System 360 Model 44 (2044)




IBM System 360 Model 44 Operator Control Panel

The 2044 was a bit of an odd duck. It did not implement the full 360 instruction set, but had enhanced support for numeric / scientific processing. There is a rotary switch on the front to select floating point precision. How did the operator know what setting to use?

The Gallery has been extremely fortunate to have found two complete System 360 Model 44 operator consoles. Below is a picture of the twins. Also visible at left is a System 370/135 console and in the background is a System 370/155 console.


Here is a picture of different varieties of lamps used in System 360 and System 370 (click for larger version). Notice the 360/44 lamps. In keeping with its odd-duckness, /44 lamps have extra long pins, while all of the other bi-pin lamps from the System 360 and System 370 lines are interchangeable.


Below is a picture of a System/360 Model 44 computer complex from an original Model 44 sales brochure. We wondered why the picture in the brochure is so different from the panel in the collection -- until we received the email below! Do you have a story to tell about this machine (or any machine in The GallerY)? Please let us know (curator@thegalleryofoldiron.com).


Here is some great information sent in by someone who knows the 2044 first hand:

The control panel you have is the ~real~ 360/44 panel. The photographs seen in early IBM brochures (pre-product-ship) were of dummy CPUs based on the model 40 frame. To accommodate the considerably greater volume of SLT circuitry and the two embedded single disk drives, the 44 frame was much taller and longer than the model 40 frame and had much taller internal circuitry gates. The 44 had much more circuitry because it did not have "microprogramming" such as the model 40 TROS or model 30 CCROS -- everything in the 44 was hardware, which accounted for its ~much~ faster performance, particularly of floating point and integer math -- not to mention that it was a 32 bit machine. Also, the 44 had a much more complex floating point unit, also free of "microprogramming." Even the channels were free of microprogramming.

As an unusual aside, each Model 44 was shipped with a rolling stool so the service person (an IBM Customer Engineer -- I was one) could reach the two single-disk drives for service.

The 44 was initially available with up to 128K (that's K) memory, but later could be upgraded to 256K with an added "L" frame at the rear.

The IBM archives talk about the minimal system programming initially available, but eventually it had a full-blown operating system, similar to 360 DOS. I was also an IBM programming support representative for the OS.

I'm certain the 44 was developed in the Hursley, England lab. The software was done at the IBM Time-Life facility in New York City, with the later Fortran compiler farmed out to a vendor.

The /44 definitely wasn't an RPQ. An unfortunate thing was that IBM salesmen tended to market it to commercial customers -- more performance for less money -- but it was a poor match; the operating system (44 OS), being so FORTRAN-focused, just wasn't up to the job.

Re: size, the gates on the 44 were one SLT back-plane board higher than the model 40 frame, which was three boards high. The gates on the 44 literally groaned when you swung them open. The 44 publicity photo is clearly a CPU dummied up on a model 40 frame. Somehow they got lights in there to show 32 bit registers (vs. 16 for the 40), but that's about it. It was vaporware.

The computer was manufactured in the Poughkeepsie plant -- at least those for North America consumption.

The initial machines came with a programming "monitor," basically a job manager, on the little single disk (called SDSD, single disk storage drive), and a FORTRAN compiler and Scientific Subroutine Package. It was produced because the 44 OS was way behind schedule, and they "had to have a FORTRAN" for the 44 to ship.

That initial release was really funny -- ~full~ of bugs. As I began digging into problems, it seemed just too buggy to me. So I started looking at the initial S/360 BPS FORTRAN package bug history. The same bugs!!! The IBM programmers at the Time-Life building in New York City had simply copied the BPS FORTRAN package, building in emulation for the non-44 instructions. But they copied from the original, very first, Version 1-0, with ~all~ its bugs, instead of using a later release. So we re-invented the square wheel on the 44.

Months later, the 44 OS was made available, which was basically 360 DOS, rewritten for the 44 instruction set (which lacked the commercially oriented "SS" instructions). The 44 OS FORTRAN compiler was farmed out to a third party programming shop. Working on it is where I first learned of such strange programming constructs as "Polish Rolls." (That's Polish, as in the country)

Later, a commercial instruction set hardware emulator was made available, to bring the machine up to the full 360 instruction set so it could run the other 360 OSs. However, it was ~really~ slow. It basically captured the invalid operation interrupt, emulated the offending instruction, and then sent the computer on its way.

