*** Please note, this page (and web site) are in early development.
Items are certainly not complete, and may be inaccurate.
Your information, comments, corrections, etc. are eagerly requested.
Send e-mail to Ed Thelen. Please include the URL under discussion. Thank you ***

Go to Antique Computer home page

Example Tour

Goal of this page
Provide an example tour of the Visible Storage area,
combining tours given by Dag Spicer, Mar 29 and April 5 & 25, 2000. [ I (Ed Thelen) have made slight "enhancements" based on personal experience with vacuum tubes and CDC6600s. ]

Table of Contents

Visible Storage

This is not the final computer museum, this is what we call Visible Storage - a euphemism for a fancied up warehouse

  • however, all the really cool artifacts are on display
  • so all you are missing out on marble floors and a cafeteria and a gift shop
  • we have a 3 acre grant of land just across the road there and we will be building a 114,000 sq. ft. building to be completed in 2005. In the meantime we will build a 40,000 square foot building near by to be completed in August 2001.

And Hanger 1, which is right beside you, will be an Air and Space Center. The Shuttle ENDEAVOR, that's the oldest shuttle, will be in there, It is coming off line in two years. So we will be part of a cluster of museums.

We have the largest collection of electronic computing artifacts in the world, over 50,000 objects.

Less than one percent (1%) of our artifacts are on display here, these are some of the greatest hits.

There is more storage area in the rest of this building, another 10,000 square foot building two blocks away and an administrative office about 1/2 mile away, also on base.

We are not just metal - like you see here around you. We also have a large collection of papers, documents, media, and video archives. If you are doing some research on computer history, we have an amazing collection to help you.

Spin-off of the Boston Computer Museum

We are basically a start-up and a spin-off at the same time, because we spun off of the Boston Computer Museum when they had trouble remaining solvent, yet they had this incredible collection that basically no one was seeing because they changed into a science center teaching kids about 1s and 0s rather than displaying this collection to adults who know what a computer is and want to know history.

So that is why we are out here - and with in about 3 years we will have the final building, a home. It will be like a real museum and the other half of what we do now is a web based museum. Its pretty rudimentary right now, so if any of you have any spare time and want to contribute CGI or JAVA skills. We are looking for those skills, just like every one else.

Please interrupt me at any time if you have a question, or if I'm going too slow or fast let me know.

Do you have any special interests in computers? Hardware? Software?

census, Hollerith, IBM

Let us start here. This is a punched card from the 1890 census. And this is a machine, a replica, done for us by IBM about 10 years ago, of the original 1980s Hollerith census machine. If you go to IBM's web site

http://www.ibm.com/ibm/history/
and click on the picture of the 1890s they have this as the oldest fossil they have to show the roots of IBM.

In 1890 the Census Bureau had just finished the 1880 census, just a few months before. And they had to start the 1890 census, and they knew they were in big trouble because millions more immigrants had come over in the intervening 10 years. - So they put out a tender to automate the process and this German/American engineer Herman Hollerith came up with this Census Machine as it was called. And it works, its really really basic - you have steel rods here that are mounted on springs and under here are discrete pools of mercury. So each hole is one little cup of mercury with a wire that goes to it and each hole here maps onto one of these counters. So, when you put the card inside this mechanism, then close this over it, and where ever there is a hole, the rod goes through, makes an electrical contact, so the corresponding counter will click over one position.

So all this is a tabulator, all it does is add. It doesn't subtract, multiple, divide or do any else. So at the end of the day, you could read these dials, and come up with your totals.

So they could do 7,000 records a day with this machine versus 700 per day with the old system. So they got a 10 fold improvement in throughput, so they finished the census in about 3 years.

It cost the same because Hollerith charged a fortune for these and became really wealthy, but they got it done in time.

Guest - and the workers died of mercury poisoning?

We would probably have to get a special permit from OSHA to put mercury in this machine - I don't know why, I used to play with mercury in high school. I'm still normal - right? ;-)

The census workers still went around and filled the forms by hand, and the cards were punch at the office.

And here is a sorter - a peripheral for the tabulator, it has 26 little slots, one slot for each letter of the alphabet. It is under solenoid control, so you could wire this up so you could sort by what ever you wanted, say first letter of the last name. So when the card was read, one of the little doors would open, and you would manually put the card in, and manually close the door. This was somewhat better than totally manual, sort of semi-automatic setup.

And it looks primitive but they still got the 10 fold improvement in throughput, and they could get data they couldn't get before either. You could aggregate data and get long term trends - which they never had time to do before.

Hollerith's company merged with other companies, one made butcher scales, and another made time clocks for employees. So in 1924, Hollerith's company, which was called "Computing- Tabulating- Recording Co." became "International Business Machines Corp." - IBM - and so this is the earliest element in the fossil record of modern computers. An the fact that IBM has it on their web site shows they are trying to make that connection.

All of the machines you see here are punched card machines. They all work on the same principle. All of the IBM machines still work. They were built at a time when most of these machines were leased. So when you have a leasing model for your business you want the things to last as long as you can. You might remember the old telephones, they were actually pretty reliable - now they are disposable items. They are now sold, not leased.

It is important to realize that even though a new technology becomes available, the older technologies do not vanish over night. As an example, transistors and vacuum tube based computers were invented within a few years of each other - however it took about 10 years for transistors to become inexpensive enough and reliable enough to rather slowly replace vacuum tubes in computers.

Another example is that punched cards are still used, in a few places and situations, even though they were invented over 100 years ago and became uncommon for computer programming and computer data about 25 years ago. Many states use punched cards in voting booths, and the U.S. Cotton Exchange uses punched cards to trace bales of cotton - each bale has a punched card with it.

And this is how you program these so called "UNIT RECORD" machines. That is for every card you have, you have a record. These machines are "programmed" by using these control panels. These machines have a fixed architecture, like a bunch of counters and latches and relays. And you actually control their interconnections by these control panels - can you pass this control panel around?

So the thing about these machine is that the coordinating intelligence (the CPU) is actually the human, the person that is physically walking around with these card "decks" and putting them into the input part of the machines, and taking them out of the output part of the machines. The person keeps the work flow intact - because all these machines do is one thing. This one for instance, the only thing it does is print out in characters the holes in the card. It prints on the card, in human readable form, the machine readable holes. That's all it does. So that is a very basic elemental function.

So what you see here is actually a subroutine library. Its a shelf with all these little routines, each one of which does one thing - that one there says "Time Cards" on it for example. There were no programs like written in BASIC or COBOL.

These two machines at the back here were actually in use until [the fall of 1998]. They were used by a major U.S. auto maker (General Motors, actually AC Delco - their Human Relations department). This model was actually introduced in 1948. The model was in use for 50 years.

