Wow. Thank you very much for your interesting information. Thats great to hear the process of people who were involved at that time. Do you know if the length of the wires werde defined?

@@CPUGalaxy Some were defined length. Those of defined length were usually "clock" wires. We would measure them manually. Longer lengths were coiled to try and keep them as neat as possible. I think I saw one of those on the back of the board you had. All other wires were just as long as they needed to be. We would route these over and under other wires to prevent the from hanging down.

Very interesting! Thank you very much. I appreciate a lot that you take the effort to provide us here some information. I wish i could make an interview with you for the channel to get all those information. 😀

did it. 😉

Not surprised it took 6 weeks to finish. Someone could probably write a formula to determine the length of time needed to assemble something based on how much of a "rat nest" it is.

😂

kzfaq.info/get/bejne/hLh5pZx-s9Onp4k.html

That actually made me laugh for some reason

More to the point: despite all that manual wiring, an Amdahl system was still less expensive than the equivalent IBM system. I think that says a lot about the profit margins on IBM's machines.

@@BertGrink that's mind blowing.

Yeah, this channel deserves the attention

Me too sent by Dave ...

@@BertGrink Believe it or not, to convert a 470 V7 into a 470 V7 A or a V7 B (less powerful variants, or even to the most powerful V8 only required moving a couple of plugs on some of the connectors that the MCC boards as seen in the video, plugged into. For a V8 conversion, if I remember correctly, you had to change out one of the MCC boards as well. It was such a simple operation and I was always amazed that a 30 minute task would cost the eventual customer hundreds of thousands of dollars. When I had to do a conversion I had to go to the Document Control Center and sign out the rework instructions which were kept in a safe. For some of the simpler updates (V7 A to V7 B for example) I could have carried out the rework without the instructions as I know the steps by heart, but you still had to go and sign out the rework instructions as that was the procedure. 😉.

yes. indeed 😅

nice idea!

@@CPUGalaxy Also maybe add rotating mirror on center axis. To show the many angles of point to point wiring versus surface mount chips on top.

could put on a slow spinning stand or even remote controlled to allow one to stop and appreciate the backplaneliness goodness...

I'd just put it in a plexiglass box and leave it sideways so you could see both sides as you walk around.

Thats cool to see somebody here who worked at Amdahl 😃

@@CPUGalaxy Those wires are considered "twin-ax" or basically co-axial wire. The chips are ECL technology and it was all air cooled. The things running along the chips are resistor packs. The chips used 3 voltages - 5.2v, 3.6v and 2v. The next generation boards are even more impressive at about 14" square. Some boards had about 2,000 wires on back and were serviced by people in the rework department in Sunnyvale, CA.

@@budude2 The MCC backside wires (i.e. twin lead) weren't coax though they provided appropriate shielding. The idea was to not have conductors behave like antennas at the related switching frequencies and to effectively eliminate transmission line effects from signal rise and fall times; hence the "terminator packs" for impedance matching. Coax wires were used for MCC-to-MCC and off-plane connections. The LSI ECL chips were powered only by -5v, provided by five slave (i.e. constant current) power supplies and one master (i.e. constant voltage). The power subsystem for the two CPU plane of MCCs could source about 1500A.

yeah, indeed!

I kept some projects like that even after prototyping :-D - sometimes I just grab a bunch of through-hole chips and solder them together using long leads from resistors/caps. If the project is simple enough, few blobs of hot glue afterwards keeps everything isolated.

@@PaaAL haha me too. Currently have a Z80 triple timezone clock that is made up of 4 boards using rats nest construction. It works just fine, so no need to change it. :)

Amdahl mainframes had built-in debugging capability called Scan. A not insignificant portion of each chip's logic was dedicated to enormous muxes which allowed the SVP to read out nearly any register in the machine. Similar to JTAG boundary scan, except it went into the chip. The logic designers could wire up any important registers or flip flops to scan, and later the system integration engineers could halt the system, issue individual clocks, and watch data flow from register to register and chip to chip. This feature combined with generous ECC and proactive patrol function on the service processor made for a very reliable, Available, and Serviceable machine. RAS. Still burned into my brain all these years later.

Thank you 🙏🏻

thats the big question 🤔

I think they simple answer is that they carefully listed all the junction points for the wires and the lengths they needed to have in a table and then they wired it up one wire at a time by hand according to the check list and then they checked each smegging wire one by one according to the list, then they checked for bad solder joints and other possibly visible defects, then they checked the function of the whole module in a test harness (e.g. in comparison to some other known good module receiving the same input). Boring, expensive and tedious. I don't think they had any magical technology to not make this tedious. For utter madness, see the rope memory used in the apollo guidance computer for ROM. It's like core memory, but since it is not writeable, you can reuse each iron core however many times you want; if the bit is a one or a zero depends on if the sense wire passes through the core or not; you could have hundreds of wires through a core. They essentially sent the programme off to be woven by some old ladies a single bit at a time. And then you could read the memory back, and check that each bit was correct, and then you could shove it into a package and pot it with epoxy. What you ended up with was a small brick contain 4 layers of 4 rows by 32 columns of iron cores smaller than a pinky nail (512 in total), with up to 192 wires through each core in a madening yarn of thin enamaled copper wire that doesn't seem to be threaded in any particular pattern; half a mile of wire packed into a casing maybe twice the volume of a 2-piece snickers bar. To make this process possible, they made a special machine wired up to a computer that would indicate which hole the needle with the sense-wire should be passed through next so the ladies (and yes, almost 100% ladies worked in making core memory) could focus on just passing the needle and not error-checking on the fly.