The machine was easy for Pugh to overlook because the volumes were really low. DEC had started to arise about that time, and took a lot of the business the 44 would have normally had.

I was a Customer Engineer servicing those computers. I was in the first hardware CE class in Poughkeepsie, and in the first Operating System class in Poughkeepsie. I installed the second machine to be installed in a customer account. In those days, CEs knew the function of every logic block (transistor/diode) in the machine. I spent the better part of a year in training, first on the S/360 IO in Los Angeles, then to Poughkeepsie for the CPU, and back to Poughkeepsie for operating system training. I was a unique cat -- CE and Programming System Representative.

The 44 CPU "broke" a lot. Not because it was intrinsically unreliable, but because it had SOOO much hardware circuitry. I'd guess three times the circuitry of the 40, twice that of the model 50. Remember, up through the 65, IBM CPUs used "microprogramming," which obviated a lot of the need for circuitry. The 44 didn't, it was all hardware -- that's why it was so fast, especially with high speed hardware registers (instead of core, which nobody ever bought). Floating point operations, up to 64 bit, screamed -- for those days. I spent many, many hours with an oscilloscope figuring out why the machine did a floating point multiply, or some other operation, wrong.

I attended an announcement meeting for the original System/360 announcement in 1964. In that meeting, they had a film of the SLT logic system -- modules, cards, boards. They showed how the internal interconnections in the back plane board would take care of virtually all the requirements for interconnecting the cards and connecting to the edge connectors. They showed 5 or six yellow wires on the back of a board to demonstrate how exceptions and engineering changes would be handled. Well, I'll tell you, on the Model 44, you couldn't ~see~ the board for all the yellow wires. They were packed in there three deep on the pins.

The first Engineering Change on the 44 took 40 man-hours to install -- replacing some cables, some boards, and wiring, wiring, wiring . . . until your eyes went bad.

Another nugget: A lot of 44s were sold to oil companies in Texas. There was a machine that could attach directly to the 44, via the Direct Data Channel, I believe, to perform ~very~ complex math equations that the oil companies required in their exploration business -- maybe processing the results of shock wave analyses of the ground. It was a "gravity" math engine -- that is, "~visually~ speaking," the equation "knowns" were dropped in the "top," and they "fell," unclocked, as fast as the electronics could process them, with the answer dropping out the "bottom." I don't think it was an RPQ, I'm certain it had a machine type. It would be great if your organization could find out something about it. (I did not work in those accounts)

As my (now feeble ) mind chews on this, I may come up with the kinds of equations those were -- it was a very specific thing.

Nugget #2: The 44 was also physically longer than standard 360 CPU's. So the front section was separated, on casters, from the standard length rear section for shipment. It was real tall, narrow-square in the other dimensions. It had this huge overhanging front panel, so it was very unstable - in fact, by itself, it would simply fall over. So they stacked huge lead plates on the bottom rear of it to keep it from falling over on its face.

Jerry Blinn, October 2003

Subsequent to Jerry's great stories of the Model 44, we found out about two devices that were attachable to do even more "extreme math." One was the IBM 2938 Array Processor (described in IBM Systems Journal Vol.8, No.2, 1969) and the other was the IBM 2937 Multiplication Summation Processor. Anyone have more information about these devices? Pictures? Functional Characteristics manuals? Stories?

And we received this from another Gallery reader:

I worked on one at the IBM site in Burlington (Essex Junction) VT from 1969-1973. Very interesting machine. At the time, the site had a 360/20 (?) as an RJE machine to a huge mainframe out in CA while my lab had the '44. It was connected to a TI-553 circuit tester since we were attempting to develop 128-bit (that's single bits ...) bipolar memory chips. Although all the software was written in Fortran, the O/S was something called DAMPS - Data Acquisition and Multi Programming System - a true multi-taking system that handled interrupts from the attached test system and could respond with the necessary signal timing. The programming model was very basic since essentially everything was sequenced to run the tester. I do remember that it had an extremely fast real-time clock, I think with 1 microsecond accuracy along with the floating point hardware.

