Amazon SageMaker AI inference now supports GPU partitioning for real-time endpoints

Summary

• SageMaker AI inference now supports GPU partitioning via NVIDIA MIG, enabling multiple models to share a single GPU.
• Up to seven isolated partitions per GPU provide hardware-enforced isolation, improving utilization for small and mid-size models.
• Feature is configurable when creating an endpoint on P4d, P5, G7, and G7e instances across several AWS regions, with no code changes required.