Super Computer

Supercomputer

A very-high-performance computer, traditionally BigIron of MainframeComputer size. Examples include the ControlData 7600 and the CrayResearch Cray-1.

More recent efforts to obtain comparably superior performance--that is, performance comparably superior to contemporary MainframeComputers--have been focused on large arrays of microprocessors, as exemplified by a recent announcement from IBM: "BlueGene/L becomes the world's most powerful supercomputer" (http://www.research.ibm.com/resources/news/20040930_bluegene.shtml). IBM's earlier "Blue Gene/L DD1 Prototype, with a sustained speed of 11.68 teraflops and a peak speed of 16 teraflops, uses more than 8,000 PowerPC processors" (http://www.research.ibm.com/bluegene/index.html). That is to say, a processor-array containing "more than 8,000 ... processors" (um, maybe exactly 8192?) is the predecessor of an architecture that has now been claimed to implement "the world's most powerful supercomputer".

Comparing clusters and big iron is an apples and oranges comparison. Clusters are a cheap way of getting a lot of CPU cycles onto certain types of problems. Luckily, some important problems fit these constraints.

Several different categories of problems, in fact. Many numerically-intensive problems are naturally partitionable onto a cluster - simulations of all sorts, for example. Many IoBound? applications fit on a cluster as well - a common architecture for a high-traffic-volume WebSite is to have a ServerFarm? of PC's all running your favorite webserver, each handling a fraction of the load. OTOH, scattering a RelationalDatabaseManagementSystem? across a cluster doesn't work very well.

Current record-holder (late 2004): EarthSimulator -- over 35 teraflops.

The claim in the hastily edited DocumentMode section was not that "11.68 teraflops" sustained, "16 teraflops" peak was the record, but that the IBM Blue Gene/L system had surpassed the existing record. All the commentator immediately above had to do was visit the URLs provided above, and the new record would have been clear: "IBM has announced that its IBM Blue Gene/L supercomputer has surpassed NEC's Earth Simulator in Japan to become the world's most powerful supercomputer. Using the industry-standard LINPACK benchmark, the IBM Blue Gene/L system attained a sustained performance of 36.01 teraflops, eclipsing the three year old top mark of 35.86 teraflops for the Japanese Earth Simulator in Yokohama, Japan." And yes, the creator of this Wikipage is very much aware that the FORTRANish LINPACK is a synthetic benchmark.

I'm amazed at the speed of response to this Wikipage, having created it as not much more than a stub either late last nite or midafternoon today (USA EDT), thinking it would be days or weeks before it caught anyone's notice. Please feel free to expand the DocumentMode section at the top so it's much more complete; Neptune knows I've already spent waaay too much time on expanding the CategoryProgrammingLanguage and CategoryComputerArchitecture in the last 2 days, when Wikipedia often--if not typically--already has more detailed information than I have time to provide. -- ClayPhipps

CategoryHardware CategoryComputerArchitecture