Posted on 06/20/2011 6:01:11 AM PDT by ShadowAce
HAMBURG, Germany—A Japanese supercomputer capable of performing more than 8 quadrillion calculations per second (petaflop/s) is the new number one system in the world, putting Japan back in the top spot for the first time since the Earth Simulator was dethroned in November 2004, according to the latest edition of the TOP500 List of the world’s top supercomputers. The system, called the K Computer, is at the RIKEN Advanced Institute for Computational Science (AICS) in Kobe.
The 37th edition of the closely watched list was released Monday, June 20, at the 2011 International Supercomputing Conference in Hamburg. The ranking of all systems is based on how fast they run Linpack, a benchmark application developed to solve a dense system of linear equations.
For the first time, all of the top 10 systems achieved petaflop/s performance – and those are also the only petaflop/s systems on the list. The U.S. is tops in petaflop/s with five systems performing at that level; Japan and China have two each, and France has one.
Bumped to second place after capturing No. 1 on the previous list is the Tianhe-1A supercomputer the National Supercomputing Center in Tianjin, China, with a performance at 2.6 petaflop/s. Also moving down a notch was Jaguar, a Cray supercomputer at the U.S. Department of Energy’s (DOE’s) Oak Ridge National Laboratory, at No. 3 with 1.75 petaflop/s.
Rounding out the Top 10 are Nebulae at China’s National Supercomputing Center in Shenzen (1.27 petaflop/s), Tsubame 2.0 at the Tokyo Institute of Technology (1.19 petaflop/s), Cielo at Los Alamos National Laboratory in New Mexico (1.11 petaflop/s), Pleiades at the NASA Ames Research Center in California (1.09 petaflop/s), Hopper at DOE’s National Energy Research Scientific Computing Center (NERSC) in California (1.054 petaflop/s), Tera 100 at the CEA (Commissariat à l'énergie atomique et aux énergies alternatives) in France (1.05 petaflop/s), and Roadrunner at Los Alamos National Laboratory in New Mexico (1.04 petaflop/s).
The K Computer, built by Fujitsu, currently combines 68544 SPARC64 VIIIfx CPUs, each with eight cores, for a total of 548,352 cores—almost twice as many as any other system in the TOP500. The K Computer is also more powerful than the next five systems on the list combined.
The K Computer’s name draws upon the Japanese word "Kei" for 10^16 (ten quadrillions), representing the system's performance goal of 10 petaflops. RIKEN is the Institute for Physical and Chemical Research. Unlike the Chinese system it displaced from the No. 1 slot and other recent very large system, the K Computer does not use graphics processors or other accelerators. The K Computer is also one of the most energy-efficient systems on the list.
Since its launch in 1993, the TOP500 List has provided a consistent metric for evaluating the performance of supercomputers. This consistency has made the list a valuable tool for tracking changes in the industry, both over the years and from list to list. Here are some notable changes between the current list and the 36th edition, issued in November 2010.
Just as the TOP500 List has emerged as a standardized indicator of performance and architecture trends since it was created 18 years ago, the list now tracks actual power consumption of supercomputers in a consistent fashion. Although power consumption is increasing, the computing efficiency of the systems is also improving. Here are some power consumption notes from the newest list.
The TOP500 list is compiled by Hans Meuer of the University of Mannheim, Germany; Erich Strohmaier and Horst Simon of NERSC/Lawrence Berkeley National Laboratory; and Jack Dongarra of the University of Tennessee, Knoxville.
Skynet or Colossus (depending on your age and reading predilections) just around the corner?
Or Data from ST:TNG?
LOL! I think our programming skills need to improve quite a bit before any of those are possible.
The NSA undoubtedly has a few that make these look weak.
While I would've agreed with you several years ago, I doubt that is the case these days.
This is some serious power--not only being produced, but being consumed.
“This is some serious power—not only being produced, but being consumed.”
Yes. Almost 10 MW for the top system now. It’s impressive at almost 1/2 million cores, but (so far) isn’t using the heterogeneous approach of adding GPU processing.
NVIDIA may become a major player in supercomputing between GPU accelerators and high-end ARM processors, which it is also going to push for desktop use. The ARM architecture is quite power efficient, and would seem to be a natural fit for these massively parallel machines.
It will be more than a little ironic if Windows supports the new NVIDIA ARM RISC processors while Apple sticks with Intel... LOL
With 80 global warming models running quadrillions of calculations every second you’d think the scientists could develop one that would model the economy and the effects of a new carbon tax on it.
Well, at 548,000+ cores, I'd say it's got more than a half million cores. :)
And according to this PDF (Page 10), The way it manages on-chip cache probably makes up for it's lack of GPU usage.
And the new interconnect model it's using is pretty sweet. After looking through the specs, I don't actually see any internal switches being used, but it's pushing out 100GB/s per core.
“Well, at 548,000+ cores, I’d say it’s got more than a half million cores. :)”
Oops, guess I misremembered the number. BTW, the PDF you references mentions “over 640K cores” (p. 2).
“And according to this PDF (Page 10), The way it manages on-chip cache probably makes up for it’s lack of GPU usage.”
Can’t be, the highly parallel GPUs pump out far more FLOPs per watt for appropriate problems than any traditional CPU. This computer simply has a whole lot of CPUs. Simpler programming, but less compute density and efficiency.
“And the new interconnect model it’s using is pretty sweet. After looking through the specs, I don’t actually see any internal switches being used, but it’s pushing out 100GB/s per core.”
The Tofu 6D mesh/torus interconnect is implicitly a switch fabric. They’ve been able to adapt MPI to their system.
Yeah--I've heard that they haven't actually finished building the machine yet.
The Tofu 6D mesh/torus interconnect is implicitly a switch fabric. They’ve been able to adapt MPI to their system.
OK--I'm not familiar with the Tofu interconnect, so I couldn't discuss specifics. It' looks nice, though.
I’m glad we’re going back in this direction. Supercomputers based on GPUs shouldn’t apply because they can’t be compared to others. GPUs can only do a limited set of operations — they do what they do very fast, but can’t do anything outside of that. They can’t be called general-purpose supercomputers.
If you want to take this to the logical extreme, we could design a chip that only does LINPACK, the Top 500 benchmark program, but does it very fast (basically, implement all of LINPACK in hardware). We could easily blow even this supercomputer away for a relatively small amount of money. We’d be at #1, but the computer would be useless for anything else. Of course it would be disallowed, but it is only further on the same road as a GPU supercomputer.
People have been playing around with this idea for years using FPGAs, so imagine a custom-fabbed multi-core LINPACK chip at 45 nm running at 3 GHz. The EFF had custom DES cracking chips made for an encryption breaking contest back in the 90s. Its 1900 chips with 24 DES crunchers each could exhaust the DES 56 keyspace in nine days (over 90 billion keys per second), while tens of thousands of PCs took months to do the same thing.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.