SIAM CSE 09 -- Emerging Many-Core Architectures

SIAM Computational Science and Engineering 2009

Scientific Computing on Emerging Many-Core Architectures

Organisers:

Mike Giles, Oxford University
Dominik Göddeke, TU Dortmund
Stefan Turek, TU Dortmund

On this webpage we have the talks presented at a two-session minisymposium at the recent SIAM Computational Science and Engineering conference in Miami. The topic was the opportunities and challenges for scientific computing on novel manycore architectures, in particular IBM's Cell processor and the GPUs from NVIDIA, AMD and Intel. The talks cover a range of different application areas, and many of them discuss the development of libraries and other tools to simplify the task of others seeking factor 10x improvements in price/performance and energy efficiency compared to more conventional multicore CPUs.

First session:

Manycore GPU Scientific Computing
Wen-mei Hwu, University of Illinois at Urbana-Champaign

Efficient Scheduling of Parallel Tasks in Dense Linear Algebra: PLASMA Framework
Jakub Kurzak, University of Tennessee, Knoxville

Scientific Computing on Graphics Processors: FMM, FLAGON Signal Processing, Plasma and Astrophysics
Ramani Duraiswami, University of Maryland

GPU-based High Performance Computing: A New Paradigm for Clinical Medical Imaging
Sanjiv Samant, University of Florida

Second session:

GPU Cluster Computing for Finite Element Simulations
Dominik Göddeke, TU Dortmund

Computational Finance Using CUDA on NVIDIA GPUs
Mike Giles, Oxford University

Extracting Parallelism from Financial Applications with SIMD, Multicore and Manycore
Shuo Li, Intel

Fluid Flow Simulations on Roadrunner
Jamaludin Mohd-Yusof, Los Alamos National Laboratory