Amazon SageMaker AI Gets GPU Partitioning for Smarter Inference

Amazon SageMaker AI now lets you split a single GPU into up to seven isolated partitions using NVIDIA MIG technology, so you can run multiple models on one accelerator instead of wasting GPU capacity. This is live on P4d, P5, G7, and G7e instances across eight regions (US East, US West, Asia Pacific, and Europe), letting you right-size your inference workloads and cut costs without changing your application code.