And the story is that General Motors got rid of it because it wasn't Y2K compliant, I don't know if the guy was pulling my leg or what.

So we are actually working to set up a working punched card installation - like 4 or 5 of these machines and having people be able to do stuff, like sort a list of names.

(Short guest discussion of the use of the drum card on the 026 key punch.)

And this machine back here, all it does is sort. There was a special on PBS on the game show in the 1950's that was fixed - did you see that one? Did you see where the producer of that show actually pointed to one of these and to make it look like the computer was generating a question or an answer, they had a bunch of cards go through this sorter, totally fake, and anyone who knew about computers would just think it was laughable, but it looked kind of scientific.

Guest - have you ever seen handwriting analysis at a fair, they put some cards through, pull them out, and somehow that is supposed to tell you about your handwriting. ....

The machine behind you (this is out of time sequence, but we did get it from the same place) it is an IBM 4381 mainframe circa the 1980's (its not that old) it totally functions - we have all the cables, manuals, software, and it goes out to the end there. (We are going to have a look at it a little later.) This cabinet here is the disk controller, - yes, it fits on a card now, and the disk drives there, about 150 megabytes, the tape ...

There is a big social architecture behind these machines - unions for example - people's jobs depend on it. It takes 8 people to do this job - even though one guy can do it on a lap top now. What to do with those other seven jobs, so they want to keep it.

Punch cards really got a big boost in the depression when Social Security checks were actually issued as punched cards, and also during World War II, there was a huge demand for equipment to keep track of all the logistical requirements of the war.

World War II, Desk Top Calculators

Many new large guns were designed at the start of World War II.

One of the requirements was to produce firing tables which were manuals that gunners took out in the field with them to aim their big artillery pieces like 155 millimeter howitzers, and rooms full of women who were called computers, (computers were people before they were machines - just like a typewriter was a person before it was a machine) sat at these machines all day long and figured out these firing tables.

They used these DESK TOP CALCULATORS. Rooms full of women sitting at these machines, reading numbers, punching in numbers, pushing a control button (+, -, *, /), reading the results from the dials, writing the results down so that the numbers could be used in the next stage of the calculation. Hours, days, weeks, months of reading, punching, writing numbers. And there is lots of room for error, human errors, machines wearing and giving bad results, ...

And they used methods, like finite differences methods, so you could tell at the end of the calculation if you got the end points right, all the intervening ones were also right. But it was extremely time consuming - and it could take them a month to come up with a firing table. This was like a real hardware/software divide, because the hardware (the gun) could be built relatively quickly, but the software to run it (how to aim it - there about a dozen variables, like air temperature, humidity, size of shell, how far you want to shoot it, [barrel temperature, powder temperature], and so on)

There was a real software problem - software lags behind hardware - just like today. [Guest laughter!

World War II, need for faster computations, ENIAC

So what the government did to solve the big problem of generating firing tables, was to encourage a couple of professors at Penn (University of Pennsylvania) to create this machine called ENIAC. We have a piece of ENIAC so lets walk down towards the door there.

Its an electronic computer, which meant vacuum tubes. This is it right here. You can see some of the tubes. This machine, when it came out, was the size of this wall [pointing above the SAGE], down to the back wall and then back, all the way up. That's how large it was. It was a giant "U" shape. That was the size of ENIAC - 18,000 vacuum tubes, 150 kilowatts power consumption, [Guest ask "How Much??"] Yes, 150 Kilowatts, and it was considered by Americans as the world's first electronic computer.

This is a pretty rare artifact, a piece of ENIAC. There is one in the Smithsonian, University of Pennsylvania (where it was developed), and may be another museum.

And I say by Americans because every country has their own inventor of the computer, the French, the British, the German, and probably many other countries.

But the ENIAC has a good claim to that. The only problem was that it was not stored program machine. So that's one thing - people say that it is not really the first computer because its got to have that stored program, the von Neumann kind of concept.

But anyway, what they learned on this machine was very useful to the later computing industry. The project was not done until 1946 - which was too late to help in the war effort - you know, for computing firing tables.

Never the less, when they did finally get it running they could do a firing table in about 10 minutes - versus a month. So it took a couple of hours to set up the problem, which was done by women again, who were programming this machine mainly by patch panel. So the metaphor of the plug panel from the punch card equipment is still in this architecture.

So patch panels and rotary multiposition switches - they would set constants up by each rotary switch zero to nine - is one digit of a 20 digit constant - endless huge racks of this filled with knobs and patch panels so it took a while to set up.

[ Vacuum tubes were the major reliability problem. There were several problems.

  1. the turn on transient was tough on the tube filaments. So:
    • filament voltage was applied in small steps. Slowly warming the tube parts.
    • leave the tubes "ON" if at all possible
  2. It was discovered that tubes ok for audio (always conducting at least a little bit) sometimes failed when hot, but not conducting for a while in some logic function. It was discovered that silicon was the cause, and the manufacturing process was modified to eliminate silicon.
  3. Tubes were run very gently, not at full rated voltage, current, or power. This greatly increased their lifetimes - many lasted well over 5 years.
  4. Because "old" tubes (in use several thousand hours) tended to fail, many high reliability installations changed out "old" functional tubes for tubes that had been "burned in" (run for several hundred hours and still tested good).
]

[Guest - "How many vacuum tubes?"] 18,000 vacuum tubes.

So it was a real problem - they learned this, basically at government expense. The next stage in this machine was basically the von Neumann type architecture.

Anyway, the ENIAC was used for hydrogen bomb calculations, after World War II, so it was used for another decade, until I think 1952. At which point there were so many problems, just like a car - so many things wrong with it. They just shut it down.

In many ways, ENIAC was a project failure - it was not completed in time to do the job it was constructed to do - generate firing tables. In other ways it was a success. People learned how to design computers from ENIAC - and how NOT to design them. For one thing, this used a decimal system, and a serial architecture (adding one digit or bit at a time). Both were rejected by von Neumann in the so-called which is what all PCs use now.

This is the last digital computer to use patch panel and hand set dial in such a major way. (There were a few dead end processors that continued this style - but the style was definitely obsolete.)

Another success of the project was that the folks that built the ENIAC taught a whole generation of computer pioneers - so it was kind of a Woodstock event. These people went out into the world and made their own machines - with the new von Neumann architecture.

the von Neumann type architecture, Johnniac

And we have an example of the first wave of machines of the von Neumann architecture. It is a one of a kind, named after John von Neumann, hence called "Johnniac". It was built at the Rand Corporation in 1953 and here you see one of seventeen machines, all in research labs like Argonne, Sandia, Los Alamos, and so on, that were based upon the von Neumann architecture. A binary, parallel, stored program - still uses vacuum tubes, and if you walk around you can actually see these wires a little better. These are your address and data lines. These piano wire sized conductors stretched the length of the machine, containing hundreds of volts - a bus.

