Looking at the current situation in parallel computing (Table 1, below), the latest hardware gives a peak performance of around 4 teraflops (million MFLOPS), with a likely Rmax of over 3 teraflops. At Notre Dame, I am currently working with a group of people that is already working on the technology and algorithms for petaflop machines—that is, 1015 floating point operations per second (a billion MFLOPS). One trend that we can see here is that these high end machines use massively parallel computing; and they do this with commodity level chips. For example, the Intel Red machine is based on Pentium Pro processors.
 
 

Table 1. Leading Parallel Computers in Late 1998.  Rates given in GFLOPS.

Rmax
Rpeak
Cray/SGI Mountain Blue (1999)  
~4000
IBM Blue Pacific (5800 processors)  
3880
Intel Sandia Red (9152 processors)
1338
1830
Cray T3E-1200E (1080 processors)
891.5
1296
IBM SP/604e (1900 processors)
547
1262
SGI Origin 2000/250MHz (512 processors)
195.6
256