IBM System/32, System/34 implementation technology?
Eric Smith
System/32 and System/34 minicomputers, or any other small IBM systems
from the early to mid-70s. I'm specifically interested in details of
the semiconductor technology, degree of integration, and processor or
system architecture.
So far I'm not turning up a lot. I had hoped to find some articles in
the IBM Journal of R&D, or the IBM Systems Journal, but the bibliography
database at the University of Utah doesn't show anything promising.
There is brief mention of some of this stuff in _IBM's 360 and Early 370
Systems_ by Pugh, Johnson, and Palmer, but of course it is really
outside the scope of that book. If there were a similar book that
covered small systems, it would be very useful to me.
Any ideas?
Thanks!
Eric
[If for some reason you want to send me a private reply, you'll have to
remove the obvious anti-spam cruft from my email address. Otherwise,
I'm perfectly happy to watch for followups in this newsgroup.]
Del Cecchi
What do you want to know? I and several of my friends who are still around here
designed the things.
--
Del Cecchi
cecchi@rchland
Del Cecchi
>
> The first systems I worked on were the '32 and '34. That was a long
> (very long) time ago.
You must be REALLY OLD.
>
> The circuit technology for both machines was an IBM bipolar technology
> that had 100 to 120 circuits per chip. Ok for it's time. It's
> internal code name was Dutchess.
It had 134 gates, 60 three way NAND, 40 4Way NAND and 34 2way nor off
chip drivers. Gate delay was about 10 ns.
>
> The processor on the '32 was a 16 bit register based machine that is
> philosophically similar to RISC. All logical and arithmetic
> operations are performed with register to register operations (16
> 16bit registers). Branches used the side effects of the register to
> register operations or tested registers directly. Memory operations
> only load data to registers or store data from registers. Memory was
> logically divided into an instruction space and a data space (aka
> control store and main store). On the '32 the two spaces were
> physically implemented as one. There was no cacheint, branch
> prediction, etc. The internal name of the processor was the CSP
> (Control Store Processor).
>
> The kernal and I/O processing of the '32 was implemented in the native
> instruction set of the CSP. The upper levels of the operating system
> and application code were in the System/3 instruction set. The '32's
> RISCish processor implemented an emulator of the S/3 instruction set
> to run the code.
>
> The S/3 instruction set is almost diametrically opposed to the '32
> instruction set. It is very memory to memory oriented with very few
> registers. All logical and arithmetic operations in the S/3
> instruction set are performed on memory to memory operations. The few
> (2) registers were used to index addresses to memory.
>
> All of this emulation was ok for one user, but pretty slow is you
> wanted to run multiple users. So the '34 took the emulator out of
> software and put it into hardware (Dutchess technology).
> This processor was called the MSP (Main Store Processor). To be
> clear, the '34 had two processors in its "CPU". The original CSP from
> it's System/32 heritage and the MSP. Neither processor supported
> virtual memory.
>
> The I/O was also interesting on the '34 since it introduced the IOP
> (I/O Processor to the mix). But that's another story.
>
> Larry
>
> PS: Don't know of any books on the subject. We were all too busy.
> :-)
Sheeit, we were all too paranoid that someone would steal all our
"secrets".
>
> In article <qh3e2tb...@ruckus.brouhaha.com>, Eric Smith
> <eric-no-s...@brouhaha.com> wrote:
> >I'm searching for articles (or books) on the implementation of the IBM
> >System/32 and System/34 minicomputers, or any other small IBM systems
> >from the early to mid-70s. I'm specifically interested in details of
> >the semiconductor technology, degree of integration, and processor or
> >system architecture.
I offered to help and heard nothing. I figure he is sitting in front of
his email waiting for stuff to fall from the sky.
> >
> >So far I'm not turning up a lot. I had hoped to find some articles in
> >the IBM Journal of R&D, or the IBM Systems Journal, but the bibliography
> >database at the University of Utah doesn't show anything promising.
> >
> >There is brief mention of some of this stuff in _IBM's 360 and Early 370
> >Systems_ by Pugh, Johnson, and Palmer, but of course it is really
> >outside the scope of that book. If there were a similar book that
> >covered small systems, it would be very useful to me.
> >
> >Any ideas?
> >
> >Thanks!
> >
> >Eric
> >
> >[If for some reason you want to send me a private reply, you'll have to
> >remove the obvious anti-spam cruft from my email address. Otherwise,
> >I'm perfectly happy to watch for followups in this newsgroup.]
