- One of the strategic goals of the HPCC program was to ``spur gains in U.S. productivity and industrial competitiveness by making high performance computing and networking technologies an integral part of the design and production processes''. The TOP500 data can give some indication what progress has been towards this goal. In Table 10.10 the distribution of the 135 industrial supercomputers worldwide is given by industry segment, as well as by MPP versus PVP, and by use in the U.S. versus all other countries.
The most surprising fact is that in the U.S. the number of MPPs in industry is almost the same as the number of PVPs. Thus even in the most competitive environment MPPs have made already major inroads. These data contradict some of the conclusions made in a controversial study [4], where it is maintained that ``conventional supercomputers ... (are) hard to displace in the market''. The TOP500 data indicate that displacement of PVP has already occurred to a much larger extent than generally believed in areas, which have been traditionally claimed by PVPs. The two industry segments where MPPs have made most progress in displacing PVPs are aerospace and geophysics. The success of MPPs in these fields can be easily explained by the nature of some of the typical applications. Seismic processing in the oil industry is an application where MPPs excel, since most of the computation is of explicit nature and done on structured grids. Similarly both electro-magnetics applications using methods of moments algorithms, and signal processing applications account for the use of MPPs in the aerospace community. The parallel applications taxonomy in [10] explains this success of MPPs.
Similarly there are areas where MPPs had no success whatsoever. Not surprisingly these are for example in the automotive industry. This industry is characterized by using unstructured finite element codes with implicit solution algorithms. These type of applications are very difficult to solve efficiently on MPPs. Another characteristic of the automotive industry is reliance on third party applications such as crash codes and structural analysis packages. So far third party developers, mostly small software companies, have been reluctant to port their software to unproven new technology. In contrast, most oil companies employ proprietary in-house codes, and have been willing to make the investment in porting applications. Thus the data in Table 10.10 reflect exactly the difficulty in porting the applications in different industry segments.
[table299]
Another observation is that MPPs so far have been replacing PVPs in established field of high performance computing use. There have been no significant inroads into new applications markets yet. Even though there are several MPPs now in financial institutions, or in chemical and biological applications, their numbers are still small. Hence MPPs have not (yet) opened up new markets for HPC technology.
There are a few other observations one can make from Table 10.10 concerning HPC use in industry comparison. For example all 10 supercomputers in the metal working and construction industry are installed in Japan, whereas in aerospace with 18 out of 22 installations the U.S. is the clear leader. The supercomputers in the automotive industry are almost evenly distributed worldwide. Every major car manufacturers owns a supercomputer (with the exception of Volvo). One can therefore assume that the use of supercomputers in industrial applications is a consequence of the leadership position each country enjoys in the respective applications field (and not vice versa). However, the topic of supercomputer use and industrial leadership deserves some further detailed investigation.