[COFF] Architectures -- was [TUHS] 68k prototypes & microcode

[Clem Cole]

Moved to COFF  - and I should prefix this note with a strongly worded --
these are my own views and do not necessarily follow my employers (and
often have not, as some of you know that have worked with me in the past).

On Wed, Feb 3, 2021 at 8:34 PM Larry McVoy <lm at mcvoy.com> wrote:

> The x86 stuff is about as far away from PDP-11 as you can get.  Required
> to know it, but so unpleasant.
>
BTW: Once again we 100% agree *on the architecture part of* *the discussion*.
And frankly pre-386 days, I could not think how anyone would come up with
it.  As computer architecture it is terrible, how did so many smart people
come up with such?  It defies everything we are taught about 'good'
computer architectural design.  But ....   after all of the issues with the
ISA's of Vax and the x86/INTEL*64 *vs.* Alpha --- is how I came to the
conclusion, *architecture does not matter nearly as much as economics and
we need to get over it and stop whining.  * Or in Christensen's view, a new
growing market is often made from a product that has technically not as
good as the one in the original mainstream market but has some value to the
new group of people.

x86 (and in particular once the 386 added linear 32 bit addressing), even
though DOS/Windows sucked compared to SunOS (or whatever), the job (work)
that the users needed to do was performed to the customer's satisfaction *and
for a lot less.*  The ISVs could run their codes there and >>they<< sell
more copies of their code which is what they care about.  The end-users,
really just care about getting a job done.

What was worse was at the time, it was the ISV's keep their own prices
higher on the 'high-value platform' - which makes the cost of those
platforms ever higher.  During the Unix wars, this fact was a huge issue.
The same piece of SW for a Masscomp would cause 5-10 more than a Sun -- why
we were considered a minicomputer and Sun was a workstation.   Same 10MHz
68000 inside (we had a better compiler so we ran 20% faster).   This was
because the ISV's classified Masscomp's competition was considered the Vax
8600; not Sun and Apollo -- sigh.

In the end, the combination of x86 and MSFT did it to Sun.   For example,
my college roommates (who were trained on the first $100K
architecture/drawing 3D systems developed at CMU on PDP-11/Unix and Triple
Drip Graphic's Wonder) Wintel running a 'boxed' AutoCAD was way more
economical than a Sun box with a custom architecture package -- economics
won, even though the solution was technically not as good.    Another form
of the same issue did you ever try to write a technical >>publication<<
with Word (not a letter or a memo) -- it sucks -- The pro's liked FrameMaker
and other 'authoring tools' (hey I think even Latex and Troff are -- much '
better' for the author) -- but Frame costs way more and Word, so what do
the publishers want -- ugh Word DOC format [ask Steinhart about this issue,
he lived it a year ago].

In the case of the Arm, Intel #$%^'ed 101-15 yrs ago up when Jobs said he
wanted a $20 processor for what would become the iPhone and our execs told
him to take a hike (we were making too much money with high margin
Window's) boxes.   At the time, Arm was 'not as good' - but it had two
properties Jobs cared about (better power - although at the time Arm was
actually not much better than the laptop x86s, but Apple got Samsung to
make/sell parts at less than $20 -- i.e. economics).

Again, I'm not a college professor.  I'm an engineer that builds real
computer systems that sometimes people (even ones like the folks that read
this list) want to/have wanted buy.   As much as I like to use sold
architecture principle to guide things, the difference is I know be
careful.  Economics is really the higher bit.  What the VAX engineers at
DEC or the current INTEL*64 folks (like myself) was/is not what some of the
same engineers did with Alpha -- today, we have to try to figure out how to
make the current product continue to be economically attractive [hence the
bazillion new instructions that folks like Paul W in the compiler team
figure out how to exploit, so the ISV's codes run better and can sell more
copies and we sell more chips to our customers to sell to end users].

But like Jobs's, DEC management got caught up in the high margin game, and
ignored the low end (I left Compaq after I managed to build the $1K Alpha
which management blew off -- it could be sold at 45% margins like the Alpha
TurboLaser or 4x00 series).   Funny, one of the last things I had proposed
at Masscomp in the early 80s before I went to Stellar, was a low-end system
(also < $1K) and Masscomp management wanted to part of it -- it would have
meant competing with Sun and eventually the PC.

FWIW:  Intel >>does<< know how to make a $20 SOC, but the margins will
suck.  The question is what will management want to?   I really don't
know.  So far, we have liked the server chip margins (don't forget Intel
made more $s last year than it ever has - even in the pandemic).

I feel a little like Dr Seuss'  'Onceler' in the Lorax story ... if Arm can
go upscale from the phone platform who knows what will happen - Bell's Law
predicts Arm displaces INTEL*64:

“Approximately every decade a new computer class forms as a new “minimal”
computer either through using fewer components or use of a small fractional
part of the state-of-the-art chips.”

FWIW:  Bell basically has claimed a technical point, based on Christenson's
observation; the 'lessor' technology will displace the 'better one.'   Or
as I say it, sophisticated architecture always losses to better economics.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://minnie.tuhs.org/pipermail/coff/attachments/20210205/505a72e2/attachment-0001.htm>