The Power of the NORC 3752

wrote, in part:

You have a very good point, although I tend to be inclined to disagree with you. (This doesn't mean I do disagree with you, just that I wish that I could.)

Since it is "easier" to perform fixed-point arithmetic than floating-point arithmetic - less circuitry is required - it is sensible to say that one can't compare MIPS to MFLOPS on a one-to-one basis. So I left out the AN-FSQ-7 primarily because it couldn't be compared well to the NORC, which *did* have floating-point.

But I do tend to feel that special-purpose computers, whether like the AN-FSQ-7, optimized for fast vector arithmetic for air defense, and, even more so, the COLOSSUS, a special-purpose calculating device designed to find Schlusselzusatz pinwheel positions, are peripheral to the history of computing.

Fixed-point arithmetic, though, certainly doesn't have this characteristic. Most things done with a computer do use just integer arithmetic. Or they don't involve arithmetic at all, except for the implicit subtraction involved in comparing things by their collating order.

An awful lot of computing is data processing, and then there's word processing, and, of course, operating systems do computing, but they seldom need to do any floating-point calculations.

I would like to see machines become available which provide more raw power for things like string searching, for example.

But, in general, a supercomputer is a device for handling very big *compute-bound* problems. And the people with *those* kinds of problems so far tend to be scientists and engineers, who definitely do use a lot of floating-point when they program.

While I am inclined, therefore, to look in the direction of number crunchers when thinking about supercomputers, I do acknowledge that you are right: all very powerful computers are deserving of recognition, whatever their preferred domain of application.

John Savard Usenet Zone Free Binaries Usenet Server More than 140,000 groups Unlimited download