- Follow this blog:
- RSS
- Email Alert
TR Dojo
Supercomputer with 68,000+ CPUs is world's fastest
Takeaway: The Fujitsu “K computer” has 672 racks, 68,544 CPUs, and leads TOP500.Org’s 37th list of world’s fastest supercomputers.
According to TOP500.Org, the Fujitsu-built “K computer” in Kobe, Japan, is currently the world’s fastest supercomputer. Able to complete more than 8 quadrillion calculations per second (petaflop/s), the K computer uses 672 computer racks and 68,544 SPARC64 VIIIfx CPUs–with eight cores each.
Although only the K computer is only half finished, it was able to achieve a LINPACK benchmark performance of 8.162 petaflops. When completed in 2012, the K computer will have over 800 racks, and be designed to reach 10 petaflops.
The K Computer has nearly twice as many cores as any other supercomputer on the TOP500 list and is more powerful than the next five machines combined. It’s housed at the RIKEN Advanced Institute for Computational Science.
Top 10 Fastest Supercomputers
TOP500.Org has been ranking the world’s supercomputers since 1993 and updates its list supercomputers twice a year–in June and November. The following 10 systems top the organization’s 37th list (released June 2011):- “K computer” (RIKEN Advanced Institute for Computational Science, Japan)
- Tianhe-1A (National Supercomputing Center in Tianjin, China)
- Jaguar (Oak Ridge National Laboratory, U.S.)
- Nebulae (National Supercomputing Center, China)
- Tsubame 2.0 (Tokyo Institute of Technology, Japan)
- Cielo (Los Alamos National Laboratory, U.S.)
- Pleiades (NASA Ames Research Center, U.S.)
- Hopper (National Energy Research Scientific Computing Center , U.S.)
- Tera 100 (Commissariat à l’énergie atomique et aux énergies alternatives, France)
- Roadrunner (Los Alamos National Laboratory, U.S.)
More supercomputer resources
- TOP500 lists the world’s fastest supercomputers - Summer 2010 (TechRepublic)
- Japanese supercomputer is fastest in the world (CNET)
- China builds world’s fastest supercomputer (ZDNet UK)
- The 80’s supercomputer that’s sitting in your lap (TechRepublic)
- Inside NASA’s world-class supercomputer center (TechRepublic)
- IBM’s Watson victorious in Jeopardy; Our new computer overlord? (ZDNet)
Get IT Tips, news, and reviews delivered directly to your inbox by subscribing to TechRepublic’s free newsletters.
About Bill Detwiler
Bill Detwiler is Head Technology Editor of TechRepublic. Previously, he worked as a Support Tech and IT Manager in the social research and energy industries.
Rated
What I would use it for:
In
How slow will Ninary Computers look when Quantum PC's come out?
What would you do with 68,000+ CPUs?
What would you do with all that computing power?
What I would use it for:
Super Computer
Farm Gold
Modeling stuff. What else?
http://www.fujitsu.com/global/about/tech/k/whatis/network/
Plus, Linux comes in handy again. I wonder what other operating systems the designers had considered.
write a letter to the editor
It would keep me going for years to come for my large raw pictures...
Well, I´d use it for search and destroy viruses....
True definition of "super-computer"...
Computers have evolved and so should the definition
Exactly!
Parallel computing can take a task, break it into many minute tasks, which will be run concurrently, and at the end, it will have arrived at the results many times faster than a single CPU computer.
Supercomputer
Always, when we hear about supercomputers they're these mammoths, usually with identifying names, rather than type names.
Unique, towering beasts.
No. The definition has been set, for a VERY long time.
What you describe is exactly the situation which programmers would try VERY, VERY HARD to avoid doing: Using the result of on calculation as input for the next. The Cray-1 could do this for address calculations, and for a data addition leading into a multiplication on each Vector Element, but using a result in any other way would break the pipeline. Such sequential, "scalar" operations were a bad thing. (And they still are.)
Furthermore, your notion of "a single CPU doing all the work" was abandoned even earlier. Computers like the Cray-1, and it's predecessors, *AND* your modern PC, all depend of "helper" processors to do a lot of the work. How much?
Well, as a specific example, the CDC 6600 didn't even have an instruction to store a register value to memory! You would store your result in one of a "special" group of registers, and then a "peripheral processor" would take over the rest of the work.
And what is a "single CPU", anyway? Again using the various Cray-1 models as the example of a "classic" supercomputer, most of them consisted of over 100,000 4-5 NAND gate chips. (Anyway, I worked there for a few years, in software.)
- - - -
This is definitely a supercomputer. It's interesting, however, that this computer is built from Fujitsu's own SPARC chips, rather than AMD or Intel.
Just one question
Deep Thought
(You're not going to like it...) Is...
42
I speak of the computer to come after me.
sparc
I'd put money on the sparc being more power efficient than an AMD or intel based system of equal capacity.