026 Keypunch Machine Arrival

October 23, 2013

We received an 026 keypunch machine from Akron, OH. It was full of card punches and dust but it has been cleaned up. Various adjusments have been made and it is now punching and printing 80 column cards. Hole alignment is off a little and the printing has a little problem with a couple of the dots in the 5x7 dot matrix. Hope to resolve these problems. We can use more cards! At the current time a shipment of old IBM card equipment is on its way to us from Brussels. Included are a 108 proof machine, a 129 keypunch and a 1442 card read-punch along with a couple of 1403 print chain cartridges.

082 Cart Sorter Arrival

June 29, 2013

We received an 082 card sorter from Chicago, IL. The 082 has been cleaned and adjusted and is sorting cards. While Fred, Gary and Jack worked on building a 1442 interface, Bill worked up changes to the math expression solving software to provide hierarchy of math operators, ie. multiply and divide execute in preference to add and subtract. This change was developed as changes to the existing program which was still in the 1441 memory. Nice feature of magnetic core memory, it remembers what is in memory when the power is turned off. These changes were not tried as the 1442 interface became ready to debug. With the 1442 interface board design complete, Gary built it using a PC AT expansion board, 2 Arduino microprocessors, opto isolators and other components. Gary programmed one Arduino and Jack programmed the other one. Debugging this interface led to having to adjust the card read hardware and due to an early microprogram bug, one read attempt caused all 8000 1441 memory locations to be cleared to blanks! It was as if the reader had read a card that was 8000 long! With problems corrected, we have actually read a card correctly into the 1441 memory, although the 1441 indicated that there was an error. A problem to investigate when we aren't working on our other devices.

129 Keypunch Arrival

February 21, 2013

We received a 129 keypunch machine from Darien, CT. We discovered that there are several problems with this unit. Unfortunately it uses SLT cards with SLD modules. Our scope shows us that at least one module on one card is not working correctly. However, we don't have logic diagrams for these cards and the modules consist of several transistors and diodes so the problem will be hard to locate and fix. (SMS cards are single layer PC boards with discrete components, SLT cards are multi-layer with complex components and very hard to obtain).

Another milestone for CT&I's IBM 1440 project

March 23, 2012

Donated to CT&I from the mountains of California by Bob Rosenbloom under the guidance of Robert Garner, IBM Almaden Research Center, one of three known IBM 1442 Card Readers in existence was delivered to the TechWorks! Workshop last week.  The device was designed by IBM San Jose in the late 1950s, per Don Rex, IBM Fellow.  Don's 1442 team was building IBM's last card handling equipment while down the hall the magnetic disk drive was being developed.  Per Robert Garner, who works closely with the 1401 restoration project team at the Computer History Museum, Mountain View, CA, the other two 1442 card reader punches are found at The National Museum of Computing, Bletchley Park, UK, and the 1130.org, a group of private collectors and enthusiasts in Berkeley, CA.

The 1442 is quite clean; the mechanics look good, with no evidence of a non-OEM mouse in residence - what a delight!  CT&I's 1440 team will need to build interfaces and cables to account for voltage and other vintage differences between the 1442 SLT cards (mid 1960s) and the SMS cards (late 1950s) used in the 1441/1447/1311 devices. Documentation for the 1440 system is coming from around the world - Mountain View and Berkeley, CA; Boca Raton, FL; Sindelfingen, Germany; Fishkill and Sommers, NY; and Bletchley Park, UK.  None the less, there are holes in the knowledge base - if you know of 1440 system documentation, please let us know.

Because the 1442 is both an input and an output device, CT&I now has a good chance of operating a complete IBM 1440 system for the 50th anniversary of the system release in Fall 2012.  Although the 1440 system performs complex math worthy of an 11th grader, visitors can't see the machine thinking - no flashing lights, no moving parts, no visible output, only a steady hum.  With an operational 1442, visitors will be able to watch a deck of cards be read in, data crunched, and cards output with results. We are still in the hunt for an IBM 1403 chain printer, vintage 1959 - mid 1960s, fast enough to play music from a card deck.

Thanks to all who are working so hard to see the 1440 restoration come to fruition. We hope you will be able to join in celebrating the 50th anniversary of the IBM 1440 System, details forthcoming as the plans emerge.  Excelsior!

