Why MIPS aren´t the point

Since recent advances in microprocessors have resulted in narrowing the gap between personal computers and supercomputers, the author gives his opinions as to what is needed in such systems in addition to, or instead of, raw processor performance. It is argued that all common presently used benchmark programs for testing new microprocessor systems should reflect the amount of work completed by an operator and machine combination. For a potential desktop supercomputer four areas must be considered when trying to determine how usable a computer is: (1) human factors-is the machine nice to work with? (2) programming-can one use a favorite language or operating system? (3) system balance-can one actually use all the system capabilities most of the time? (4) growth potential-is there a painless upgrade path? Each of these points is briefly explored. Trading peak performance for these features should result in better overall performance.<>