Optimizing GPU Utilization: Understanding MIG and MPS

Efficiently sharing GPUs between multiple processes and workloads in production environments is critical — but how? What options exist, what decisions need to be made, and what do we need to understand to make them? We'll explore two technologies NVIDIA makes available for GPU sharing: CUDA Multi-Process Service (MPS) and the muli-instance GPU (MIG) technology introduced with the NVIDIA Ampere architecture. Learn how these services interact, what you can do in each, and how to apply both in different scenarios to efficiently share GPUs with current best practices.