The discussion of the high level of government support for MPPs in the U.S. in section 10.3 should imply that there is a large installed base of supercomputers in the government sector in the U.S. This is indeed the case as Tables 10.7-10.9 show.
In Tables 10.7-10.8 the distribution of the TOP100 machines across different sectors is shown with respect to number of installations and performance in [tex2html_wrap1344]-Gflop/s. Here government refers to all government agencies and laboratories, which ranges from basic research environments to more production oriented machines as in the national weather bureaus. The categories university and industry are self-explanatory. The column with vendor machines includes all the machines used by supercomputer vendors in house for benchmarking and software development.
From the data in Table 10.7 it is clear that the distribution of the largest supercomputers in the U.S. is fairly consistent with the distribution in Japan and Europe. In all three regions/countries government accounts for about half of the largest supercomputers and industry for around 20%.
Comparing the distribution of the TOP100 with the TOP500 is quite intriguing. First it is notable that industrial installations with about 27%of the total have world wide a much larger share among the TOP500 systems. This is to be expected since one can expect the really large machines to be predominant in research environments. However, what is surprising is that the distribution between industrial and government installations is quite different in the corresponding countries/regions. In Japan only about 21%of the supercomputers are installed in government laboratories, whereas 40%can be found in industry. With 39%government and 20%industrial use the situation in Europe is exactly reverse. The U.S. takes a middle position, somewhat closer to the situation in Europe, with 33%government use and 25%industrial use.
[table274]
[table280]
[table286]
These results seem to imply that industrial use of supercomputers in Japan is considerably more widespread than in the U.S. While this is probably true in general, the situation in the U.S. is biased because of the effects of HPCC. It is not that U.S. industry is using significantly less supercomputers than Japan, but rather that U.S. government laboratories due to government spending are using considerably more supercomputer power than their Japanese counterparts. It is the goal of the HPCC program that this heavy use in government laboratories will have eventually a positive effect on the industrial use of high performance computing technologies. If this type of technology transfer will occur remains to be seen. Future version of the TOP500 report will provide the necessary data.