<technical report>
Performance Evaluation of a Large Scale Reconfigurable Data-Path Utilized for Scientific Application

Creator
Language
Publisher
Date
Source Title
Source Title
Vol
Issue
First Page
Last Page
Publication Type
Access Rights
Related DOI
Related URI
Relation
Abstract 最近,要求メモリバンド幅を抑えつつも高性能な科学技術計算を可能とする,大規模再構成可能データパス(LSRDP)が提案された.本稿ではこのLSRDP について,種々のメモリバンド幅のもとでの性能評価を行った.対象アプリケーションは,1 次元の熱伝導方程式,2 次元のPoisson 方程式,2 電子積分計算の3 種類である.その結果,熱伝導方程式,2 電子積分計算については,汎用プロセッサのみでの計算...に比較し大きく性能向上を果たしたが,Poisson 方程式については逆に実行時間が増加した.これは,主記憶からLSRDP へのバースト転送利用のためにLSRDP 計算にて行っている主記憶上での入出力データの並び替えステップが,実行時間の大半を占めてしまうからである.
Recently, Large Scale Reconfigurable Data Path (LSRDP) processor has been proposed for the reduction of required memory bandwidth in a high performance scientific computing. In this paper, performance evaluations of LSRDP at various memory bandwidths have has been demonstrated for 1-dimensional Heat and 2-dimensional Poisson partial diffrential equations, and for Electron Repulsion Integral (ERI) calculation as target benchmark applications. Executions times for Heat and ERI applications were decreased compared to original execution by the general purpose processor. On the other hand for Poisson application, execution times were increased, since data sorting steps, which are needed for burst transfer from main memory to LSRDP, consume great amount of execution time.
show more

Hide fulltext details.

pdf kataoka08_1 pdf 319 KB 286  

Details

Record ID
Peer-Reviewed
Subject Terms
ISSN
NCID
Type
Created Date 2009.04.22
Modified Date 2020.11.27

People who viewed this item also viewed