ISC'14

June 22–26, 2014
Leipzig, Germany

Presentation Details

 
Name: The EigenExa Library – High Performance & Scalable Direct Eigensolver for Large-Scale Computational Science
 
Time: Thursday, June 26, 2014
11:00 am - 11:20 am
 
Room:   Hall 4
CCL - Congress Center Leipzig
 
Breaks:10:30 am - 11:00 am Coffee Break
 
Speaker:   Toshiyuki Imamura, RIKEN/AICS
 
Abstract:   EigenExa is a high performance and high scalable eigenvalue solver for dense matrices on a peta-scale parallel supercomputer system. EigenExa is used as a powerful mathematical tool and a building block for parallel numerical simulations such as quantum physics, quantum chemistry, drug design, statistical data analysis, and so on. From the latest benchmark, EigenExa successfully solves full diagonalization of a one million dimension matrix taking advantage of the full-nodes (82,944 processors) of K computer within one hour. The performance records 1.7 PFLOPS (16% of K computer's theoretical peak performance). It is a world largest scale and highest performance for solving an eigenvalue problem of a dense matrix. This presentation reports the review of modern parallel eigenvalue solvers and the trend of the parallel algorithms for high performance numerical libraries. We will also discuss an impact of such a challenging eigenvalue computation to Exa-scale computing.