Here at the end of the machine are meters indicating voltages. These are grasshopper fuses - indicating fuses, if a circuit pops, you can see it.

Experience with ENIAC and this Johnniac demonstrated that vacuum tubes were not quite ready for prime time in a digital (ON/saturation or OFF/cutoff) mode. Until computers, tubes were always run in a linear way, not on/off, and there were troubles as a result of the changed mode. There was a lot of work back and forth between IBM, RCA, and users to get the characteristics right. Its kind of surprising because tubes have been around for quite a while, but they were never used in that mode until computers. They finally found the silicon, so useful in transistors, was a big problem in this new mode if there was any silicon in the cathode.

This is 1953, and this computer was used in early artificial intelligence by pioneers like Newell & Simon It used core memory. Does anyone know what core memory is? Yes - Old-timers - . It used 4 K 36 bit words core memory for high speed memory, and drum memory for medium speed memory. And what it could do is that there were about a dozen of these terminals around the Rand building after about 1961. And they had a time sharing executive running on this machine way before Multics at MIT.

It was a time sliced algorithm, round robin. The problem was that the time slices were in minutes - not in milliseconds or microseconds. So you had a light on here that started flashing when your time was about to end. You would then enter some key sequence to swap your job from core into drum, where it would live until your next time slot. There were about a dozen stations, so each station would get a crack every two hours.

The operating system was called JOSS - "Johnniac Open Operating System". There was one error message - "EH?". When it did not understand something it would type that out.

This is still way beyond other people's experience.

[Guest question - was this machine used to break codes?

No, they used very specialized machines, the British for instance used machines called "bombs" that were basically reverse Enigma machines. We have an Enigma at our office. There are thousands left. After the war the British gave many to developing third world countries, wasn't that nice of them? They gave them the machines that they could read the codes of. For their embassies and so on. Enigmas are really cool machines, easy to understand. The one we have works. Basically its just a flash light circuit, a key, a scrambling mechanism, and a lamp - that's it. So its quite easy to build and they are very rugged.]

Anyway, the ENIAC was used for hydrogen bomb calculations, after World War II, so it was used for another decade, until I think 1952. At which point there were so many problems, just like a car - so many things wrong with it. They just shut it down.

SAGE

OK - the next machine we are going to look at is a really specialized military computer called SAGE. Has anyone heard of SAGE? Yeh, OK -

This is a control console for the SAGE.

Well SAGE stood for "SemiAutomatic Ground Environment". Which doesn't tell you much about it. But it was basically a 1950s era Star Wars system. The U.S. Air Force created a series of 23 interconnected stations around the U.S. which monitored the air space above the whole continent, and the idea at the time was that mad Soviet bombers on suicide missions would come over Greenland and the eastern seaboard and drop an atomic bomb on New York or Washington. They wouldn't have enough fuel to fly back to Russia.

So that was the threat, a bomber threat, that this was designed to counter. By the time the last "Direction Center" as they were called came on-line in 1961, the whole system was obsolete. So the point I'm going to make is that like the ENIAC, it was a project failure, like ENIAC wasn't finished until after the war - but it had astounding spin-offs - it was a real success actually if you look at the bigger picture. So, finishing off the failure mode of this, in 1961, ICBMs had been perfected - which are 50 times faster than a bomber - if not more.

So the system simply could not handle this much higher speed.

We had a talk on this about a year ago. People that worked on it, hardware guy - software guy and they called it a great peace time defense system, which was great as long as you didn't need it. And that was part of the success. In political terms, the Soviets didn't know its limitations and so had to assume that it was actually quite good.

And the idea was that this would pickup a bomber and scramble jet fighters and electronically guide missiles, including firing missiles from the intercepting fighters, to shoot down the incoming bombers. There were also missiles like the Nike Ajax and Hercules that this could designate targets to. A lot of them were actually nuclear tipped missiles.

This hand held unit is called a "light gun". I encourage you all to try it. And what you see in SAGE is the roots of so many technologies, one of which is Air Traffic Control. You had your whole air space (your sector) in front of you and every object in here is actually given a vector-character descriptor, just like you see in air traffic control, so you could know if it was friendly or not, and you could designate any particular point of interest, and it would magnify it and tell you more about it.

It was a sophisticated system, like the action you take could take many directions - from you could "shoot it down" to "no, its an airliner". You notice the built in ashtray and cigarette lighter? It you think in even more abstract terms this is the forerunner of the light pen and even the mouse, because its an interactive device between your hand and the screen.

Using the light pen, you could focus in on a particular plane, get more data, pass the plane off to another control console.

And this is the first large scale network. The 23 centers were linked together by modems - which were developed for this project. It was developed at WesternElectric/BellLabs, and they were responsible for hitching these all up using phone lines. There were thousands of people using these 23 machines.

Another new feature was the on-line real-time data base control.

This thing ran on drum memory, there were no interrupts, it just executed one task after another. It was extremely reliable. We were talking about the mean time to failure of the ENIAC and Johnniac which was about 8 to 10 hours at this time in the mid-1950s. This SAGE was down about 10 minutes a year, thousands of times more reliable.

The way they did it was - IBM was the main contractor. They did extensive testing on these modules, including marginal checking. They would raise and lower the tube voltages to verify that there was plenty of reserve life in the tubes.

And they would swap these modules out according to a fixed maintenance schedule - even if there was nothing wrong with them. These modules are hot swappable. This was in the mid 1950s.

And it had duplex CPUs - that was another big development. There was one CPU on warm standby all the time. So that makes a huge difference in reliability - the system never really went down.

There were two separate computers, and this is not even 1% of the machine. The whole SAGE system is about 4 football fields, when you include the generators, the phone system, the air conditioners, a bunch of radar stuff. - Four football fields of equipment, stacked four floors.

We got this in 1982 from Canada. It was in North Bay, Ontario. It was the last functioning SAGE system. NORAD and the Canadian government let us take a few pieces of it to preserve it. I don't think anyone else has it. Maybe the Smithsonian has a little bit.

The total SAGE system (23 centers) cost 8 to 10 billion 1960 dollars. So it was quite expensive - the cost is technically still classified, so the money is just an estimate. It cost way more than the Manhattan project (the atomic bomb project). To get today's dollars you can multiply by about 10 - maybe a hundred billion dollars.

