Berkeley Lab - Scientific Computing Seminar

Date:

Tuesday, July 1, 2008

Time:

1:00pm-2:00pm

Location:

Building 50F, 1647 Conference Room

Seminar Speaker:

Tobias Gradl
Department Informatik
Universität Erlangen-Nürnberg, Germany

Title:

A Multigrid Solver for Massively Parallel Finite Element Computations

Abstract:

Among the various discretization methods in numerical simulation, Finite Elements are often favoured for their flexibility when unstructured meshes are required. As supercomputers have never performed very well on unstructured data, block-structured approaches have always enjoyed attention. HHG ("Hierarchical Hybrid Grids") is a Multigrid solver for hierarchically structured meshes. It is designed to solve very large simulations on massively parallel computers. Having solved a problem with 10¹¹ unknowns in 1.5 minutes on 9170 cores of an SGI Altix, it currently keeps the world record in Finite Element linear system solving. Using a consistent Multigrid approach is a crucial factor at these problem sizes, because Multigrid methods are the only solvers that scale linearly with the number of unknowns. During my three-month stay at Lawrence Berkeley Lab, I would like to make HHG fit for the upcoming generations of ever larger and increasingly diverse supercomputers. I am looking forward to many fruitful discussions with everyone who wants to share some knowledge about parallel performance analysis and tuning.

Sponsor of Seminar:

Erich Strohmaier

Scientific Computing

Contact Esmond G. Ng EGNg@lbl.gov