The spontaneous reaction in the U.S. can be described as rather surprised and hectic; obviously the Americans were caught on the wrong foot. The comments in the various technical periodicals and also the daily press culminated in the expression "Computenik" in the New York Times, in allusion to the Russian Sputnik of 1957, which had likewise shocked the Americans and became crucial for the further progression of the American space programme.

Of course, the ES project was well-known in the U.S. from its beginning in 1997. Despite the Japanese steadily exposed the completion date of March 2002 this was probably not taken too serious in the U.S. or perhaps there was no faith in the Japanese capabilities to complete the project in time.

What is so special about the ES? It was developed decidedly for the study of the Earth, e. g. to enable global climate research and particularly to analyse global warming. In addition, disaster forecasts (e. g. earthquakes and typhoons) are focused by the Japanese government, who was financing the five-year project with approximately $ 350 to 400 million. Thereby the ES programme is distinguished from the American Department of Energy's (DOE's) militarily oriented ASCI programme, in which supercomputers are developed in order to simulate nuclear weapons.

The ES is based on NEC vector technology, 5,120 vector processors are connected via a single-level crossbar switch. The main memory has a capacity of 10 Terabytes; the secondary memory comprises 700 Terabytes of disk storage and 1.6 Petabytes of mass storage. In order to accommodate the computer system, air conditioning and wiring a three-storied building with the size of four tennis courts was erected, causing approximately 1/3 of the total costs.

To set up its Linpack world record the ES solved a set of 1,041,216 linear equations in hardly six hours. The matrix of this set of equations was 8.7 Terabytes large and fitted into the main memory. The ES represents the provisional end of now 55 years of computer development. The unimaginable factor of 3.6 quintillions lies between the Zuse Z4 built by Konrad Zuse in 1947 with a performance of approximately 10 Flop/s and the ES with 3.6 *10**13 Flop/s. If we assume that the Zuse Z4 could achieve the performance of 10 Flop/s at the solution of this giant set of equations (of course impossible due to its small main memory capacity), then it would need approximately 3*10**9 years for the solution of this set of equations, which is about 1/3 of the time since the Big Bang. As mentioned above, the ES makes it in just six hours.

The increase of performance over 55 years corresponds in average to the factor of 1.76 per year, which is an exponential growth and larger than postulated by Moore's Law, i.e. 1.58 per year or doubling of performance every 18 months. Besides the effect of Moore's Law, the supercomputers' increase of performance stems from more and more processors.

On the SC2002 Conference in Baltimore last month, the Americans reacted to the ES by a governmental announcement for 2004/ 2005 and presented the ASCI Purple with an peak performance of 100 Teraflop/s, as expected and in addition "Blue Gene/L" which will be used to develop and run a broad suite of scientific applications including the simulation of very complex physical phenomena of national interest, such as turbulence, prediction of material properties, and the behaviour of high explosives. Blue Gene/L is expected to have a theoretical peak performance of up to 367 Teraflops with 130,000 processors. In Baltimore US Energy Secretary Spencer Abraham has announced that the DOE has awarded IBM a contract valued at $ 216 to $ 267 million to build these two fastest supercomputers in the world.

Taking into account that the Japanese fabricate only approximately 8 % of the TOP500 systems and fell far behind Europe in the use of supercomputers, with the ES they succeeded in developing a "Computenik" which will keep the lead for some years and, moreover, will for sure "promote" the American supercomputer development.

Reprinted with permission from the "SC2003 Newsletter 01/2003" at http://www.isc2003.org/.