And like Star Wars, national missile defense, it did not work either. It is a technical solution to what is really a social problem - the Russians and Americans can't play together very well. (I'll just leave it at that.)

Another thing about SAGE was that it was very good for IBM. They got about a half a billion dollars in profits from that contract - and some people speculate that with out that contract they would not have become the really dominant player in computers - because they used that money to develop the 360, which is still the most successful computer of all time - over 100 billion dollars in sales.

Core Memory

We are stopping here to show you core memory. This is from the Whirlwind which is the first computer to ever use core memory. In 1953, Jay Forester at MIT, a grad student/instructor there, developed the core memory for the Whirlwind.

If you look over the 50 years of computer history - from about 1950 to 2000, (I'm sort of shaving off the earlier developments), core memory was the dominant memory technology. Is not DRAM or anything else. Core has been around for over 30 years or so. It will take another 10 years or so for DRAM becomes the leader in terms of longevity.

Anyway, the way it works is it uses the little donuts or toroids of ferrite - powdered Ferro-magnetic material, and you magnetize it clockwise or counter clockwise to get a zero (0)or a one (1). So there is an X-Y matrix to address the individual core, and there are one or two sense lines - there are 4 or 5 different schemes for core memory. Some of them are destructive read with regeneration and so on - that's not important. Each donut is actually one bit, so here you have a thousand bits. So how many bits are in this whole thing (pointing to the 4x4 array of 1024 bits each but with one piece to the array torn out.)

(A guest mentions 15 K bytes.)

Yes, I accept either 15 K bytes or 16 K bytes. You noticed that the one memory plane is missing. :-)

And it is about a dollar a bit to make.

So, if you go down to Fry's [big local electronics store] how much is 64 meg DRAM? It's about $64, right? Roughly - That about a dollar a megabyte, instead of a dollar a bit. So its a factor of about 10 million times cheaper. That is how much the price has dropped.

Kind of a neat thing about core memory is that it is non-volatile - so even when the power went off, all your stuff was there. They still use it in space craft. It has certain inherent radiation hardening characteristics.

[ Some early "recipes" for making the core material were very temperature sensitive - and some systems - like the IBM STRETCH - immersed the cores in transformer oil to keep a more uniform temperature. ]

This is core. Ten years later it looked like that [pointing up where?] - this is about 1965, so its getting smaller and smaller. In about 1980, which was the last year that core was really made in volume, it was really small. I have a pepper shaker on my desk with the cores, and you can just shake it out - its incredibly small - like it fits through a salt shaker. So core memory hung around as long as it could. I have a sample of about the last core plane built, and its a core plane surrounded by ICs - like driver ICs like they are gobbling up all the real-estate on this circuit board.

Any questions on core?

Guest - how fast was it?

Another guest - I fixed General Electric computers, the first generation was 18 microsecond read/write cycle time. The next generation was 5 microseconds, the next generation was 2 microseconds. The CDC 6600, back there, had a core memory cycle time of 0.475 microseconds, 475 nanoseconds.

Guest - what is DRAM time in comparison?

Another guest - common DRAM is about 60 nanoseconds, SRAM can go 5 to 10 nanoseconds.

There's some neat retro computing, I don't know what you call them - hackers? I'm sort of one, involved with them in a way - where you build these really strange hybrids, like a PIC micro controller that uses core memory - some weird old and new technologies.

Seymour Cray

Right under this memory is one of Seymour Cray's first machines. It is a military machine. Seymour Cray is considered the father of super-computing and basically defined the whole industry for 30+ years. He made the fastest machines - that was the stated goal of all his equipment. Just build the fastest, and money was no object. And during the Cold War, money was no object.

There were some gross fights for money between the various military services, and the traumas of the 1973 defense spending cutbacks are well documented. The book "Shield of Faith" by B. Bruce-Briggs describes the almost shameful fights for money for air defense between
- the Army (pushing Nike missiles)
- and the Air Force (pushing BOMARC missiles).

Most of his customers were federal weapons labs at Los Alamos, Sandia, Livermore, and so on. This had 10,000 transistors, cost half a million dollars, and was Seymour Cray's first mass produced machine. He did come from a company called ERA which made magnetic drums for the Navy, the precursor to the NSA (National Security Agency).

And right next to it is a prototype built by Seymour Cray for one of his machines. Hand made - so this is kind of neat. You can see a core memory right here that he built as a little scratch pad memory. Core memory is kind of neat because all you have to do is count the cores to find how big the memory is.

Guest - that looks like a log of counting

OH - not so bad as it looks. Most memory planes read out one bit per plane, if the computer had a 16 bit word, you used 16 memory planes, one bit per memory plane. And the memory planes were almost always square, so you could count the cores on one side, square that number, and that is the number of cores in the memory plane. A common size was 64 cores on a side and 64 squared is 4096. So there are 4096 cores in one of these grids.

PDP-8

Computers were considered enormous things, almost like a cyclotron. Some of the high end scientific computers were in that range, ten to fifteen million dollars.

Right behind you here, in 1964, you have really the first successful mass produced computer, the so called "mini-computer", the PDP-8. This sold for $18,000 in 1964, which was about a tenth of what people mentally thought a computer should cost.

It was advertised pictured in the back seat of a Volkswagen Beetle (convertible) to make the point that it was really small. And these sold like hot-cakes because they were within the spending authority of middle management - $20,000 to $25,000 was typically their cap. They could buy these, and they sold by the tens of thousands to not only government and universities, but process control.

A lot of corporations put their toes into process control using this equipment. There was a lot of stress at the beginning putting a process under control of a computer. There was a big sugar factory in Boston that was an interesting history of them negotiating with Digital about being concerned about dumping hundreds of tons of sugar into the Charles River because they were worried that this would not be reliable enough and might run amok. And maybe make them bankrupt.

Part of the BART system runs on a PDP-8, not this exact model, a model that was made about 3 years after this one.

IBM-1401

One of IBM's first mass produced systems was the 1401. Did anyone use a 1401? IBM sold or leased 12,000 of these systems. This is a huge huge number, probably the largest single computer family made until that time, or until the System 360.

The CPU is there, and if you have digital watches they probably have more logic in them than is in this CPU. Here is the card reader/punch, that is your "I" and your "O", Input and Output. Here is the 1403 printer which could do 600 lines a minute. The model 2 could do 1200 lines a minute. That's really fast - paper would go flying out of these things.

