With NASA fights for the chunk of US budget burdened by the economical crisis and the running expenses of the Iraq and Afghanistan campaigns – the Japanese are accelerating their space travelling efforts with the deployment of thier latest supercomputer. 3,008 Fujitsu Sparc64-VII CPUs were selected for a supercomputer that should help JAXA [Japanese Aerospace Exploration Agency] to accelerate its spaceship development and hopefully, flight to Moon and Mars.
The system is based upon 3,008 Fujitsu SPARC64VII quad-core processors ticking at 2.52 GHz each. The hardware specifications are positioned somewhere between AMD and Intel processors, with the "small" omission that this is a pure-blooded RISC architecture, not an "x86: CISC that became RISC with CISC translator" one. Each CPU is consisted out of four cores – eight threads by using Simultaneous Multithreading [same MT tech as Intel], 64+64 KB of L1 cache and 6MB of L2 cache.
These 3008 processors make a total sum of 12,032 cores, or 24,064 threads. You also have 96TB of RAM to play with, 1PB [Petabyte] of storage and 10 PB in tapes for backup purposes. The system was launched on April 1st, 2009, and achieved full speed seven days later, scoring 110.6 TFLOPS on Linpack benchmark. What made this system special was its efficiency – Linpack is notoriously sensitive to latency, as we disclosed in Nova’s Nehalem-EP articles. The Japanese experts managed to achieve 91.19% efficiency [theoretical peak: 120 TFLOPS] – in comparison, a system with 752 Nehalem-EP based processors would achieve 71.42 TFLOPS at 100% efficiency [at a fraction of the price].
While this system is massive, we wonder what would happen if the system by design included ATI FireStream of nVidia Tesla cards. Unfortunately for ATI and nVidia, their development focus is on x86 systems.