>
> ---------------------------------------------------------
> IBM Server Group - Architecture, Performance and Design
> home: ldwhit -at- ibm.net
> work: ldw -at- us.ibm.com
> (Note the address modifications to reduce the spam load.)
del cecchi
Larry Whitley
(very long) time ago.
The circuit technology for both machines was an IBM bipolar technology
that had 100 to 120 circuits per chip. Ok for it's time. It's
internal code name was Dutchess.
The processor on the '32 was a 16 bit register based machine that is
Larry
In article <qh3e2tb...@ruckus.brouhaha.com>, Eric Smith
<eric-no-s...@brouhaha.com> wrote:
>I'm searching for articles (or books) on the implementation of the IBM
>System/32 and System/34 minicomputers, or any other small IBM systems
>from the early to mid-70s. I'm specifically interested in details of
>the semiconductor technology, degree of integration, and processor or
>system architecture.
>
Eric Smith
technology, Del Cecchi <dce...@ibm.net> wrote:
> I offered to help and heard nothing. I figure he is sitting in front of
> his email waiting for stuff to fall from the sky.
And I appreciate your offer to help. But since I know very little about
the machines to start with, I don't even have very specific questions.
That's why I was hoping someone had written a book or at least published
some papers in technical journals.
I'm interested in these systems for two reasons:
(1) I may have an opportunity to acquire a System/34, but I won't get
manuals and software with it. I was wondering if I would be able to make it
do anything. At the moment the answer appears to be "no".
(2) You had previously mentioned that the 5100 hardware used similar
components to System/32 and System/34. So it seems logical that if I can
find published material on the latter, it might help explain the former.
Was the "Dutchess" bipolar logic technology based on non-saturating circuits
(like TTL), or saturating (like ECL)? Was it electrically compatible with
an industry-standard logic family, such as 74xx TTL or 10K ECL? Since my
5100 only has the minimum RAM configuration (16 Kbytes), it would be nice to
build a new RAM card for it.
Speaking of which, are the module connectors that IBM used in those products
still being used today? Are they available as a standard product from
connector companies (such as AMP), or are they unique to IBM?
The PALM ("Put All Logic in Microcode") processor used in the 5100 sounds
similar to Larry Whitley's description of the System/32 CSP. Is there in fact
a close relation? Was the PALM designed to be the processor for a low end
System/32 machine? Was PALM used in any other products?
And what were the UC.5 and UC1 processors? What machines used them?
Paul DeMone
[snip]
> Was the "Dutchess" bipolar logic technology based on non-saturating circuits
> (like TTL), or saturating (like ECL)?
Ooops you have that backwards. Saturating a BJT is bad for fast logic
because it greatly increased charge storage and slows it down.
BTW, the IBM stuff doesn't sound like ECL because 10 ns prop times
is way too high.
[snip]
--
Paul W. DeMone The 801 experiment SPARCed an ARMs race of EPIC
Kanata, Ontario proportions to put more PRECISION and POWER into
dem...@mosaid.com architectures with MIPSed results but ALPHA's well
pde...@igs.net that ends well.
Del Cecchi
>
> In response to my query about System/32 and System/34 implementation
> technology, Del Cecchi <dce...@ibm.net> wrote:
> > I offered to help and heard nothing. I figure he is sitting in front of
> > his email waiting for stuff to fall from the sky.
>
> And I appreciate your offer to help. But since I know very little about
> the machines to start with, I don't even have very specific questions.
> That's why I was hoping someone had written a book or at least published
> some papers in technical journals.
I think you are out of luck, sorry. A bunch of IBM Manuals but they are
obsolete.
>
> I'm interested in these systems for two reasons:
>
> (1) I may have an opportunity to acquire a System/34, but I won't get
> manuals and software with it. I was wondering if I would be able to make it
> do anything. At the moment the answer appears to be "no".
I think that is correct unless somebody has something laying around
their cabinet somewhere.
>
> (2) You had previously mentioned that the 5100 hardware used similar
> components to System/32 and System/34. So it seems logical that if I can
> find published material on the latter, it might help explain the former.
The logic was a "masterslice" or gate array as it is called now.
>
> Was the "Dutchess" bipolar logic technology based on non-saturating circuits
> (like TTL), or saturating (like ECL)? Was it electrically compatible with
> an industry-standard logic family, such as 74xx TTL or 10K ECL? Since my
> 5100 only has the minimum RAM configuration (16 Kbytes), it would be nice to
> build a new RAM card for it.
