My Site

360/40 CPU

The thing with all the blinky lights. The number of blinky lights went up and down with the model of the CPU. More powerful CPUs had more lights.

The CPU stood about head high and over 6 feet long. Ours had a cabinet across the back that held 128K of memory. The first 128K was in the first box. The blinky lights were on a panel on the front of the CPU. Below that was a desk area. On the right, at 90 degrees, was the console. I think you could get the console on either side, but most, possibly all, I saw had the console on the right.

The console was a selectric typewriter.

Front Panel

In the lower right of the panel there were three dials. These had the address you wanted to IPL, Initial Program Load, from. Normally they were a disk drive, but sometimes you loaded from the card reader or a tape. There were a couple buttons, load, interrupt, and others I don’t remember.

There also were a couple of lights in a group, wait, lamp test, and a few others I don’t remember. When the CPU was running there were times it would have to wait for things like tape drives or unit record stuff and while it was not processing the wait light would come on. Usually it was only on dimly cause the CPU didn’t wait long. The lights were encased in a plastic case with two pins sticking out the back. You could grab them and pull them out. You had to know if a light was off or it was dark because it was burned out so there was a switch, lamp test. Push it down and all the lights came on. For some unknown reason the lamp test light had a higher voltage than all the others. For grins you would take a bulb out, stick it in the lamp test socket, hit the lamp test switch, and the bulb would get bright and burn out. Don’t do this too often of the CE, IBM guy who looks after the hardware, will yell at you.

There is a row of, I think, 32 switches for the address. The reason for 32 is that the 360 used 31 bit addressing, far beyond this to explain why 31 and not 32. Above that were 32 switches for the data. You could put an address in the low switches and data in the row above, press the store button, next to the load and interrupt, and it would store the data. The 360 worded with bytes, half words, and words. If you have a word it had to be on a word boundary. Later families of CPU did not have to align words. (That is kind of correct, but there were only 18 switches. Made a bunch of sense since nore would produce addresses the CPU did not have. They were in groups of 8 with 4 black and 4 white handles each.)

Another thought. Why are there 32 address switches. A 40 can have a max of 256k of memory. Do I remember wrong or did IBM make it look nicer?

They are REALLY nice switches. A slit in the face with a flat piece of metal sticking out. On the end is a round plastic piece with a metal cap on the end. They stick out 1-2 inches. They are straight out when off. Flip them down and they are on. They are stable in both positions. Move the switch part way and it will snap the rest. The machine cost a couple million dollars and you can see the quality everywhere.

Above the address and data row there are a whole lot move lights. They have a description of them underneath and they are grouped. I remember a time or two in the years I worked with them when the IBM CE would look at a few of them. Most of the time they did nothing more than blink. They represented the internal state of the machine. I think there was a light that came on when the last arithmetic instruction had an overflow. Things like that. If a machine had a problem and halted you MIGHT look at the lights and be able to figure out what happened.

There were 2 hour counters. One counted the hours you used the machine, You could rent based on the number of hours you used or just rent the entire machine or buy it. The other counter had the number of hours the CE had the machine, important not to pay for the CE if you were paying by the hour. The CE had a key he would put in and turn to swithc between counters.

On the top right was the emergency button. Not a button. A thin stalk came out with a round thing on the end. You grabbed the thing, fingers behind the flat part, and pulled. I never got any real training or information on this. Never saw one pulled. The rumor was that after it was pulled a CE had to do stuff to allow it to be pressed back in. Burning out bulbs was one thing but we never messed with this switch.

The front was divided into separate panels. The bottom was angled differently so you had a large area on top that sloped away from you and a shorter on on the bottom that sloped toward you. The top was divided into smaller panels that had related stuff. The address and data switches were on one panel and another had lights for channels, etc.

As an operator most of the front was unused. You occasionally would change the device you IPLd from, do an IPL. On rare occasions you might start/stop the machine. Some optional software used the interrupt. That was pretty much all.

I used the address and data switches to play with the computer. The rest was a mystery.

As machines increased in power they got more lights. The 50 had a couple rows with a thing below that you could rotate and change the labels and see different things.

Insides

The panels on the sides were hinged and you could open them. Then you saw the logic gates. A bunch of pins sticking out with yellow wires going everywhere.

One of the jobs the CE might have would be to replace an 8 inch wire with a 12 inch wire. This would be because the extra 4 inches would slow the signal down enough to make the machine more stable.

There were a LOT of pins. An expanse about 4 feet by 6 feed with a pin every quarter inch. A wire would wrap around one pin and go to another. It made no sense how they went. You could have a whole bunch of wires crossing at some point. The whole thing was just a mass of yellow wires.

The pins went through and if you opened a gate, a lot more heavy duty than the covers, little cards were plugged in to them all. The cards, maybe 3x 6 inches, had electronic components on them.

The machine was designed in 1964, a time before most integrated circuits. Most of the electronic components were discreet. A card might have a couple of transistors, a few resistors, and a capacitor. Someone back at IBM figured out what to put on the cards, where to plug them in, and how to wire them.

Memory consisted of magnetic rings. A doughnut shaped thing with 3 wires. They were a 0 if they were magnetic in one direction and 1 in the other. two of the wires were used to pick an X and Y location. If you wanted to know what X128 and Y43 was you would send current through those two wires. The third wire went through all the doughnuts and it would sense what direction he active doughnut was.

A bunch of these frames were stacked to give you a cube that would hold 64K of memory. Two of these cubes could be in the main box and an additional two could be put in the box on the back. Every 8 bits of memory had an additional 1. Because the memory was so unreliable each byte used the extra bit to be sure an even number of bits was on. If you read a byte and an odd number of bits was on one had flipped and the memory was invalid.

I’ve seen pictures of laties at a bench stringing these things up. The doughnuts were so small I can not imagine getting 3 wires through, let alone having them all lign up nicely. The ladies got over 2.3 million of them, without any errors, together for our machine.

My second machine, also a 40, had 128k of memory and we got another 128, not from IBM. The second 128k was solid state and in a much smaller box. The install did not go smooth and took over a day. There was also a service problem. Any memory problem was IBMs and now IBM only was responsible for half.

The I/O channels also lived in there. The area for the bus and tag cables was a fairly large area.

General

When the machine hung you often would need to see all of memory. You changed the load address to the card reader, put a special deck in, and pressed load. It read some cards in and dumped all of memory to the printer. 32 bytes a line. One of the sites we rented time from did not want to spend the time copying down the lights, so they had a Polaroid camera and would use that to take a picture and then IPL immediately. You could also dump memory to a tape quicker than print it out and then print it out later.