phipps@fortune.UUCP (Clay Phipps) (03/28/84)
A 64-bit processor could have some commercial applications.
Not only can you manipulate integers in the range
-9_223_372_036_854_775_808 .. +9_223_372_036_854_775_807
(that's nine quintillion, folks !), but you can also manipulate
numbers stored as 8 ASCII digits (i.e., COBOL "display" arithmetic)
or 16 BCD digits (i.e., COBOL "computational-3" [or is it "...-1" ?])
in registers. This can be done rather fast: 64-bit ASCII addition
can be one half the speed of the corresponding 64-bit integer addition.
Now, it's true that the average COBOL program spends most of its time
accessing files and copying memory instead of performing arithmetic,
but the average FORTRAN program also spends more time accessing memory
(including branches, which are just instruction fetches)
than performing arithmetic, and that doesn't stop FORTRAN hackers
from drooling over machines with fast arithmetic.
All of which brings us back to the fundamental "von Neumann Bottleneck":
memory access time (slow) and its relation to processor speed (fast).
-- Clay Phipps
--
{allegra amd70 cbosgd decwrl!amd70 harpo hplabs!hpda
ihnp4 megatest nsc oliveb sri-unix ucbvax!amd70 varian}
!fortune!phipps