The 1401 could be used in either of 2 ways:

  1. as a print server
  2. as a stand alone computer for a small business
As a print server, they would use this in combination with a larger main frame installation. Main frames at that time were serious systems, for an insurance company or something, would rent for $100,000/month - $200,000/month - really expensive, so you did not want to waste computer cycles printing out payroll checks or what ever. so they wrote them to a tape then bring them down to this 1401 system which do all the printing and low level tasks.

So you might notice here that most IBM equipment has elapsed hour meters. You can see one here. The reason is that your friendly IBM Customer Engineer would come and read these once a month, as well as lube the equipment. So when you rented this, you rented it on a per shift basis - one shift, two shift, three shift. You were given a little margin, but if you had a one shift contract, and you went into two shift territory you really got dinged. Basically, your rent almost doubled. So that was one of the functions of the Customer Engineer was to check these meters.
John Van Gardner - who was an IBM CE - reports (March 2006)
The CE did not read the meters once a month and report it for rental purposes. The Customer had a form to fill out and they had to do it. You may have seen a CE read and record a meter but that was done on some machines and sent to the plant for trouble analysis and things like mean time between failures. We also reported the meter reading at the completion of an installation when the machine was turned over to the customer and the final meter reading when the machine was discontinued.

Its hard to appreciate how expensive computer time really was unless you had to pay for them. It really had to be parceled out very carefully. Any questions on this?

Guest - does any of this still work?

Yes, much of this punch card stuff, like the key punches, still work. That's another lesson you learn from technology is the interplay of technology. When you have a leasing model, like the old phone company used to, where you would just lease the telephone - you couldn't buy them - it was in their interest to make them really reliable, because every time they came out they lost money. But if you buy stuff, and it breaks down, well that's too bad - you can buy another one.

----------- little related story about IBM's mastery of data processing ---------
About 25 years ago, the company I was working for purchased a new "no name" accounting/data processing system. I was describing some of the usual trials and tribulations involved with the hardware and software to a friend. My friend was not properly sympathetic.

"Well, What do you think they should have bought?" I asked.

"IBM, of course" he replied.

It took a while for me to recover. We had both worked for both competitors of IBM for years. And we both agreed that IBM was the current "Evil Empire". How could he possibly recommend IBM for the task?

Sensing that I was gasping - he soothed.
"Look at the IBM printer, it works well?"

"Yes" I reluctantly replied
"It is the most reliable that you know of?"
"Yes" again I replied
"It is the easiest for the operators to:
- insert paper
- align paper
- open the hood
- stack paper
has the best looking print,
and the best maintenance?"
"Yes" was the only possible reply

"You can get one for about $25,000, about $13,000 more than your no-name printer. Over a period of 10 years do you think the frustration, down time, and ruined runs are worth the $5 per day you are saving with the no-name printer? Will you really save money?"

There was really no answer - and so it was with all the equipment accounting hardware and software.

(And to make matters worse, the no-name company went out of business in the next year, - leaving my company "high and dry", - and painfully converting back to IBM, - and with a new director of Data Processing.)

----- end of true story -----
----- quoted text is approximate ---------

Kitchen Computer, one of many "Dead Ends"

Behind you here is a machine called a "Kitchen Computer"

What we try to show at the Museum is that not everything is a success. It didn't go from the Abacus to the Pentium and not make a lot of interesting detours.

In 1969 this was put on the cover of the Neiman Marcus catalog for $10,000. It is actually a stock Honeywell mini-computer. It came out in 1965. So Neiman Marcus sold this as a way for housewives to keep their recipes on. It was $10,000 which was the price of a house in 1969. There was no I/O so you had to use the keyboard switches to program, directly in decimal, every character, totally ridiculous - right?

Guests - much laughter

You know - set up your address, - load your address - load your data you know, lather, rinse -

So you actually bought a two week programming course with it - you know the ridiculous thing is - if a house wife could afford $10,000 on this, she would probably have here own cook. That was a good annual wage. That is more of a marketing story there.

IBM Stretch

This is a piece of a machine called the IBM Stretch, finished in about 1960 - and was an internal research project, although IBM wound up making 8 of them. This is a tiny fraction - we have actually the entire system. If we unpacked it, it would take up three times this entire exhibit area.

Stretch was really important because it was basically a sandbox for hardware engineers and software people to - they were told by IBM management, to push the envelope, do what ever you can in which ever your specific field is, do the best you can. As a result, it had 765 instructions, like the ultimate CIS [Complex Instruction Set] machine, if you include all the modes and uses of registers and so on. It could be decimal OR binary.

It was about 8 million 1960 dollars - and basically the only people who ordered this were weapons laboratories, and one went to NSA as the fastest machine, one always goes there, at least one. Anyway it was a really important machine because it has a whole bunch of concepts that microprocessor designers re-discovered 35 years later - like this is 1960 meaning that design started in the late 1950s.

It had things like look-ahead, pipelining, register renaming, speculative execution, it just goes on and on. All these things that guys at AMD and INTEL think are hot technologies have been in mainframes for 40 years. They don't know their history. They keep re-inventing the wheel. I had David Patterson in here and we had a good talk about all the good things that came out of Stretch. It was really amazing.

A lot of people think that if IBM had not done Stretch, that IBM would not have had the internal know-how to do the IBM 360, including manufacturing technology, packaging technology, and just architecture, what works, what doesn't work.

And all the folks are still alive. We are trying to make a talk on Stretch in the next 6 months [this is April 2000] involving people that worked on this machine.

Control Data 6600 & 7600

This machine in the corner is kind of the Rolls Royce of computing in 1964. It is the Control Data 6600 -

Guest - I did my thesis on serial 13, the one where they turned off the cooling with out turning off the power. This was at the University of Texas

This is half the machine, it actually forms a big cross. We only show half to save space because we have so many goodies here.

It was the fastest machine in the world, about 8 to 10 million dollars - in 1964. It was especially good at scientific processing because it had such fast "floating point" processors. It could do a 60 bit floating point multiply in 1 microsecond, and it had two independent multiply units. It had a total of 10 independent arithmetic or logical or memory access functional units. It could issue 10 instructions per microsecond (one every 100 nanoseconds) when correctly coded for the right kind of problem. I hand coded a Fast Fourier Transform subroutine for a competitive benchmark that ran at 8 instructions per microsecond.

Another nice feature was the 10 "Peripheral Processors". These little 12 bit processors had independent program memory and could also access the main memory. They were used to

  • schedule/control the main processor,
  • perform I/O, such as drive the printers, tape units, disk units, etc.
  • drive the operator's console, and other required tasks.
Can you tell me why they have some of these large loops of wire in here?

Guest - timing

That's right timing, That's the secret of good comedy too - timing.

Guest - they also had heaters in there and refrigerators and compressors in there

Right - you can see that right there, each part of the "X" of the machine has a compressor right there.