It was shottky ttl, ran off 5 volts and is 74xx compatable.
>
> Speaking of which, are the module connectors that IBM used in those products
> still being used today?
No.
Are they available as a standard product from
> connector companies (such as AMP), or are they unique to IBM?
I think they are unique. How many do you need?
>
> The PALM ("Put All Logic in Microcode") processor used in the 5100 sounds
> similar to Larry Whitley's description of the System/32 CSP. Is there in fact
> a close relation? Was the PALM designed to be the processor for a low end
> System/32 machine? Was PALM used in any other products?
No No No (other than 5110 and 5120)
>
> And what were the UC.5 and UC1 processors? What machines used them?
Other IBM microprocessors UC==Universal Controller
Various controllers primarily used with mainframes like 3081.
del cecchi
Julian Thomas
>Other IBM microprocessors UC==Universal Controller
>Various controllers primarily used with mainframes like 3081.
Then there was the infamous 5 bit engine used in the 303x consoles!
--
Julian Thomas: jt 5555 at epix dot net http://home.epix.net/~jt
remove numerics for email
Boardmember of POSSI.org - Phoenix OS/2 Society, Inc http://www.possi.org
In the beautiful Finger Lakes Wine Country of New York State!
-- --
To a cat, "NO!" means "Not while I'm looking."
Eric Smith
> Was the "Dutchess" bipolar logic technology based on non-saturating circuits
> (like TTL), or saturating (like ECL)?
Paul DeMone <pde...@igs.net> writes:
> oops you have that backwards. Saturating a BJT is bad for fast logic
> because it greatly increased charge storage and slows it down.
You're right, of course. I guess I need to proofread my postings a little
better against "thinkos".
> BTW, the IBM stuff doesn't sound like ECL because 10 ns prop times
> is way too high.
Good point.
Frank A. Adrian
Eric Smith wrote in message ...
+AD4APg- BTW, the IBM stuff doesn't sound like ECL because 10 ns prop times
+AD4APg- is way too high.
+AD4-
+AD4-Good point.
But there were other non-saturating logics around at the time. At least one
I'm familiar with looked like the front end of an ECL circuit, but didn't
have the emitter followers on the output (just some diodes for level
shifting). It had extremely limited fanout (as you could well imagine) and
in most cases went slower than ECL. It also took about the same amount of
power. So why was it used? Those were the days when bipolar transistors
(and the resistors for the emitter follower stage) were large. Lower
component count +AD0APg- smaller chip size. I think it was called DCL (for Direct
Coupled Logic). In addition, it may have been a hold-over from discrete
circuitry days. But. to be honest, this is from so far back, all of this is
fuzzy. Maybe one of the digital cognoscenti could fill us in on what the
other non-saturating logic families were.
faa
BBReynolds
>I'm searching for articles (or books) on the implementation of the IBM
>System/32 and System/34 minicomputers, or any other small IBM systems
>from the early to mid-70s. I'm specifically interested in details of
>the semiconductor technology, degree of integration, and processor or
>system architecture.
>
>
The physical architecture of IBM products of the 1970's has had little
literature made available (sorry for that horrible sentence). The logical
architecture did get some amount of "press". For the S/38, there was a full
issue of the Systems Journal, issued in landscape format: I brought one of
those back from a COMMON meeting, and gave it to a cow-orker in my Philadelphia
workplace. He went bonkers about the
architecture of the S/38, and started a campaign for IBM to hire him to become
a guru on the S/38; he eventually convinced them to hire him on, and
subsequently wound up in Rochester, as editor of the IEEE Computers and Society
journal.
The Series/1, which was built out of the same parts box as other IBM machines
of the time, had a formal book issued by IBM/SRA, written by James D.
Schoeffler
of Cleveland State University: again, the logical, rather than the physical
architecture is addressed---the book is Schoeffler, James D., <IBM Series/1:
The Small Computer Concept>, Atlanta, IBM General Systems Division, 1978
(SH20-0237-0) and Palo Alto, Science Research Associates, 1978 (13-4330). In
the case of the
Series/1, in the early days, a maxed-out 4955-E00 required nine slots to have
256KB with address relocation and floating point; and the end of the Series/1
product line, a 4956-J00/K00 would have all of the processor parts, and 2MB on
a single 6" x 4" card.
