Posted On: Oct 15, 2021
Amazon Elastic Container Services (ECS) now enables customers to manage containerized GPU-based workloads running on-premises infrastructure using Amazon ECS Anywhere. With Amazon ECS Anywhere GPU support, customers can deploy GPU based applications that need to remain on premise due to regulatory, network latency, data residency, or other requirements. Additionally, enterprises can make use of their existing investment on GPU compute capacity to run machine learning, 3D visualization, image processing and big data workloads without the need to transfer data to the cloud.
Amazon ECS Anywhere, launched in June 2021, is a capability in Amazon ECS that enables customers to more easily run and manage container-based applications on-premises, including virtual machines (VMs), bare metal servers, and other customer-managed infrastructure. With this release, customers can add GPU instances by adding the --enable-gpu flag to the Amazon ECS Anywhere installation script. Once the script is installed, customers will be able to assign a number of GPUs to particular containers in their task definition. Amazon ECS uses this as a scheduling mechanism to pin physical GPUs to the desired containers for workload isolation and optimal performance.
Customers can use Nvidia and CUDA drivers with Amazon ECS Anywhere by following the steps to install the drivers as provided here. To learn more, read our blog or check out our documentation. This feature is now available in all public Amazon Web Services Regions, including Amazon Web Services China (Beijing) Region, operated by Sinnet and Amazon Web Services China (Ningxia) Region, operated by NWCD.