Guest - Each of those modules is a standard size module, and in some areas they are denser than in other areas. And in areas where they weren't very dense, it would get too cold, so they actually had heater modules to keep the temperature distribution constant. I was hanging around the SEs [Service Engineers] when they were installing the thing, and they bring in the four pieces, and all the cables and plug it in. And they said "See that, it doesn't do anything, its just a heater."

The thing that blew people's minds was that this was a very clean machine architecturally, very simple, even now you can read the essentially "Principles of Operations" of the machine, and you can understand it. Whereas if you try to understand a Pentium or an AMD, that is a couple month project.

This was a very straight forward architecture, and so when people got it they said "this looks so good on paper, why does it look so messy - why are there big loops of wire. and the explanation is that the machine is tuned - like a Maserati - not a Pinto. so every wire is cut to within 1/16 th of an inch to account for the propagation delay of signals through out the machine. Jack Rubin points out "if I were concerned with _real_ quality, I would make sure the reference would be to a pre-war (WWII) Maserati!"

What's the clock cycle time? One instruction every one hundred nanoseconds, assuming no register or functional unit conflicts. So that is a 10 megahertz clock time.

There were all made by little Scandinavian ladies who had needle point experience, and so Seymour Cray had these ladies build his machines - he did not let men touch these machines, The wires are all color coded for length, and twisted pair for common mode rejection. And like our friend pointed out, built in air conditioning system pumping freon throughout the whole machine.

They sold over a hundred of these, so this made Control Data into a big company. Control Data had been a small company making small machines like you see there - the 160A, 1604.

Seymour Cray was this legendary figure, even more legendary now that he is dead. Seymour was building this machine, the 1604. After he had built 10 or so, he went to his supplier to get more transistors. The supplier said "I don't have any more, those transistors were all surplus." Seymour was building his machine with below par transistors, yet it worked well. He could build circuits that used sub-par parts that worked. That shows how brilliant Seymour was - he could compensate for lousy parts and yet build the fastest machine around.

So this serial # 1, it came from Lawrence Livermore Labs.


CDC 7600
The machine behind you is the follow-on. That is just a little piece of it with the wood grain panels. You can always tell what decade a computer is from by their package styles. So that is from 1969/70 and also serial # 1, also the fastest machine in the world at the time. That was what Seymour's design goal was. Money was not a large consideration.

Guest - What kind of languages were used on the machine.

Fortran for sure -

Guest - I put LISP on it, that's what I did. I was the LISP developer, distributor, maintainer, what have you, for this Control Data series machines. The instruction sets were effectively identical, and binary compatible.

And when you got the 6600 you essentially got no software, I don't think you even got an operating system.

Guests - Yeh you did, you got an operating system - SCOPE - no - actually two, you could get And COMPASS the assembler, and FORTRAN, Dick Nelson was Seymour's resident compiler writer, and he could write FORTRAN compilers in his sleep.

Control Data is one of the few companies that actually successfully sued IBM for anti-competitive practices. IBM kept announcing these supercomputers that were never even made - or even started. And Control Data was able to prove it. They settled out of court with IBM giving Control Data millions in cash and a division of IBM, the Service Bureau Corporation.

Magnetic Data Storage, disks, tapes

One of the things with super-computers is that they can produce oceans of data, so what do you do with it?

There is a three part hierarchy of memory that is used. The faster the memory, the higher the cost. So you have to trade-off speed vs. cost.

  1. core memory that is really really fast, but costs a lot, and is right in the machine for fastest access
  2. this you have medium speed memory, which in this case is disk
  3. the slowest and cheapest was magnetic tape


We have discussed high speed core memory. This is still using cores for high speed memories and registers are implemented in core.

-------------- comment! ---------------- no "super-computer" used cores for registers!!! That trick was reserved for mini computers that were really interested in saving a buck at the expense of speed. Remember that a core memory read/write was 1 microsecond or more in most machines (even in the 6600, the cycle time was 0.475 microseconds) ------- end of comment -------------------


And in the medium range speed memory you have this,
Disk Pack and Drive
- which might be called the ZIP Disk of the 1970s. Its a 100 megabyte removable cartridge and I encourage you all to lift it. You don't have to go to the gym if you work out with these.

So they used to live in one of these machines here, the size of a washing machine. When you closed the door it purged out the air in there and kept it kind of filtered, but - you can't get away with that now.

A typical super-computer room would have 40 or 50 of these going, incredibly loud, moderate speed.

IBM 3380 Disc System
Well, speaking of memory, I want to show you this. This is a classic IBM dual disk drive here, two 500 megabyte drives, heads from both sides. Mid 1980's

Guest - is this a Winchester?

Well - not quite a Winchester yet.

(This is just 2 drives, the controller is in this other cabinet.) There are at least two important technologies that this machine imbeds, one you can see, the other you can't.

There is a really important historical innovation here, can anyone tell me what it is?

It is this - a floppy disk! Floppy disks were not invented as a means of personal data storage. They were invented as a means of updating macrocode for this machine.

The thing you can not see is the IBM channel architecture which was basically one of the building blocks for SCSI. So this was kind of interesting.

If you are a hardware guy, being able to change/update things is firmware is your best friend because you really can't make mistakes. If you can make little mistakes you can update them with floppy disks - it was a beautiful thing - anyway that is where the floppy came from. Then people started using them for other things.

These things are built incredibly well, you could probably drop it out of an airplane and it would still work. IBM just made things really well. They still do.

IBM has an historical disk collection in San Jose, because most of IBM disk drive technologies were created in San Jose. They have agreed to donate their entire collection of historical disk drives to us when we get our new building


For long term storage, low speed - low cost, you would have one of these, something to do with magnetic tape. This is a tape robot, this could go on all day long, find tapes, mount tapes like you see there, pull them off, put it back into storage.

And there were various other schemes like these magnetic cartridges, cartridge based machines that held an equivalent amount of information. So - dependent upon how important in the hierarchy, you would more high speed storage than if you were an intern or grad student.

So one thing Control Data did was to poke fun at IBM was to take an IBM storage system model number and add a zero at the end. So that is the IBM number of the equivalent machine. So you pretty much knew what the machine was if you knew the common IBM model numbers. (A little foot note in the history of computing :-)

Analog Computing Section, forgotten chapter in computing history

Norden bombsight from World War II

IBM 4381

.

Connection Machines

Connection Machines - we actually have four of them. This is CM-1, there is a CM-2 - they look the same), this is a CM-5. This is a different approach than Seymour Cray's very fast basically uni-processor configuration - even though they have some aggressive parallelism in some cases.