The parts bin was an important part of IBM product design at the time: although
logical architectures might be different, form factors and peripherals were
shared across product lines: the same line printer mechanism was used by the
3791, the
S/32, the 3776, and the Series/1 4973; the cabinet of the 3791 was shared with
the
S/32; the drive of the 3791 was shared with the S/32 and the Series/1, etc.,
etc.
There are some memory cards for the 3274 which are identical to those of the
early Series/1's, etc., etc., etc.
--
Bruce B. Reynolds, Systems Consultant:
Founder of Trailing Edge Technologies, Glenside PA and Niles IL:
Sweeping Up Behind Data Processing Dinosaurs
Paul DeMone
>
> Eric Smith wrote in message ...
> +AD4APg- BTW, the IBM stuff doesn't sound like ECL because 10 ns prop times
> +AD4APg- is way too high.
> +AD4-
> +AD4-Good point.
>
> But there were other non-saturating logics around at the time. At least one
> I'm familiar with looked like the front end of an ECL circuit, but didn't
> have the emitter followers on the output (just some diodes for level
> shifting). It had extremely limited fanout (as you could well imagine) and
> in most cases went slower than ECL. It also took about the same amount of
> power. So why was it used? Those were the days when bipolar transistors
> (and the resistors for the emitter follower stage) were large. Lower
> component count +AD0APg- smaller chip size. I think it was called DCL (for Direct
> Coupled Logic).
I think it was more commonly called Current Mode Logic or CML. IIRC Honeywell
used it extensively in their large systems. Whether or not it was slower
depended on fanout and loading. It was possible to combine both ECL and CML
to optimize speed vs area vs power.
> In addition, it may have been a hold-over from discrete
> circuitry days. But. to be honest, this is from so far back, all of this is
> fuzzy. Maybe one of the digital cognoscenti could fill us in on what the
> other non-saturating logic families were.
Schottky clamped TTL was a popular non-saturating bipolar logic family. These
circuits looked a lot like a conventional TTL with the multi emitter fround
end (actually DTL in low power Schottky) and totem pole output stage. The
difference was that a Schottky barrier diode was connected between the base
and collector of the pulldown BJT in the totem pole. The Schottky diode is
a junction between a metal contact and lightly n doped silicon. The diode
conducts current if the output falls much below the base voltage and thus
keeps the BJT out of deep saturation. The Schottky diode is quite easy to
implement on an IC, just make a contact to n diffused area without placing
an n+ contact plug.
Del Cecchi
snip
> keeps the BJT out of deep saturation. The Schottky diode is quite easy to
> implement on an IC, just make a contact to n diffused area without placing
> an n+ contact plug.
except the metalurgy has to be right to get the diode drop small enough
to clamp and not too small to destroy the noise tolerance by wrecking
the down level. Especially in Dutchess that didn't have the phase
splitter but just a resistor pull up.
Del Cecchi
>
> Eric Smith wrote in message ...
> +AD4APg- BTW, the IBM stuff doesn't sound like ECL because 10 ns prop times
> +AD4APg- is way too high.
> +AD4-
> +AD4-Good point.
>
> But there were other non-saturating logics around at the time. At least one
> I'm familiar with looked like the front end of an ECL circuit, but didn't
> have the emitter followers on the output (just some diodes for level
> shifting). It had extremely limited fanout (as you could well imagine) and
> in most cases went slower than ECL. It also took about the same amount of
> power. So why was it used? Those were the days when bipolar transistors
> (and the resistors for the emitter follower stage) were large. Lower
> component count +AD0APg- smaller chip size. I think it was called DCL (for Direct
> circuitry days. But. to be honest, this is from so far back, all of this is
> fuzzy. Maybe one of the digital cognoscenti could fill us in on what the
> other non-saturating logic families were.
>
The first IBM monolithic ICs were CSEF current switch emitter follower
called MST, monolithic systems technology. Used in 370 and system 3.
Followed by dutchess, ttl. Followed by Golf, DTL and Tango (bizarre
dtl) and Granite CML.
Keith R. Williams
<fad...@uswest.net> wrote:
>
> Eric Smith wrote in message ...
> +AD4APg- BTW, the IBM stuff doesn't sound like ECL because 10 ns prop times
> +AD4APg- is way too high.
> +AD4-
> +AD4-Good point.
>
> But there were other non-saturating logics around at the time. At least one
> I'm familiar with looked like the front end of an ECL circuit, but didn't
> have the emitter followers on the output (just some diodes for level
> shifting).
