Re: [math-fun] When I'm 64

8 Jun 2017

      On 2017-06-07 18:54, Tom Knight wrote:
...
One of the machines the Livermore S1 project was supposed to replace
was a weird variant of the ANFSQ-7, a 32 bit machine.  But an Admiral
apparently heard about the wonders of parity checking, and insisted
his new machine have parity. The new machine became a 30 bit machine,
with two bits stolen for parity. Tasteful.
AN/USQ-17.  I was told in St. Paul that it was the last UNIVAC Seymour Cray
ever designed.  He refused to cooperate with the rest of the designers, who
then came up with nothing, so UNIVAC had to build Cray's.  It evolved to
the UNIVAC 1206, via the similar looking
https://en.wikipedia.org/wiki/AN/USQ-20#/media/File:AN_USQ-20.jpg .
It was incredibly slow, 115μec to divide, but it was built like a bank
vault.
Those bay doors had meshing brass teeth.  It was air cooled with a
deafening scream.  You could Bernoulli-levitate a football in the exhaust
from the stack.

I wrote it a DDT-style interactive debugger called NIP, which stood for
nothing in particular.  I had to lobby really hard for UNIVAC to
disconnect a device built into the FlexoWriter stand that disabled input
for many seconds following output, so that no one could waste precious
machine time interactively.

The console was amazing, availing every flipflop in the processor as an
illuminated button.  The CPU
had a crazy RPT instruction that repeated the succeeding instruction
a number of times specified by the address field, optionally bumping
the address field of the repeatee.  With the machine halted, it was
possible to set up a zero-instruction "program" to single-step compute
small factorials: RPT ADVANCE MULTIPLY IMMEDIATE 1.

Some years ago I war-storied this list about the Big Red Fault Light.

These were 1's complement machines that (usually) evaded minus 0 via
hardware subtract with end-around borrow, instead of the usual adder
with end-around carry.

UNIVAC remained unable to design a new processor, so they kept
repackaging Seymour's
with hilariously marketerized instruction names, concealed by a
"Real time, high level language, problem oriented compiler" that was
actually a braindead assembler.  To index location foo, you had to write
RPL·Y+1·W(FOO) .  Of course the "compiler" for some reason would
not let you add 2 via RPL·Y+2·W(FOO).  UNIVAC never told its
programmers there was no add 2 instruction.

They renamed all the instructions yet again, selling now a "business
 machine" under the names M460 and M490.  30 bits.  Half a billion cents.
Do > $5 million in business and suddenly you're $5 million in the red.

At my Navy job, the 1206 supplanted the even weirder P(ackard) B(ell)
250, http://www.cca.org/tech/rcs/pb250.html , 22 bits/wd.  Totally
volatile magnetostrictive delay line (serial) memory.  Even the active
registers!  The divide instruction took 3 MILLISEC, unless you "sequence
 tagged" it, meaning to skip over as many instructions as the number of
specified quotient bits!  Depending on signs, there were, I think, seven
ways the quotient and remainder could be wrong.  But the machine had a
completely reliable square root instruction!  It is atrocious that
subsequent
computers didn't.

Memory was divided into 256 word loops.  Except some for data were
255 and some were 257, so a trivial program could modify a table by
repeatedly modifying the "same" location.  We programmed in octal.
--rwg
...
...
On Jun 7, 2017, at 8:48 PM, Bill Gosper <billgosper@gmail.com> wrote:
The MIT Life hackers once got an absolutely surreal demo of the "PDP-3"
in
...
...
the basement of Searle Medidata, out on rte 128, by it's sole caretaker,
builder(?),
and programmer Charles Corderman.
If memory serves, some salesman leaked that DEC was building a 36 bit
challenge
to IBM's mainframes, and Searle staked much on this belief.  When it
became
clear that DEC was abandoning the project, Searle insisted on seeing the
plans,
and charged Corderman with completing them.  The machine computed parity
nearly
everywhere.  What a waste of bits.  So Corderman changed it to a 37bit
machine!
He then built an *entirely* menu-driven OS, where the only pointing
device
was
a light pen!  To make things weirder, the Type 30 display had a purple
phosphor.
On the console was a Flexowriter, powered off.  We expressed suspicion
that
he had programmed his (utterly peculiar) OS on it.  "No, but it does
work,"
he
said, powering it on, and waving his light pen.  The Flexowriter typed a
lowercase
"r", and he powered it off.
(Kids:  The only data from a light pen is a single bit, saying whether or
not it saw
the most recent point flashed on the (non rastered) screen.  During one
second,
the machine was only fast enough to light a very few thousand of the 2^20
addressable points.  The pen's field of view depended on how close it was
to
the screen, but it needed to be pretty close.  How can you possibly do
anything useful with such a thing?)
This is the machine with which Corderman found his famous Switch Engines,
http://www.conwaylife.com/w/index.php?title=Switch_engine , bizarre c/12
diagonal puffers, one of which shoots (Doppler shifted) gliders out the
front!
Or did I dream all this?  --rwg
On 2017-06-07 16:39, Fred Lunnon wrote:
...
Manchester Atlas I with 48-bit words.    WFL
On 6/8/17, Eugene Salamin via math-fun <math-fun@mailman.xmission.com>
wrote:
...
My first computer was the IBM 650.  It had 2000 10-decimal digit words
stored on a rotating magnetic drum.  Senior year in high school class
in
applied math in 1959, IBM gave us access to their computer.  It was
punch
cards, card sorters, and line printers.
--  Gene
On Wednesday, June 7, 2017, 3:52:26 PM PDT, Victor Miller
<victorsmiller@gmail.com> wrote:
And 60 bit (cdc).  I also programmed a univac 418 -- 18 bit 1's
complement
On Wed, Jun 7, 2017 at 18:40 Tom Knight <tk@mit.edu> wrote:
...
You left out the 40 bit machines…
...
On Jun 7, 2017, at 3:50 PM, Henry Baker <hbaker1@pipeline.com> wrote:
FYI --
"Will you still need me, will you still feed me
When I'm 64?"
I've now lived through:
decimal computers (IBM 1620, IBM 1401)
36-bit/15-18bit-address computers (IBM 7040, IBM 7090, GE635, PDP-10,
DEC-20)
12-bit computers (PDP-8)
Microcoded machines (Xerox Alto, Lisp Machines)
1-bit computers (ICL DAP 32x32x1 "square word", Connection Machine)
4-bit computers (Intel 4004)
8-bit computers (6502, Z-80, 8080, 8051 (still used inside "USB flash
drives"!), 6800, etc.)
16-bit computers (PDP-11, Moto 68000, 8086, etc.)
32-bit computers (IBM 360, VAX, 80386, ARM, etc.)
64-bit computers (everyone!)
It remains to be seen if I'll make it to 128 or 256...
Son of a bits!
---
https://apple.slashdot.org/story/17/06/07/0538233/apple-to-phase-out-32-bit-...
...
...
...
...
...
Apple To Phase Out 32-Bit Mac Apps Starting In January 2018
Apple will be phasing out 32-bit apps with iOS 11, and soon the
company
will make the same changes on its macOS operating system.  During its
Platform State of the Union keynote at the Worldwide Developers
Conference,
Apple told developers that macOS High Sierra will be the "last macOS
release to support 32-bit apps without compromises."

math-fun mailing list
math-fun@mailman.xmission.com
https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
_______________________________________________
math-fun mailing list
math-fun@mailman.xmission.com
https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun

Bill Gosper

tags

participants (1)