Nat Semi ns32k, was Re: Whitechapel Computer Works MG-1

Steven M Jones classiccmp at crash.com
Mon Dec 1 00:04:04 CST 2014


On 11/30/2014 05:21 PM, Chuck Guzis wrote:
>
> I never got to that point--the NSC chip was so late that it wasn't
> even a candidate.  I seem to recall some discussion that the early
> 16032 chips were a little on the flaky side, but that may have been
> rumor.

It was a fact, but even if/when they (mostly) fixed it the "rep"
continued to dog them. There are certainly indications they had those
kinds of issues with products besides the 32k family.

And I think they always struggled to achieve the higher-clocked yields
and performance they needed across designs. This is all based on odd
bits of reading over the years as I wasn't interested in (and had barely
heard of) the ns32k until Bruce Culbertson shared his homebrew 32016
design around 1990 - long after the competition had left Nat Semi behind.



On 11/30/2014 04:54 PM, Jules Richardson wrote:
>
> How did they compare in price with the m68k? Because that chip (pretty
> much) "just worked", so if the price was comparable then I can
> certainly see why it gained such acceptance and the ns32k line fell by
> the wayside. 

I'm not clear on the pricing, but the ns32k was designed and delivered
as a complete chipset to support multi-user, virtual memory systems with
a 32bit-wide CPU design. Even though the MC68000 was slower at 32bit
operations it was available several years earlier, and (I gather) was at
least equally targeted at "standalone" use like a traditional MPU, or
using 6800-family peripherals, so that as the price curve fell it could
move into mid- and low-end applications and develop an enormous volume
advantage... And then you've got a solid track record, vast adoption,
and the forward-compatibility of the instruction set with the
fully-32bit 68020 when it shows up circa 1984. Even though the 32032 was
released that year, and the 32016 had been out for two years, NSC still
had some nasty errata and only a 24-bit physical address space, and were
both much slower than the '020.

You can certainly find examples of the 32016 being used stand-alone, but
not with anything like the ubiquity of the 68000...

In terms of performance, you can still find a lot of lively "discussion"
in a NetNews archive that includes net.micro.16k, net.micro.32k, and
comp.sys.nsc.32k. The 32032 only outpaced the 32016 when the 32bit wide
data bus came into play despite coming two years later, and still topped
out at 10MHz. The 68020 seems to have beaten them both handily clock for
clock, and of course that eventually went up to what, 30MHz? The 32332
arrived in 1985 but only reached 15MHz and was never more than 50%
faster best case than the '032. The 32532 of 1987 was a complete
redesign and finally seemed to trump the 68030 and 80386 - only to be
pummeled by the emerging RISC designs...


Despite the issues with early silicon, there were some notable adopters.
Encore and Sequent delivered multiprocessor systems based on the ns32k
family. Siemens-Nixdorf built a few models running SINIX.  Definicon and
OPUS shipped add-in boards for PCs, using the x86 for I/O. Tektronix
delivered a couple models of UTek workstations. More notable than
"successful" since this wasn't commercial, but Niklaus Wirth et al at
ETH Zürich built their Ceres workstations using the ns32k family.

Bill Jolitz, later of 386BSD fame, built and sold the very portable
Symmetric S/375 using a 32016 chipset (and running 4.2BSD). And you've
got others like Whitechapel, Acorn, BBC Micro - CompuPro even shipped an
S-100 32016 board.

But for my money the most interesting two designs were Bruce
Culbertson's homebrew 32016 design, and the pc532 by Dave Rand and
George Scolaro (both of whom had been part of the Definicon team).
Culbertson ported MINIX to his design, and later to the pc532. Both
designs were freely available, but Dave and George organized a run of
around 200 of the complex multi-layer pc532 system board in the early
90s and a group-buy of the more expensive parts like the processor
complex. The pc532 would eventually run MINIX, Mach, and NetBSD/pc532.

BTW - If you've got a spare pc532 sitting in a closet, I'd like to get
one... ;)



On 11/30/2014 04:54 PM, Jules Richardson wrote:
> One of these days, I think I'd quite like to find one of the FPUs

The NS32081 enjoyed some popularity as an FPA with non-ns32k systems. In
fact the only application note in National's 1984 Series 32000 Databook
interfaces the 32081 with a 68000...



On 11/30/2014 05:46 PM, Al Kossow wrote:
>
> National released 4.1BSD for it. Opus Systems had it running on their
> PC coprocessor card.

Nat Semi also offered some flavor of System V Release 2 by 1986-87. And
I think SysV was offered by OPUS and some other OEMs.

Never have found out what software - if any - CompuPro offered for their
32016 board.



--Steve.



More information about the cctalk mailing list