Diodes are better than emitter-followers at level-shifting?
I don't remember any non-saturating IBM logic other than CSEF
(Current-Switch-Emitter-Follower - ECL to "outsiders" ;-). Sure
they used cascaded logic for such things, but they were all ECL
based AFAIK.
There were some strange beasts, but Dutchess and it's follow-ons,
"Golf" and "Tango" were decidedly TTL, or more accurately SDTL. I
believe the S38 was all Tango. Tango was indeed an interesting
technology.
> It had extremely limited fanout (as you could well imagine) and
> in most cases went slower than ECL.
"Some" cases? The ECL of the time was MST4 (made by TI) or HPCL
(made by IBM). In that timeframe it may have even been HPCL-F
(MS-255 IIRC). There is no way that any TTL of the time could
catch MST4/HPCL. The 308x was a TTL machine and used a strange
variant of TTL (ground-down to minus 1.5V), but that was an
exception. BTW, it also had a variant with an emitter-follower
after the TTL driver. The open-collector driver was known as the
"long-line" driver for rather obvious reasons.
> It also took about the same amount of
> power. So why was it used?
I'm not sure what technology you're talking about here. I think
you may be confusing many.
> Those were the days when bipolar transistors
> (and the resistors for the emitter follower stage) were large. Lower
> component count +AD0APg- smaller chip size. I think it was called DCL (for Direct
> Coupled Logic). In addition, it may have been a hold-over from discrete
> circuitry days. But. to be honest, this is from so far back, all of this is
> fuzzy. Maybe one of the digital cognoscenti could fill us in on what the
> other non-saturating logic families were.
I haven't seen any other non-saturating logic other than ECL
(CSEF) that made it into production. The real problem is that the
current-switch dissipaes power in either state. TTL only in one
(though it has a huge dI/dT), but it doesn't matter much.
They're both too power hungry for the densities of today. CMOS
only dissipates power when switching and has knocked all that
archaic bipolar stuff into history.
----
Keith
Del Cecchi
>
> On Fri, 2 Apr 1999 05:59:20, "Frank A. Adrian"
> <fad...@uswest.net> wrote:
>
> I don't remember any non-saturating IBM logic other than CSEF
> (Current-Switch-Emitter-Follower - ECL to "outsiders" ;-). Sure
> they used cascaded logic for such things, but they were all ECL
> based AFAIK.
>
> There were some strange beasts, but Dutchess and it's follow-ons,
> "Golf" and "Tango" were decidedly TTL, or more accurately SDTL. I
> believe the S38 was all Tango. Tango was indeed an interesting
> technology.
>
later called, MS-381 with a follow on in MS-431. It was similar to the
MS-359 used in the 3081 only in single chip air cooled modules.
Tango was interesting all right. Goofy might be more like it.
> > It had extremely limited fanout (as you could well imagine) and
> > in most cases went slower than ECL.
>
> "Some" cases? The ECL of the time was MST4 (made by TI) or HPCL
> (made by IBM). In that timeframe it may have even been HPCL-F
> (MS-255 IIRC). There is no way that any TTL of the time could
> catch MST4/HPCL. The 308x was a TTL machine and used a strange
> variant of TTL (ground-down to minus 1.5V), but that was an
> exception. BTW, it also had a variant with an emitter-follower
> after the TTL driver. The open-collector driver was known as the
> "long-line" driver for rather obvious reasons.
The off chip drivers for Purdue were Emitter followers mostly except for
the long line drivers. MS255 was used on the 3033 which was done on a
very short schedule due to Gene Amdahl. IBM sold many more than they
thought they would.
>
> > It also took about the same amount of
> > power. So why was it used?
>
> I'm not sure what technology you're talking about here. I think
> you may be confusing many.
>
> > Those were the days when bipolar transistors
> > (and the resistors for the emitter follower stage) were large. Lower
snip
> ----
> Keith
Ron Nicholson
>It had 134 gates, 60 three way NAND, 40 4Way NAND and 34 2way nor off
>chip drivers. Gate delay was about 10 ns.
What was the CPU cycle time (if synchronous) and/or the approximate
micro and macro instruction dispatch rate? How about the memory cycle
time? Do any System/32 benchmarks still exist that could be translated
to run in some form on current hardware (a WorkPad for instance)?
--
--
Ron Nicholson mailto:r...@sgi.com http://reality.sgi.com/rhn/
#include <canonical.disclaimer> // only my own opinions, etc.