This is a massively parallel machine, 16,384 CPUs in here. And the thing is, like Beowulf, yes, it is a super-computer, but only in a certain problem space. So these are really cool, and they had their hay day in the early to mid 1980s and a super-computer company that can last 5 years is doing really well. Seymour Cray's companies are basically the only American companies that have been in business for 30 years or so.

So these two machines were very popular. This is from MIT's Laboratory for Computer Science and the other is from Dow Jones and Dow Jones used it to play the stock market. There are some econometric models that you can put through there and they can be made parallizable and can tell you if it is a good bet to buy a certain kind of instrument. You know, equity or future. And you have a minute and a half to do it and this Connection Machine was the only computer that could do it with in that window. So they bought them, for like $5 million which is an hour's trading for a good stock broker.

Guest - We used one at ?Advanced Systems? over here on Shoreline to do image processing. Same kind of thing. We would send images through them, asking "is this a tank", "is this a tank".

Yes, anything that can be put into a table, are usually parallelizable, weather, hydrodynamic equations, computational fluid dynamics,

The Connection Machine-5 over there actually has a switch so that you can make it look like it is working even when it isn't. So the Customer Engineers used to set that switch when they were working on it because it was so upsetting to the customer to see this 5 or 10 million dollar machine just sitting around not working. [Much laughter from the guests.] Kind of a "Hurry, here comes the customer switch".

You could tell with experience what code the machine was running. Each CPU had 2 LEDs, 32,000 LEDs in this machine. It looked awesome when it was running, and people could tell what was happening. If the CPUs in only one quadrant were working, maybe the problem was not divided correctly (there were lots of programs or parts of programs that you could not divide up) One interpretation is that you are not taking advantage of the parallelism inherent in the machine. Just kind of thrashing in this one little quadrant. So it is really kind of neat, besides looking cool, it is really useful diagnostically. And programmers are always doing these funny things like on the Control Data console - the two tubes look like two eyes, and they programmed them to look like two eyes and peer around and appear to follow you.

Mini-Computers, DEC, et.al.

And the Whirlwind at MIT put out so much RFI (Radio Frequency Interference) that grad students who worked on it at night - when you signed up for the machine, you got total control of the machine, not timesharing. They could tell from 2 or 3 blocks away if the machine was running or down/dead.

We found a warehouse in New Hampshire that had almost everything that Digital Equipment Corporation ever made. It was actually part of the Boston Computer Museum, but not part of the museum - some complicated arrangement. We worked out something with COMPAQ who had acquired Digital a year ago, and they donated everything that was there. So in 3 50 foot long trailer loads (shipping paid by COMPAQ) we got that equipment - some here, and some in another 10,000 square foot - 30 foot high - warehouse.


PDP-1
Why don't we walk into here for a sec, and we will segue into the mini-computer which was basically designed among other people, by Gordon Bell. And this is the Digital Equipment PDP-1, and the whole concept of a mini-computer was counter-mainframe - that you should give people a computer they can interact with directly, they can re-boot it, re-load it, they can boot strap it, they had total control of the machine. Not like the existing paradigm at the time which was big computer room, glassed in with these priests in white lab coats that take your cards, and if they like you they will run your program. You come back in three days, so you are missing a coma, OK - so you try again - that's the iteration process.

Versus this which is hands on, right now, what we understand today is a computer. Well the guys at MIT and Gordon Bell (who is a co-founder of this museum) - they had this mental mind set from coming from MIT which was a computer of course you interact with a computer, what else is there? Right.

---- something screwed up in above paragraph -----

They figured IBM was not the way to go, cycles being parceled out, really sparingly, and so on.

This is a PDP-1, these hacker guys (when hacker was not a dirty word) from MIT created this program called "Space War" in 1962 (kind of a little "shoot 'em up game") and it was played on this about 8 years ago - we brought this machine back up. There is a JAVA applet that runs original PDP-1 code.

There are a lot of machines - we are not going to walk around all of them now - maybe later when you have a little coke. But it a history of Digital Equipment's mini-computers around the periphery of this room.


Model 360 - IBM's famous computer family
This is a model 30 - which is the entry level system. You know, a good Palm Pilot has more power than this. All of this, this, this - its is in your hand.

Here are the tape drives, 100 - 150 megabytes.

----------------------------------
Can we verify that?
usual tape was 1200 feet?  I think - 14400 inches
  density	max char
  char/in 	per tape
     200      2.8 meg
     555      7.9 meg
    2000     28.  meg
----------------------

East German clone of DEC VAX
Kind of interesting. You can see some Russian lettering, and some German lettering.

Micro-Computers, "PC"s

Altair
This is the machine that really started it all from the hobbyist point of view - the Altair - this is what was revealed to the people of his generation (our generation I guess) that there was a real demand here - a pent up market. And Bill Gates actually wrote his first BASIC for this machine. We have the paper tape - he donated it to us. BASIC 1.0

One of the stories that you hear in badly researched books is that he wrote it on the Altair - he didn't - he wrote it using a PDP-10 at Harvard. He did not have an Altair nor an 8080 so he debugged it running Paul Allen's 8080 emulator on the PDP-10. So he basically went down there [to Albuquerque, New Mexico] and it worked - first time. It was very good emulation, he deserves credit for that.

---------- a possible nit pick ------------- "Hard Drive" by Wallace and Erickson says that Paul Allen did the emulator, Gates did the BASIC, and that Paul Allen went to Albuquerque and gave the original demo - which indeed worked. ---------- end of possible nit pick ---------

And here is a ruggedized version of the Altair.


The Apple-1, two of them here, and these are the computers that Steve Jobs and Steve Wozniak sold for $666.66 and you got a blank PC board, and hand full of parts, and a 10 page photocopied manual. And you had to add your own keyboard, power supply. You also needed to buy an RF modulator to paint the screen on your family TV. A cassette interface was available as an optional extra at $40. Has anyone ever used a cassette interface?

Guest - It Sucks

Yes, its quite bad. It is slightly better than nothing.

Based on an order for 50 of these from the BYTE SHOP in Mountain View, just 2 miles away from here on Castro Street, Steve Jobs got parts to make the order, and Apple was born. (The BYTE SHOP doesn't exist any more.)

These are really collector's items now. You see them on e-bay for about $40,000 - but lots of things go on e-bay for silly prices.

Look here, you can see someone doubled the memory by piggy-backing the DRAMS and lifting the chip select. This permitted you to expand memory without cutting traces on the mother board.

They made 220 of them, and a whole bunch of them were recalled when the Apple-II came out. Which is over there.

