25 Oct
2010
25 Oct
'10
2:22 p.m.
On 25 Oct 2010 at 17:14, William Donzelli wrote:
IBM did a study at the start of the 1960s and found
that most data out
in the real world was financial records made of mostly BCD, and having
a word length of 32 bits made a great deal of sense, both in storing
and moving data. I think Amdahl originally wanted the S/360 to be
based on a six bit character, but lost the argument to the study.
Is the study available for perusal somewhere? (e.g. FJCC) I'd like
to see their logic.
32 bit was a terrible choice when your floating point word was 8 bit
exponent and 24 bit mantissa, normalized to a 4 bit boundary,
however.
--Chuck