non-binary hardware (was: Absolute size of 'short')

[Mike Albaugh]

>From article <6266 at venera.isi.edu>, by lmiller at venera.isi.edu (Larry Miller):
>     There was also the IBM 1620, a BCD machine.  Yes, decimal, but
>     all arithmetic was performed using table lookup, floating point
>     in software, so I guess it could be called a nonbinary machine.
	There were _lots_ of BCD (and bi-quinary) machines back when
16K was a lot of memory and a large percentage of problems needed mainly
I/O with a few calculations on each item. The size and speed of the decimal
to binary and back conversion routines would have made them the major
bottleneck. Incidentally, one of the neatest things about the 1620 was that
numbers of arbitrary size could be added/subtracted/multiplied/divided (as
long as they fit in memory). I know at least one number-theory freak who
resented the "restriction" of only 256 byte long numbers when we upgraded
to a System/360. Anyway, the 1620 was still "binary", in the sense of any
one signal being in one of two states.
	The Soviets built a balanced base 3 (radix -3) machine in the
late fifties (reported in IEEE transactions, as I recall), but it too
used binary logic, with each digit stored as two bits.
	I believe Signetics built a fast math chip of some sort (Maybe a
Multiplier?) which actually used base four inside (four discrete current
levels in IIL) and converted to binary TTL outside. This was written up
in EDN or the like about 1980.

> Larry Miller				lmiller at venera.isi.edu (no uucp)
| Mike Albaugh ({decwrl!turtlevax!}weitek!dms!albaugh) voice: (408)434-1709
| Atari Games Corp (Arcade Games, no relation to the makers of the ST)
| 675 Sycamore Dr. Milpitas, CA 95035
| The opinions expressed are my own (My lawyer isn't listening)