Accelerating Sparse Iterative Solvers and Preconditioners Using RACE - Supercomputing Spotlights

Explore cutting-edge advancements in high-performance computing through this webinar from the SIAM Activity Group on Supercomputing's Supercomputing Spotlights series. Delve into Christie Alappat's presentation on accelerating sparse iterative solvers and preconditioners using RACE. Learn about the challenges and solutions in optimizing the sparse matrix-vector multiplication (SpMV) kernel, a critical component in computational science algorithms. Discover how the RACE library framework's level-based approach can significantly boost performance, achieving 1.5-4x speedups on modern multicore processors. Examine the integration of RACE with the Trilinos framework and its application to communication-avoiding s-step Krylov solvers, polynomial preconditioners, and algebraic multigrid (AMG) preconditioners. Gain insights into the performance benefits and challenges of implementing these optimizations, and understand how they can improve total solver time by 1.3x - 2x while maintaining numerical accuracy.

Supercomputing Spotlights: Accelerating sparse iterative solvers and preconditioners using RACE