CSE Community Seminar | April 4, 2025

Presenter Sungho Shin, Assistant Professor, Department of Chemical Engineering, MIT

Talk Title Solving large-scale, sparse, constrained nonlinear optimization problems on GPUs

Abstract

In this talk, we delve into recent advancements in solving large-scale sparse nonlinear optimization problems on GPUs. This new approach combines three key elements: (1) the SIMD abstraction of nonlinear optimization problems, (2) the condensed-space interior point method approach, and (3) NVIDIA’s recently released sparse linear solver, CUDSS. By combining these components, we provide a comprehensive nonlinear optimization solution capabilities. We present numerical experiment results showcasing over 10x speed-ups when applying this method to large-scale AC optimal power flow problems. We will conclude the discussion by examining the potential implications of these findings for the future of power system optimization and decision-making.