We hired an S/E from TX, Robert Hansard, to write a driver for the IBM 2260 display ( a real boxy 14" display - not like the CAD 2250 21" display).

The machine was finicky as the article stated. We were on an A/C, raised floor room. The CE's could never get the temperature controls working, and found that removing the doors and some of the sheet-metal ducting (removed and piled in the corner) kept the machine operational.

The other story was that the machine could operate at a speed faster than the 360/65, the fastest normal business machine at the time (excluding the 360/67 tandem, 360/75 and 360/90?, all specials for either NASA or big insurance companies like Prudential). With the pre-consent decree educational discounts offered by IBM, the universities were buying the '44s and running business applications on them ... the students would write the emulators to get the legacy (what legacy in 1968?) code or new code in Fortran to work. Additionally, IBM killed off the machine when they saw its potential and refused to upgrade the O/S, so a SHARE group arose of all the universities that were now stuck with these dead-end machines, making mods and keeping them alive.

A couple of other notes. The machine address length would allow the addition of memory up to 384K ... a real odd size, completely determined by that address field. One school wanted to bump it up to the '65 1MB size (not sure if they did it or not) by completely modifying the address bus to handle the change. I think the cost of the memory was something like $25K for 50KB .... ouch.

I think it was the Arizona School of Mining and Technology (it's been a long time) had a machine with the normal IBM 1443? slide-bar printer attached. It could print 144 cols wide and was noted for the large type bar that slid back and forth with hammers behind each print position timed to hit the correct character as it passed. Well, the students at the school were printing thesis docs and the printer's MTBF was something like 10 hours. IBM was called every time it broke, almost to the point where they kept someone on site. The school asked for a faster printer (the 1443 was something like 150-300 LPM) like the IBM 1401 that screamed along at 1100 LPM (core dumps would do 1600 LPM due to limited chracters needed) but IBM said the development cost was too prohibitive ... at the same time that we had the 1401 attached and running at our site and the East Fishkill site that had the only other '44 in the company!

The machine also supported 600 bpi tape drives. When IBM came out with the 3000 series ? which I think could do 3200 bpi, one of the universities asked IBM for an RFQ. IBM refused, saying that aside from attaching it to the '44, there were no diags to support the attachment (was that OLTEP?). The school hackers re-wrote the diags from the '65 that had them attached and -gave- the diags to IBM and insisted that they then rent the new drives to them.

The other machines used to control test systems were the 1130 with a SAC channel and the 1800 which was a true process control machine (sold a lot to the oil companies). The next generation was the S/7 (System/7 .... not to be confused with the AT&T; SS7 - Switched System 7). They were all 16 bit machines. The next generation was the Series/1 running EDX - Event Driven Executive ... again running with 16 bits. When internally they were going to develop an upgrade to the S/1, we asked for a true 32 bit machine ... which we were told was an unnecessary requirement for current systems. When the newer systems were delivered, we noted that the new 5MB disk drive ... was unable to directly address the sectors ... because the number could not be handled in 16 bits ... so there was a lot of cylinder/sector/track manipulation to "create" a number that would address the information.

Hope you find this helpful.

Bob Sparling, January 2005

And here is a note from someone who knew both the 2044 CPU and the attached 2938 Array Processor.

Dear Curator,

The IBM 2938 was indeed used by many of the Geophysical analysis companies back in the 60's.

I was an operator in the late 60's (1968/9), working at "Western Geophysical Co. of America" in West London and we had 3 x 360/44's installed. Each had an array processor attached. I don't think I have any pictures of the box, it was a pretty dull box, just a rectangular box similar in size to the 44's CPU frame.

I became a FORTRAN programmer and although I didn't write progs for the 2938, I recall the APM calls mentioned in the text.

Western GeCo became "Western Atlas" -- HQ in Houston.

I found this in the IBM research web site that may be of some use to you. You may need to acknowledge copyright etc. It's an extract from the IBM Systems Journal relating to the 2938. [ibmsj0802C 2938 array processor.pdf]

Best Regards,
John


Return to The Gallery of Old Iron home page