MitchAlsup <MitchAlsup@aol.com> writes:
>On Friday, December 30, 2022 at 1:16:36 PM UTC-6, John Levine wrote:
>>
>> But I agree that a lot of people figure that whatever results they get
>> from their FP code is good enough, and the ones who understand
>> numerical analysis budget a lot of time to get uncooperative hardware
>> to get decent results.
><
>This occurred across the FP numeric spectrum including IBM and the
>BUNCH {Boroughs, Univac, NCR, CDC, and Honeywell}--if you were

It is spelled Burroughs.

One of the Burroughs mainframe lines only suppported BCD
(base 10) floating point, with a two digit exponent (-100 to +100)
and a 100 digit mantissa.

On 12/31/2022 12:20 AM, Quadibloc wrote:
> On Friday, December 30, 2022 at 3:50:35 AM UTC-7, Anton Ertl wrote:
>> Quadibloc <jsa...@ecn.ab.ca> writes:
>>> IBM's bread and
>>> butter, their core business, was commercial customers doing data
>>> processing.
>
>> Was this the case in the early 1960s? Certainly the IBM 704 and
>> followons have been much higher-profile than the commerial computers.
>> And even the IBM 1620, a BCD computer with variable number of digits
>> per number, is described as scientific computer.
>
>> It seems to me that scientific uses were a matter of considerable
>> prestige to IBM. Note the reaction of IBM to the CDC 6600.
>
> I could be mistaken, but while computers used for research certainly
> made some money for IBM, my understanding was that even then,
> the dollar volumes were considerably smaller than those generated
> by commercial data processing - which was, of course, IBM's
> business _before_ computers, with unit record equipment - and so
> if it hadn't been for the visibililty and prestige value of advanced
> scientific computers, IBM would have given them less emphasis than
> they did.

Absolutely agreed.

> IBM wasn't a technology company first and foremost. It was, instead,
> fundamentally a *sales* oriented company. It worked by understanding
> the businesses of its customers, and providing software useful to them.

Yes, absolutely. But there was another, frequently, under appreciated,
aspect of their success. The quality of their peripherals, especially
things like printers and card readers was outstanding - again coming
from their history of unit record equipment. High level managers, who
make big decisions like buying new computers, read reports. And having
high speed printers that lined up and printed characters perfectly on
those reports was a valuable attribute not shared by some of their
competitors.

--
- Stephen Fuld
(e-mail address disguised to prevent spam)

On Friday, December 30, 2022 at 4:50:35 AM UTC-6, Anton Ertl wrote:
> Quadibloc <jsa...@ecn.ab.ca> writes:
> >IBM's bread and
> >butter, their core business, was commercial customers doing data
> >processing.
> Was this the case in the early 1960s? Certainly the IBM 704 and
> followons have been much higher-profile than the commerial computers.
> And even the IBM 1620, a BCD computer with variable number of digits
> per number, is described as scientific computer.
>
> Looking for commercial systems, I found the IBM 1401, of which 12000
> were built, and which was rented out starting at USD 2500/month (no
> sales price given). A 7090 could be rented for USD 63500 in 1960;
> according to <http://gunkies.org/wiki/IBM_7094>, 255 IBM 7094 were
> produced; this means that the IBM 7094 generated roughly half the
> revenue of the IBM 1401.
> >If the 360 sort of worked for scientific uses, without necessarily
> >being optimized for them, that was good enough.
> It seems to me that scientific uses were a matter of considerable
> prestige to IBM. Note the reaction of IBM to the CDC 6600.
>
> My guess is that many users did not care much about numerical
> properties of the FP implementations, so computer architects did not,
> either (cf. Seymour Cray's take on such issues). Stuff like
> denorms-to-zero and compiler options like -ffast-math show that little
> has changed.
> - anton
> --
> 'Anyone trying for "industrial quality" ISA should avoid undefined behavior.'
> Mitch Alsup, <c17fcd89-f024-40e7...@googlegroups.com>

The early history (50s & 60s)of BCD versus binary (my opinion):
Early business applications didn't do much calculation, so the efficiency of binary was not
enough to overcome the cost of converting between the two. Besides people were
more comfortable with decimal. I used both the IBM1620 and the CDC1604, in Fortran.
No matter the relative slowness of the 1620, it was faster and more accurate than a slide rule.
There were few scientific computers using BCD, the Univac LARC being the most notable (and the last).

One of my fantasy architectures is radix 100 and/or 128 with variable length data. A way to
do both worlds. Variable length data support eliminates all the extended length subroutines
on a small simple machine (eg the 8 and 16 bit processors). Viability???

According to Quadibloc <jsavard@ecn.ab.ca>:
>I could be mistaken, but while computers used for research certainly
>made some money for IBM, my understanding was that even then,
>the dollar volumes were considerably smaller than those generated
>by commercial data processing - which was, of course, IBM's
>business _before_ computers, with unit record equipment - and so
>if it hadn't been for the visibililty and prestige value of advanced
>scientific computers, IBM would have given them less emphasis than
>they did.

Until the 1940s, a computer was a person, usually a woman, who used a
desk calculator and a paper spreadsheet to do relatively complex
calculations. Howard Aiken had persuaded IBM to fund and build the
ASCC better known as the Mark I in 1937, and the Manhattan project
used a lot of IBM punch card equipment to do physics calculations.
(Wikipedia says one of the first Mark I programs was a Von Neuman bomb
design problem in 1944.) IBM built the SSEC in the late 1940s and then
the Defense Calculator which became the 701 in the early 1950s. In the
late 1950s the STRETECH (7030) was a very large scientific machine
that was a commercial failure but produced technology used in 709x and
S/360.

So even though commercial data processing was IBMs cash cow, they had
been involved in scientific computing as long as there was scientific
computing.

Also keep in mind that the line between scientific and commercial
computing has always been fuzzy. SABRE was certainly a commercial
project, but it ran on a pair of 7090s.

Finally, remember that the goal of the S/360 project was to replace
their many incompatible product lines with one, so the had to have
something to offer the 709x customers. Too bad the floating point was
so bad, although I think that even if the FP had been as good as the
7090's, which was quite good, the Cray would still have beaten them on
price/performance and the numerical guys would have dealt with it.

--
Regards,
John Levine, johnl@taugh.com, Primary Perpetrator of "The Internet for Dummies",
Please consider the environment before reading this e-mail. https://jl.ly