See pictures from the arrival here

1442 Card Read-Punch Model 6/7 Arrival

March 13, 2012

We received a 1442 card read-punch model 6/7 from California. Our 1441 was designed to connect to either a 1442 Model 1 or 2 card read-punch which used SMS hardware as the 1441 does. The 1442 Model 6/7 has SLT logic and no translation hardware to handle character translation between the Hollerith card code form and 1441 BCD form. Bob Horan, retired Link machinist, corrected a twisted side plate in the card hopper and made some new machine legs for the bent and missing ones. Fred, Jack and Gary started thinking about designing an interface to allow the 1442 to be connected to the 1441. This interface has
 to convert signal voltage levels between the SLT and SMS hardware and translate character codes between the Hollerith card code form and 1441 BCD form, in both directions, one way for reading, the other way for punching.

Several retired IBM CEs worked on improving the performance of the selectric printer on the 1447 console. Fred and Bill repaired three broken "teeth" on our 1440 system special type ball by using JB Weld. Anybody have one of these special type balls? We could use it.

With the selectric printer in better shape, we tried printing all the characters. They all printed except for letters F and O, those hung up the machine. This problem took a while to track down and was finally traced to a ribbon cable connecting signals between two circuit panels. One end of the cable was in its slot but not plugged in and hard to spot because it uses short paddle cards. The 1441 hung because the cable problem caused the 1311 disk file interface to think that it was being addressed. The unplugged cable was probably a result of an oversite of reassembly when the machine was cleaned. After getting more comfortable with programming the 1441 and writing various subroutines for multiply, divide and typewriter I/O, Bill got
 more adventurous and developed programming to solve math problems using floating point software supporting add, subtract, multiply, divide, square root and parentheses to change exection order. Bill based this code package on a math expression solving program he had written for a PC. He converted the code from PC BASIC to the 1441's symbolic machine instructions and, at the beginning, had no idea if the program would fit into the 8000 character memory of the 1441. As the 1442 card reader and 1311 disk file were not yet operational, any programming was restricted to the 8000 character memory that was available. As it turned out, about 5000 of the machine's 8000 character memory locations are involved in doing this task. As we didn't have an operational card reader nor any way to punch the cards required at this point, programs were entered into the 1441 memory via the console
 keyboard. The debug process lead to finding a couple of additional errors in instruction execution. Again, bad transistors. There is also a yet unresolved problem of an occasional extra bit being turned on in a character in memory when data is typed into the machine and the parity bit gets flipped so that the problem entry can not be found by a memory scan for bad parity entries! Often the extra bit appears in an instruction's memory address field and leads to an unwanted use of an index register.

To process a math expression, it is entered via the 1447 keyboard,
 solved by the 1441 and the answer is typed in scientific notation on
 the selectric printer.

Trying to write machine code by hand and figuring out the instruction operand encoded addresses was a pain

July 13, 2011

Trying to write machine code by hand and figuring out the instruction operand encoded addresses was a pain. So, Bill wrote a PC based Assembler to support program generation for our 1441 processor. He used a symbolic coded multiply program listing and a machine coded divide program listing found in a 1401 system reference manual as a starting point to develop routines to do multiply and divide operations as our 1441 was not built with these instructions. The closest our 1440 system comes to multiply and divide are two blank buttons on the 1447 console that would have been operational had the machine been equipped with the multiply/divide option.

Several errors were found in the manual's multiply source listing, we corrected those. The machine coded divide program instructions wereconverted to symbolic form, assembled and entered into the 1441. It worked great, once! As it was designed, the program modified itself as it executed and had to have the instructions reinitialized before it could be used again. Bill discovered that our 1441 was built with the index register option so he revised the program to make use of it thus eliminating the need to reinitialize instructions and it was also then freed of the restriction that it reside in low core (addresses 000-999).

We tested various 1441 machine instructions. When execution errors occurred, Fred, Gary and Jack sprang into action. They got out the oscilloscope and ALDs (the machine's logic diagrams that we were lucky to receive with the hardware) and tracked down each problem. Bad card components were replaced (usually a bad germanium transistor was the culprit) and we were back in business.

Initial cleanout finished

June 17, 2010

Now that the initial cleanout of mouse debris, testing of power supplies, correcting the plugging of mislabeled cables, checking of clock signals and mechanical fixes to the selectric printer had been done in 2010, we were ready to exercise the 1440 system.