Thank you 😊

it's as messy as your house after one of your parties

Some were the same but mostly not. For example, some chips were used repeatedly if they were providing a slice of say, an address generator but certainly not if it was part of a state machine controlling operations on and movement of extended precision floating point bits in the E-Unit (i.e. arithmetic unit).

I worked on these boards in Sunnyvale, CA circa 1981-1983. It was my first job with a computer company, after having graduated from business school in Ohio. The board wiring assembly area had at least a hundred machines doing rework, and always full of assembly workers, mostly female immigrants from China, India, and other Asian countries. Some of these ladies would work 7 days a week non-stop and would drive to work in Sting Rays, BMWs, and Mercedes. The laser machines were in operation across three shifts and even on weekends. The demand for workers was so high that the only requirement for hiring was your dexterity ability with handling tweezers. Such simple requirements for such a high paying job with overtimes was very foreign to me. I made much more per hour as an assembler at Amdahl, than I did as an inventory auditor with a BS in Business Administration, at a local retail chain. After about year of rework experience at Amdahl, I saved up enough funds to quit and go back to school at Heald College, Technical Division in San Jose. I graduated with an associates degree in Electronics Technology and immediately went to work as a bench technician at Avantek Telecommunications. Later went on to get my BSEE and worked for a number of technology companies through out the San Francisco Bay Area. Going back working at Amdahl, those laser rework machines were simply amazing technology for the time. The machine did most of the work, you just had to make sure the laser gun was lined up over the solder pad, and with your tweezers, move the existing wire out of the way so the laser could do its job of soldering the wire to the pad. Then just let it route the wire to the next pad.

wow, very super cool document you liked. Great, Thank you!! 😍

Amdahl 470/V{5,6,7,8} cooling was strictly air cooled, drawn from under the raised floor and accelerated to high velocity through clear plastic vertical conduits over the front of the MCCs. Since the MCCs were mounted in an vertical plane, the ones at the top were exposed to airflow already warmed by the two MCCs located beneath; all taken into account. At the end of the day there was no cooling infrastructure purchase nor maintained strictly for the system; a selling point. Power on the other hand was demanding, requiring a 400HZ, 3-phase motor-generator. I have no idea the cost of one but as they say, if you have to ask, you probably can't afford the mainframe. ;)

great idea 😃👍🏻. Thanks

Long overdue!

Not sure about the pioneering, SMD was developed in the 1960s, but it did take a long time to become widely used. My guess is that secondary PCB would have routing issues (a lot of criss-crossing signals so you would need a lot of layers) and you don't want to spin a new board for every engineering change, wiring is easier.

The MCC was a production board. Wire wrapping was not within physical design constraints and probably posed serious reactance problems. Capacitors would have slowed signal propagation and affected signal edges. All Amdahl CPU signal paths were EM transmission lines.

Amdahl claim to fame included but wasn't predominantly about its cooling system. The name of the game was "price/performance", and Amdahl left IBM in the dust on that criteria.

Wire wrapping a backplane was 1960's technology the mainframe world.

😄

The MIPS performance was completely dependent on an instruction mix. The instruction pipeline was designed with something around fifty interlocks that could stall most stages, affecting completion times. Every machine instruction had defining equations that included potential pipeline interlocks and worst case timing impacts in their definition.

@@scottgever1290 That sounds pretty unique and fascinating to me. Thank you for the detailed information. This is advanced stuff. Mainframes always had quite unique designs and equally special processing capabilities. Tons of research effort, engineering and science must have gone into all that.

Yes, along with multilayer boards with up to 16 layers of internal copper. Very fragile boards to work on, as the layers and the vias tend to break when you attempt to remove the solder, and the tracks peel off easily. There are whole video series from Pace dedicated to fixing those boards, using their equipment and repair kits.

Here's a photo of the wafer. The thing that stands out the most, is the fact that it's not the traditional regular grid of many chips on one waver. This thing was a massive undertaking, and I can only imagine where computing would have gone, had they ever succeeded in making it work. i.imgur.com/AaZBhIu.jpg

would make a pretty nice calculator

sounds like a virtualization nightmare

Crosstalk prevention - twin lead on the backside and good EE design on board (i.e. limited parallel trace runs, ground planes, etc.) External EMI sources weren't a problem, but there were plenty internal to the frame so again, good EE design utilizing board ground planes, shielded twin lead, inter-MCC coax wires, etc.

Yes, it is quite amazing, but is it more so than the fact that a lot of financial institutions still use software written in COBOL? A language that has very few practitioners left, and according to rumours, those that are still alive can pretty much dictate their own salaries.

Governments too. I'm working on a Z14 mainframe every work day - we have a team of COBOL programmers as well as a bunch of Java folks, and we still have some Bus and Tag equipment in operation. It's mostly Linux workloads on the mainframe, but we have CICS and batch processes from the 80s that still are used.

Not at all. There are tens of thousands of COBOL progs in India, and they don't dictate price