Van Snyder wrote:
> I'm curious about present and near-future 64-bit
> workstations. Maybe you
> can help me. I'm interested in the following questions for HP/Intel
> Merced, HP PA-8000, and offerings from Sun, DEC, SGI and IBM:
>
> 1. What workstations/chips are "64-bit"?
>
> 2. Does "64-bit chip" mean it has a 64-bit (or wider) processor bus?
>
> 3. 64-bit integers? Bigger (and smaller) ones?
>
> 4. Is single-precision floating point 64 bits? Is it IEEE?
>
> 5. What size is double-precision floating-point?
Rather than answer each of these individually, I'll explain what is
typically meant by "64-bit" when referring to CPU architectures. Briefly,
it means that the processor supports virtual memory with a 64-bit address
size. The size of numeric datatypes is not relevant. More to the point, a
64-bit processor doesn't "automatically" double the size of classic numeric
types. While there may indeed be (and usually is) a 64-bit integer type,
most implementations I'm aware of still use 32-bit integers as the "default
INTEGER" type. (But most also have an option to select 64-bit integers as
the default as well.)
Floating point types are unaffected by the address size of the processor,
and no, REAL isn't automatically DOUBLE PRECISION on such a processor.
Consider that 16-bit processors had 64-bit floating types available.
The size of the processor bus is also not relevant to the term "64-bit".
SGI was first to market with a 64-bit processor, but OS support for 64-bit
virtual address spaces didn't follow until several years later. Digital's
(now Compaq's) Alpha was first with a complete 64-bit workstation solution.
HP (PA-RISC), Sun (UltraSPARC) and IBM (POWER3) also brought out 64-bit
systems over time, though again OS support tended to lag the chip
capability. Intel doesn't have a 64-bit processor product - Merced is still
vapor.
> BTW, is there an IEEE standard for binary floating point with
> more than
> 80 bits, either in place or under development?
I'm not aware of such. All the vendors with 64-bit workstations (not Intel)
support some sort of "quadruple precision" datatype. Other than IBM, the
format used is "IEEE style", with 15 bits of exponent and 113 bits of
fraction (1 hidden). IBM uses two 64-bit values with an overlapping
fraction to give extended precision but not extended range (at least this is
how it was the last time I looked into it - this may have since changed.)
Steve Lionel
Fortran Engineering
Compaq Computer Corporation
Nashua, New Hampshire
[log in to unmask]
Compaq Fortran web site: http://www.compaq.com/fortran
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|