The historical trend is that computational performance increases by
about two orders of magnitude a decade, and this is a trend that we should
expect to see continue into the foreseeable future. What is happening is
that we have been riding the rapid growth stages of a continuing series
of technological advances. As growth in one type of advance levels off,
rapid growth due to some other advance kicks in, and thus growth continues
at a rapid pace.
Looking at single-user PC performance in Figure 1 (right), the two orders of magnitude per decade growth rate is apparent, putting us over 100 MFLOPS (millions of floating point operations per second) today. This is based not on a theoretical peak rate, but the rate on solving a 100 x 100 system of linear equations (the LINPACK-100 benchmark). For reference, Figure 1 (right) also shows a couple of old machines that I used to work with: the CDC Cyber 175 mainframe and the Cray X-MP supercomputer. Clearly, we now have come a long ways since then, especially in price/performance ratio. The Cray X-MP cost around 8-10 million dollars; a comparably performing machine today would run around 1000 dollars, and by Christmas shopping time in 1999 we should expect this to drop to around 200 dollars. The PC is rapidly becoming a "disposable" applicance.