BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20121115T213000Z
DTEND:20121115T220000Z
LOCATION:255-BC
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: Multigrid methods are widely used to accelerate the convergence of iterative solvers for linear systems.  We explore optimization techniques for geometric multigrid on existing and emerging multicore systems including the Cray XE6, Intel SandyBridge and Nehalem-based Infiniband clusters, as well as Intel's forthcoming Knights Corner (KNC) Coprocessor.=0AOur work examines a variety of techniques including communication-aggregation, threaded wavefront-based DRAM communication-avoiding, dynamic threading decisions, SIMDization, and fusion of operators.=0AWe quantify performance through each phase of the V-cycle for both=0Asingle-node and distributed-memory experiments and provide detailed=0Aanalysis for each class of optimization.=0AResults show our optimizations yield significant speedups across a variety of subdomain sizes while demonstrating the potential of multi- and manycore processors to dramatically accelerate single-node performance.=0AOur analysis also indicates that improvements in networks and communication will be essential to reap the potential of manycore processors in large-scale multigrid simulations.
SUMMARY:Optimization of Geometric Multigrid for Emerging Multi- and Manycore Processors
PRIORITY:3
END:VEVENT
END:VCALENDAR
BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20121115T213000Z
DTEND:20121115T220000Z
LOCATION:255-BC
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: Multigrid methods are widely used to accelerate the convergence of iterative solvers for linear systems.  We explore optimization techniques for geometric multigrid on existing and emerging multicore systems including the Cray XE6, Intel SandyBridge and Nehalem-based Infiniband clusters, as well as Intel's forthcoming Knights Corner (KNC) Coprocessor.=0AOur work examines a variety of techniques including communication-aggregation, threaded wavefront-based DRAM communication-avoiding, dynamic threading decisions, SIMDization, and fusion of operators.=0AWe quantify performance through each phase of the V-cycle for both=0Asingle-node and distributed-memory experiments and provide detailed=0Aanalysis for each class of optimization.=0AResults show our optimizations yield significant speedups across a variety of subdomain sizes while demonstrating the potential of multi- and manycore processors to dramatically accelerate single-node performance.=0AOur analysis also indicates that improvements in networks and communication will be essential to reap the potential of manycore processors in large-scale multigrid simulations.
SUMMARY:Optimization of Geometric Multigrid for Emerging Multi- and Manycore Processors
PRIORITY:3
END:VEVENT
END:VCALENDAR