of the EC2 instances with GPUs. I used g3s.xlarge as it seamed enough for what I needed to do.
3. In your app Dockerfile add following lines: ENV NVIDIA_VISIBLE_DEVICES all ENV NVIDIA_DRIVER_CAPABILITIES compute,utility
4. In your app root folder create .ebextensions/01-nvidia-docker.config file and paste following code there: ```yaml option_settings: aws:autoscaling:launchconfiguration:

   I had to increase storage space for the docker to run my container

   BlockDeviceMappings: /dev/xvdcz=:24:true

commands: configure_docker_run: # I had to update EB startup script so it passes --runtime=nvidia when running docker command: sed -i 's/docker run -d/docker run --runtime=nvidia -d/' /opt/elasticbeanstalk/hooks/appdeploy/enact/00run.sh # Following line will be required if docker is updated to 19.1 # command: sed -i 's/docker run --gpus all -d/docker run -d/' /opt/elasticbeanstalk/hooks/appdeploy/enact/00run.sh ```

Now you can enjoy Docker container running on Elastic Beanstalk with NVIDIA GPU enabled.