Japan operates a number of centers for supercomputing which hold world records in speed, with the K computer becoming the world's fastest in June 2011. and Fugaku took the lead in June 2020, and furthered it, as of November 2020, to 3 times faster than number two computer.
The K computer's performance was impressive, according to professor Jack Dongarra who maintains the TOP500 list of supercomputers, and it surpassed its next 5 competitors combined. The K computer cost US$10 million a year to operate.
Previous records
Japan's entry into supercomputing began in the early 1980s. In 1982, Osaka University's LINKS-1 Computer Graphics System used a massively parallel processing architecture, with 514 microprocessors, including 257 Zilog Z8001 control processors and 257 iAPX 86/20 floating-point processors. It was mainly used for rendering realistic 3D computer graphics. It was the world's most powerful computer, as of 1984.
The SX-3 supercomputer family was developed by NEC Corporation and announced in April 1989. The SX-3/44R became the fastest supercomputer in the world in 1990. Fujitsu's Numerical Wind Tunnel supercomputer gained the top spot in 1993. Japanese supercomputers continued to top the TOP500 lists up until 1997.
The K computer's placement on the top spot was seven years after Japan held the title in 2004. NEC's Earth Simulator supercomputer built by NEC at the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) was the fastest in the world at that time. It used 5,120 NEC SX-6i processors, generating a performance of 28,293,540 MIPS (million instructions per second). It also had a peak performance of 131 TFLOPS (131 trillion floating-point operations per second), using proprietary vector processing chips.
The K computer used over 60,000 commercial scalar SPARC64 VIIIfx processors housed in over 600 cabinets. The fact that K computer was over 60 times faster than the Earth Simulator, and that the Earth Simulator ranked as the 68th system in the world 7 years after holding the top spot, demonstrates both the rapid increase in top performance in Japan and the widespread growth of supercomputing technology worldwide.
Supercomputing centers
The GSIC Center at the Tokyo Institute of Technology houses the Tsubame 2.0 supercomputer, which has a peak of 2,288 TFLOPS and in June 2011 ranked 5th in the world. It was developed at the Tokyo Institute of Technology in collaboration with NEC and HP, and has 1,400 nodes using both HP Proliant and NVIDIA Tesla processors.
The RIKEN MDGRAPE-3 for molecular dynamics simulations of proteins is a special purpose petascale supercomputer at the Advanced Center for Computing and Communication, RIKEN in Wakō, Saitama, just outside Tokyo. It uses over 4,800 custom MDGRAPE-3 chips, as well as Intel Xeon processors. However, given that it is a special purpose computer, it can not appear on the TOP500 list which requires Linpack benchmarking.
The next significant system is Japan Atomic Energy Agency's PRIMERGY BX900 Fujitsu supercomputer. It is significantly slower, reaching 200 TFLOPS and ranking as the 38th in the world in 2011.
Historically, the Gravity Pipe (GRAPE) system for astrophysics at the University of Tokyo was distinguished not by its top speed of 64 Tflops, but by its cost and energy efficiency, having won the Gordon Bell Prize in 1999, at about $7 per megaflops, using special purpose processing elements.
DEGIMA is a highly cost and energy-efficient computer cluster at the Nagasaki Advanced Computing Center, Nagasaki University. It is used for hierarchical N-body simulations and has a peak performance of 111 TFLOPS with an energy efficiency of 1376 MFLOPS/watt. The overall cost of the hardware was approximately US$500,000.
The Computational Simulation Centre, International Fusion Energy Research Centre of the ITER Broader Approach/Japan Atomic Energy Agency operates a 1.52 PFLOPS supercomputer (currently operating at 442 TFLOPS) in Rokkasho, Aomori. The system, called Helios (Roku-chan in Japanese), consists of 4,410 Groupe Bull bullx B510 compute blades, and is used for fusion simulation projects.
The University of Tokyo's Information Technology Center in Kashiwa, Chiba, began operating Oakleaf-FX in April 2012. This supercomputer is a Fujitsu PRIMEHPC FX10 (a commercial version of the K computer) configured with 4,800 compute nodes for a peak performance of 1.13 PFLOPS. Each of the compute nodes is a SPARC64 IXfx processor connected to other nodes via a six-dimensional mesh/torus interconnect.