Amazon Elastic Container Service (ECS) integrated support for NVIDIA Multi-Process Service (MPS), allowing multiple ECS tasks to share a single NVIDIA GPU concurrently. This advancement boosts GPU utilization and reduces infrastructure costs, especially for workloads such as small-scale inference models, AV/ADAS simulations, and multi-model serving that do not fully utilize the GPU.
Users specify GPU memory and optionally maximum GPU compute shares per container, enabling ECS to pack tasks onto GPUs with available capacity without code changes. The MPS control daemon is monitored for health on managed instances, with ECS automatically replacing unhealthy instances to avoid manual intervention. This support was made available in all AWS commercial and GovCloud Regions.
Source & further reading
Amazon ECS now supports GPU sharing with NVIDIA Multi-Process Service (MPS) ↗Source published 2026-10-27. This report was prepared with AI assistance from the linked primary-source material. Claims and limitations should be checked against the original announcement or research.