The Road to TSUBAME and Beyond

Abstract

TSUBAME (Tokyo-tech Supercomputer and Ubiquitously Accessible Massstorage Environment) is a new supercomputer installed at Tokyo Institute of Technology in Tokyo, Japan, on April 1st, 2006, facilitating over 85 Teraflops of peak compute power with acceleration, 21 Terabytes of memory, and 1.6 Petabytes (initially 1.1 Petabytes) of online disk storage, “Fat Node” as well as fast parallel interconnect supercomputers. TSUBAME became the fastest and largest supercomputer in Asia in terms of performance, memory and storage capacity etc. It recorded 38.18 Teraflops performance for the June 2006 Top500 [1], making it the 7th fastest supercomputer in the world at the time and fastest in the Asia-Pacific, besting the Earth Simulator. TSUBAME continues to improve performance in three consecutive Top500s, and with lucid use off acceleration, has now marked 48.88 Teraflops. At the same time, being PC architecture-based, TSUBAME, being a large collection of PC servers, allows for offering much broader services than traditional supercomputers resulting in a much wider user base, including incubation of novice students. We term such architectural and operational property of TSUBAME as “Everybody’s Supercomputer”, as opposed to traditional supercomputers with very limited number of users, thus making their financial justifications increasingly difficult.