The TOP500 continues to demonstrate the dominant position the U.S. assumes in the world both as producer and as consumer of high performance computers. In Table 2 the total number of installed systems in the major world regions is given with respect to the origin of the computers.
If one considers in Table 2 the country of origin then it is striking that 423 out of the TOP500 systems are produced in the U.S., which amounts to 85% of all installed systems. Japan accounts for 12% of the systems, and Europe produces only 3%. The extent of the American dominance of the market is quite surprising, and has been even increasing from the previous report, when the U.S. share was 84%. For years, in particular in the mid 80's, there were ominous and ubiquitous warnings that the American supercomputer industry (which was essentially Cray Research at that time) is highly vulnerable to an ``attack" by the Japanese vertically integrated computer giants Fujitsu, NEC, and Hitachi. Obviously this has not happened. How much various efforts such as the NSF Supercomputing Initiative in the mid 80's, or more recently the HPCC Program have contributed to the current vast superiority of the U.S. high performance computing industry, remains to be investigated. It is interesting to note that one view expressed outside the U.S. [13] is that strengthening the U.S. HPC industry and easing the transition to MPP was the only rationale for the HPCC Program.
The numbers for Europe are actually better than last year (15 machines in Nov. 95 versus 12 machines in June '95). This situation is probably not going to change, since one of the remaining two European vendors (Parsytec) will no longer focus on the HPC market. With lack of immediate access to the newest hardware, and the absence of the close interaction of users with vendors as is prevalent in the U.S., the best the European High Performance Computing and Networking Initiative can accomplish is maintaining the status quo of Europe as a distant third in high performance computing technologies.
Table 2: US Share of Total Number of Installed TOP500
Systems
Table 3 is analogous to Table 2, but instead of the number
of
systems, the aggregate performance in 
-Gflop/s is listed. Table
3
demonstrates a truly astounding event in 1995: within six months the total
number of
installed Gflop/s in the U.S. increased from 1392 Gflop/s in June to 2660
Gflop/s
in November 1995. This is an increase of 92% in only six months. At the same
time
growth in other regions was substantial but not quite as high: from 709 Gflop/s
to
1234 Gflop/s or 74% in Japan, and from 457 Gflop/s to 801 Gflop/s or 75% in
Europe.
What is more astounding is that this growth did not happen by installing a
few very large machines. Instead a large number of machines were installed,
which now
occupy medium to lower ranks on the TOP500 list. One conclusion from this data
is that
the HPCC initiative in the U.S. has succeeded in the sense that the
infrastructure
for HPC is dramatically changing. A large number of institutions now has access
to Gflop/s
level computing for machines which cost not much more than $ 1M. Only five
years ago
this compute power was accessible only to the elite few institutions being able
to spend
tens of millions of dollars. We can anticipate exciting times for HPC: more and
more
people in the U.S. will have access to inexpensive computational modeling
tools. It will
be worthwhile to examine what this revolution will do to economic productivity
measures
such as the GDP in the U.S.
Table 3: US Share of Total Rmax (in Gflop/s) of Installed
TOP500
Systems.
In an international comparison one should however also consider the relative size of countries and their economies. Here we present a new TOP500 set of statistics. In Table 4 we list the a measure of the supercomputer density by ranking the top ten countries with the highest number of supercomputer per capita. Population date are from the ``Interactive 3D Atlas" and date from 1992.
Table 4: Population (in thousands) per TOP500
supercomputer.
Table 4 shows that on an international comparison most industrialized countries are providing about one supercomputer per 1 - 2.5 million inhabitants. The number of US installations is no longer that dramatically different from the rest of industrialized countries. It should be mentioned that the among the major industrialized nations the big anomaly with respect to supercomputing usage is Italy. In Italy there is only one supercomputer per 9.6 million inhabitants, far below the number of all other western European countries.