Introduction to the HPC Challenge Benchmark Suite - eScholarship
Introduction to the HPC Challenge Benchmark Suite - eScholarship
复制标题
HPC 挑战基准套件简介 - eScholarship
DOI：
--
复制
发表时间：
2005
期刊：
影响因子：
--
通讯作者：
D. Takahashi
中科院分区：
文献类型：
--
作者：
P. Luszczek;J. Dongarra;D. Koester;R. Rabenseifner;Bob Lucas;J. Kepner;John D. McCalpin;D. Bailey;D. Takahashi
HPC挑战基准套件Piotr Piotr Luszczek 1，Jack J. Dongarra 1,2，David Koester 3，Rolf Rabenseifner 4,5，Bob Lucas 6,7，Jeremy Kepner 8，John McCalpin 9，John McCalpin 9，David Bailey 10，David Bailey 10和Daisuke Takahashi Takahashi田纳西州诺克斯维尔山脊国家实验室11号11号木星4高性能计算中心（HLRS）5大学Stuttgart，www.hlrs.de/people/rabenseifner 6信息科学研究所7南加州大学8 MIT Lincoln Lab 9 IBM Austin 10劳伦斯·伯克利国家实验室11 tsukuba University 11 tsukuba Uniession ok oak 2 oak 2 Oak 1 2005年3月1日，2005年3月1日，和带状/潜伏期测试）试图跨越高空间和临时区域空间。 HPC挑战测试是可扩展的，而DARPA HPC计划租赁的HPC挑战率可以帮助定义数据集的函数，这是最大的HPL矩阵性能界限的函数。在HPC体系结构中使用内核具有内存访问模式的挑战比1个高性能计算的测试的测试 - Linpack（HPL）台式标记在Top500列表中，该套件旨在增强Top500列表，提供了DARPA高生产率计算SYS-BONDER的基准，以tems（hpcs）[1）[1 ]专注于提供新的基因 - 内存访问特征的函数，例如，经济上可行的高生产率计算和临时位置的序列，并提供一个国家安全系统，尤其是用于包括其他测试的工业。 ，HPL，矩阵乘数 - DGEMM，平行量度性能，可编程性，可移植性，RO-LEL矩阵转台 - Ptrans，FFT，Randomaccess，Abtness和untness和最终，高端1中的生产率这项工作得到了DARPA，NSF和计算（HEC）域的部分支持对政府HPC摘要的重要价值
Introduction to the HPC Challenge Benchmark Suite Piotr Luszczek 1 , Jack J. Dongarra 1,2 , David Koester 3 , Rolf Rabenseifner 4,5 , Bob Lucas 6,7 , Jeremy Kepner 8 , John McCalpin 9 , David Bailey 10 , and Daisuke Takahashi 11 of Tennessee Knoxville Ridge National Laboratory 3 MITRE 4 High Performance Computing Center (HLRS) 5 University Stuttgart, www.hlrs.de/people/rabenseifner 6 Information Sciences Institute 7 University of Southern California 8 MIT Lincoln Lab 9 IBM Austin 10 Lawrence Berkeley National Laboratory 11 University of Tsukuba 2 Oak 1 University March 2005 and bandwidth/latency tests – b eff ) that attempt to span high and low spatial and temporal locality space. By de- 1 benchmark suite has been re- sign, the HPC Challenge tests are scalable with the size The HPC Challenge leased by the DARPA HPCS program to help define the of data sets being a function of the largest HPL matrix performance boundaries of future Petascale computing for the tested system. systems. HPC Challenge is a suite of tests that exam- ine the performance of HPC architectures using kernels with memory access patterns more challenging than 1 High Productivity Computing those of the High Performance Linpack ( HPL ) bench- Systems mark used in the Top500 list. Thus, the suite is designed to augment the Top500 list, providing benchmarks that The DARPA High Productivity Computing Sys- bound the performance of many real applications as tems (HPCS) [1] is focused on providing a new gener- a function of memory access characteristics e.g., spa- ation of economically viable high productivity comput- tial and temporal locality, and providing a framework ing systems for national security and for the industrial for including additional tests. In particular, the suite user community. HPCS program researchers have initi- is composed of several well known computational ker- ated a fundamental reassessment of how we define and nels ( STREAM , HPL , matrix multiply – DGEMM , paral- measure performance, programmability, portability, ro- lel matrix transpose – PTRANS , FFT , RandomAccess , bustness and ultimately, productivity in the High End 1 This work was supported in part by the DARPA, NSF, and Computing (HEC) domain. The HPCS program seeks to create trans-petaflops DOE through the DARPA HPCS program under grant FA8750-04- systems of significant value to the Government HPC Abstract