Amazon ECS Now Supports GPU Sharing With NVIDIA Multi-Process Service (MPS)
AWS, Tuesday, October 27th, 2026
Amazon ECS now lets multiple tasks share a single NVIDIA GPU using Multi-Process Service to improve utilization and cut costs.
Amazon ECS now supports GPU sharing through NVIDIA Multi-Process Service, allowing multiple ECS tasks to run at once on a single NVIDIA GPU.
It targets workloads needing only part of a GPU, such as small-model inference and embedding generation, to raise utilization and lower costs.
Users set the GPU memory each container needs and can cap its compute share, and ECS places tasks on GPUs with spare capacity without application code changes.
The feature works on ECS Managed Instances, where ECS also monitors the MPS daemon and replaces unhealthy instances, and on ECS on EC2, in all commercial and GovCloud (US) Regions.