NCHC´s Formosa V GPU cluster enters the TOP500 ranking

Author

Chih-Wei Hsieh ; Chou-Yi Chou ; Tsung-Che Tsai ; Yu-Fen Cheng ; Shen-Hsiu Kuo

Author_Institution

Nat. Center for High-Performance Comput., Hsin-chu, Taiwan

fYear

2012

fDate

3-6 Dec. 2012

Firstpage

622

Lastpage

624

Abstract

Top 500 list is a competition for supercomputer rank in the world twice a year. How to employ the high performance LINPACK (HPL) to obtain the maximal LINPACK performance (Rmax) plays a key role in this competition. In this paper we propose these strategies for NCHC (National Center for High-Performance Computing) Formosa V GPU cluster that it is a GPU-based heterogeneous cluster to stand 232^nd place in the 39^th 500 list which is announced on June 18^th, 2012. We took three months to test this system and improve the hardware. We also cast a weighed function into HPL to make data decomposition more efficiency. Rmax is improved from 65 TFLOPS to near 90 TFLOPS!

Keywords

graphics processing units; mainframes; parallel processing; software libraries; GPU-based heterogeneous cluster; HPL; NCHC Formosa V GPU cluster; National Center for High-Performance Computing Formosa V GPU cluster; Rmax; TOP500 ranking; data decomposition; high performance LINPACK; supercomputer rank; top 500 list; Cloud computing; Computer architecture; Conferences; Graphics processing units; Hardware; Random access memory; Supercomputers; GPU clustert; HPL; TOP 500 list;

fLanguage

English

Publisher

ieee

Conference_Titel

Cloud Computing Technology and Science (CloudCom), 2012 IEEE 4th International Conference on

Conference_Location

Taipei

Print_ISBN

978-1-4673-4511-8

Electronic_ISBN

978-1-4673-4509-5

Type

conf

DOI

10.1109/CloudCom.2012.6427507

Filename

6427507