[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

old intel processors (was Re: Compaq?intel vs API ...)

> From: alvin <alvin@iplink.net>
> Date: Wed, 27 Jun 2001 17:09:10 -0400
> Does anybody remember the intel product iapx423(I think the prefix may
> be wrong). After spending large numbers of Millions of dollars they took
> what they learned and made it the 286.

That's sort-of close but a ways off from actual history.

The iAPX 432 effort started up around or shortly after 1975.  It
was a 32-bit ultra-CISC architecture, with _no_ registers at all.
The instruction stream was bit-addressed, which required a huge
extractor/rotator in the instruction decoder.  There was
essentially no assembly language (other than Ada).  Here's a
quote from an Intel publication: "8086 (1978): Intel's new 16-bit
architecture, originally designed to fill the gap while waiting
for completion of the 32-bit iAPX 432 CPU, ..."  The 432 was way
ahead of its time in many ways.  In hindsight, in other ways, it
was not on the right track.  The 432 development effort ended
shortly after 1983.  The 432 effort and various X86 efforts of
the time undoubtedly cross-pollinated.

You might actually be thinking of the 80960 architecture.
There's a whole book on the 80960 architecture, if you can find a
copy.  It is by Glen Myers and Dave Budde.

The first 960 chip was contemporary with the 386, taping out
about 3 months after the 386.  It had originally been designed
with a byte-addressable very CISC instruction set, but it was
changed to a RISC instruction set late in the design after it was
found that the instruction decoder wouldn't fit on the die (along
with the rest of the design), even at the largest die size that
could be made on the equipment available for the process of that
time.  This first 960 chip had about 1.4X the transistors of the
386 and was rougly 2X as fast.

The second pair of 960 chips were superscalar.  They were roughly
contemporary with the first Pentium (trademark and so on).  They
were (of course) much faster than the first 960 chip.

The 960 architecture was actually a _VERY_ nice 32-bit (and
33-bit) architecture, with four levels of the architecture.  At
the high end, it was designed to do rather elaborate object-based
capability-based addressing and protection.  At the low end, for
embedded use, it could be done without floating point or address
translation of any kind.

The design of the floating point unit from the first 960 chip
found its way (with a few tweaks) into the 387 math coprocessor,
and probably heavily influenced the 486's FP unit.  There was
sharing of ideas and some raw technology between the X86 and 960
groups over the years.

In about 1991, the Oregon 960 design group was disbanded and many
of the engineers worked on the P6 (Pentium Pro).  Some members of
this group had already done the 486DX2 and a few other
proliferation chips.

I worked in the 960 chip design group from 1983 to 1992, then
(with no love for the X86 architecture) on the P6 and Pentium 4
group until 2000.  I use a UP2000 at home.  I find it sad to see
superior architectures die for marketing and political reasons.

Thanks for listening to today's history lesson.  :-)

Robert Riches

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index] []