Understand this story
On November 7, 2026, AWS Machine Learning announced that Amazon SageMaker Inference now supports GPU partitioning based on NVIDIA Multi-Instance GPU (MIG). This capability can split a single physical GPU into up to 7 hardware-isolated partitions, each with its own compute units and memory, thereby enabling deployment of more models than the number of GPUs on a single instance. This capability applies only to endpoints that use inference components; single-model endpoints, serverless endpoints, and heterogeneous instance pools are not supported, and the partition layout must be selected when the endpoint is created. Salesforce participated in requirements definition and initial validation, and stated that MIG helps consolidate inference workloads, improve GPU utilization, and reduce inference costs. There are currently no further developments.
AI-generated from public reports
Latest update · Discussion
Unite.AI joined this discussionDevelopments
1 confirmed developments
- Source:AWS Machine Learning
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 7
Unite.AIRelated discussion
Unite.AI joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
AWS Machine LearningFirst-party
SageMaker Inference supports NVIDIA MIG, partitioning a single GPU across multiple model endpoints
Amazon SageMaker Inference now supports GPU partitioning based on NVIDIA Multi-Instance GPU (MIG), which can split a single physical GPU into up to 7 hardware-isolated partitions, each with its own compute units and memory, thereby enabling deployment of more models than the number of GPUs on a single instance. This capability applies only to endpoints that use inference components; single-model endpoints, serverless endpoints, and heterogeneous instance pools are not supported; the partition layout must be selected when creating the endpoint. Salesforce participated in requirements definition and initial validation, stating that MIG helps consolidate inference workloads, improve GPU utilization, and reduce inference costs.