On Sun, May 9, 2021 at 3:58 PM Larry McVoy <lm@mcvoy.com> wrote:

> Pretty sure Bill Jolitz either had a 32016 Unix or he was also selling
> similar boxes.  National couldn't get it together to produce bug free
> chips or maybe we'd all be running that, pretty nice architecture
> (in theory).
>
He did.  It was about the size and form factor of the Compaq 'luggable.'
IIRC hBill used a flavor of the multibus, which made a large power supply
and more expensive peripherals (although Sun, Masscomp, and Apollo were
also using Multibus in those days -- as Multibus was cheaper than the
Unibus or QBUS).    Bill's computer showed up at about the same times as
the Sun-2, which was 68010 based.

As Yale Pratt once said of the NS32016 ISA, 'it was a VAX/780 cleaned up
but on a single chip'.   Still technically a CISC, but did had fewer
specialty features (more like PDP-10 in that what Bill Wulf used to call a
'regular and complete' architecture).  Conceptually the ISA more powerful
than the 68000 ISA, but it lacked a good compiler (*a.k.a.* 'production
quality' compiler) that could exploit it, and as you point out, Nat Semi
had a difficult time producing them.    Funny, the compiler issue was SOP
for the chip folks in those times as they had traditionally relied on 3rd
parties to build the compiler for them - they would put out an assembler
(usually written in Fortran-IV), make the assembler 'open source' and let
the ecosystem go from there.  Compiler houses (such as Green Hills, MCA,
Frieberghouse, and PG) had to decide if they wanted to invest in that new
ISA or not if they could not get someone else to pay them to write one for
it.  The systems folks (like DEC/DG knew that was not good enough).   Sun
was just awakening to that fact and I >>think<< it would have been around
then that they started their real compiler team (Rob G -- you probably know
more than I do here).  Anyway, I always felt it was the compiler that
killed them as much as Nat Semi's well-known Si production issues.

Also, so many of us were ignoring the PC because it was based on the x86,
that we did not realize that the PC's ISA bus was 'good enough (Masscomp
never built anything with it, and Apollos did not until they finally did
the DN3000 a couple of years later and I think it was not until RR that Sun
tried).  But in the end, the 386 'won' for economic reasons -- by being on
the PC ecosystem much cheaper peripherals snd eventually got the ability to
solve the addressing issues so you could run large programs on it, so the
fact that it had the worst ISA did not matter.

I've always wondered if a Nat Semi NS32016 based system running in a PC/AT
form factor had appeared that was priced like a PC/AT if that might have
had a chance.

Clem


ᐧ
ᐧ
ᐧ