GPU Accelerated Sparse Inverse Factor Matrix-Vector Multiplication Solver

Solving systems of linear equations (Ax = b) is a fundamental yet computationally intensive task in engineering applications, especially in power system simulations, where such operations often dominate execution time. Current power system simulators predominantly rely on LU decomposition-based linear solvers for transient simulations, whether executed on CPUs or GPUs. While traditional LU decomposition is the preferred method in most simulators, its inherently sequential nature limits performance gains on modern GPU architectures.

 

To overcome the current limitations in linear solvers, researchers at Arizona State University have introduced a Sparse Inverse Factor (SIF) solver that exploits the high throughput matrix-vector multiplication on GPUs. This approach leverages precomputed sparse inverse LU factors to enable parallel matrix-vector multiplication on GPUs, overcoming inefficiencies of traditional sequential LU-based solvers. Implemented using Julia language on high-performance workstations, it delivers substantial speed-ups in single and batch scenario power flow analyses while maintaining numerical accuracy and stability. By eliminating serial forward and backward substitution, the method significantly reduces computation time.

 

This technology introduces a novel GPU-based implementation of the Current Injection Method which significantly accelerates power flow computations in large-scale distribution systems.

 

For more information about this opportunity, please see