They actually accepted them as a trade in - so that took a bunch of them out of circulation.

In 1976, three machines came out within 3 months of each other.

  • the Apple-II
  • the Commodore PET
  • and the TRS-80 - "Trash-80"
the Heath Kit


Couple of other machines - the "PC" did not exist until 1983, when IBM said it did. We think of it as generic, but it was really tied to the IBM brand name until a few years later.

- the LISA {probably Apple III - thanks Udo Schmitz} was interesting. It had a real manufacturing problem, every thing was in sockets, and there was some kind of bi-metallic junction formed between the socket and the IC. And with time, you would get these high resistance contacts - there would really be no connection any more. So the fix - from Apple - was to physically pick it up about 6 inches and drop it. As a way of re-seating the ICs - Its on paper with the Apple letter head - it says pick it up and drop it.

That might have been OK for the sockets, but it can't have been too good for the CRT [monitor]

What you have seen is just hardware. We collect software, technical manuals, sales literature, ephemera (tee shirt, coffee mugs) There is actually a book on the entire product history of Apple, based on tee shirts - its nothing but a history of their tee shirts.

It looks kind of trivial and silly, but it in fact tells you about the cultural context. Like all those bags and bags of stuff that you get at COMDEX - we try to collect that because usually it just vanishes - its produced by the millions but it vanishes - nobody hangs on to it for very long.

A couple of other things

- NEXT Cube
- B Box
- Anniversary MAC

Guest - DEC made a RAINBOW

Right, had an 8086 and a Z-80 because they didn't know which way the market was going to go. DOS or CPM?


This is a NOTE-TAKER done by Alan Kay. Alan Kay was here last Friday saying "Hi", and (shameless name dropping) and this is his prototype. The first portable computer ever done. And we show the OSBORN , because it is a total - let's say borrowing from the NOTE-TAKER.

Guest - the IBM PC was kind of like that too.

Right, and the EAGLE and the COMPAQ , this whole sewing machine form factor.

Does anyone remember the Radio Shack Model 100?

It looked like a laptop, laptop form factor, with LCD display. And it came out in the early 1980s. Anyway, it is interesting how the two form factors battled it out.

If you go to the Gates Building at Stanford, on every floor, as you go up the elevator, you will see displays that we did there. You can see little stories of computer history.


This is Computer Space. This was done by Nolan Bushnell - the guy who founded ATARI - and it still works. It was the first commercial computer game, it still works. This is the only one in existence so far as we know.

Chris - give us the word about this game

Computer Space was the first mass produced coin operated video game. It was based off of Space War which was famous, which Nolan Bushnell had seen it when he was at Utah. And you will notice there are these wonderful buttons - and this is the whole reason it did not work. They only sold about 2,000 of these. It was so complex to play, such as to fly straight you had to press both of these buttons at the same time. If you wanted to turn you released one a little bit.

There is no microprocessor in it. It is actually discrete components soldered onto a board. He had a ton of components left over. So he sat around thinking "I can either loose my shirt completely, or I can try to build something else.

So he came up with Pong which made HUGE money for him. This is the only one of its color and type, this is a little later than some, which still works, that is in existence anywhere - and its lovely.


And this is an IMP. Has anyone heard of an IMP?

Guest - yes - Interface Message Processor

Yes - This is kind of a router - one of the first that the Internet was built on. Everybody had a mainframe, and you would connect an IMP to the mainframe, and at the other end there would be an IMP connected to a mainframe, and the IMPs would talk to each other.

How many Hosts are on the Internet now? 60 million?

There is a map here showing the entire ARPA net at that time, so I encourage you to look at it.

The map was made in the 1974. There were about 30 hosts. And these are based on Honeywell mini-computers. Its what Cisco makes now in little boxes.

And Len Kleinrock at UCLA is thinking of sending his, the one the first transmission was sent on. And then we have a donor in Arizona who has the Xerox Sigma 7 computer that was the other end - it was at SRI. So if we can get those two things, we will have the original two ARPANET nodes. That’s as small as you can make a network, two nodes.

IBM-1620, Restored

Let me just show you this - this is a restoration project we did. We just got this running after 40 years. It took a team of 8 experts a year to get it running again. It is an IBM 1620.

This shows how we view computer history. It is important to let people understand what we did and what IBM did. So everything we did is in red wires here, and the problem was that this core memory in here disintegrated - it was 40 years old. So we designed and built a semiconductor replacement, that you can see right here - and wired it in. But it is very easy to see what we added or changed. Just about a month ago [March 2000] ran the level zero diagnostic which means everything is working on this - after 40 years.

So there is a team of really cool volunteers that basically have day jobs, like all of us, and yet they were really into resurrecting these - on both the hardware and software side. So I encourage you, if anyone is interested in those kind of projects to let me know.

Guest - Is your goal to get everything up and running, or just a few things?

Everything is impossible. Like how do you turn on a Cray-1? The power requirements, the motor-generators, the air conditioning, the restoration, the maintenance, the special semi-conductor parts that are no longer made, the technical experts, ... - The job is basically impossible. So some of the factors of selecting artifacts for restoration are:

  • cost
  • interest
  • availability of experts
This was not really our first choice of machine in terms of importance to computer history. Of more importance would be the XEROX ALTO, DEC PDP-1, the IBM 1401, but there were guys around here that really liked this machine, and they put in thousands of hours to get it working. And we did.

Another thing this kind of project does is that it really shakes the tree, not just in terms of getting people to work on the machine. These three hundred thousand punched cards of software for this machine (here) we got as a result of someone hearing about this project. This professor at Purdue, who was cited by the fire marshal to get rid of these. And we got them with a week to spare, or they were going to go. So, 300,000 punched cards, all software for this machine. We now probably have the world's biggest collection of software for this machine - the 1620 -

Fifty megabytes, so half a ZIP disk. And what we are doing is reading it in with this card reader, through a PC interface, we bring a lap top, dump it in there, then burn a CD.

We have an emulator right here which we built. You probably wonder what all this life support stuff is. Its got all this ultra high tech stuff. The problem is that we did not have a console until late in the project, or card reader or card punch. So we just built fake ones here, a VME based system that plugs into a PC.

So we take the software from the PC and upload into this. It actually runs on this, and we also have a JAVA emulator that runs this. So you can go on the web and run this same software, and it looks exactly like this control panel. All the lights move exactly right.

And from our point of view, probably the emulator is the way to go to preserve it for the future. Because the guys that got this going - this thing needs routine maintenance and 50 years from now they probably will not be around - while, depending on your outlook, JAVA might be. So that's what we are hoping anyway. So if you do an emulation, its pretty much platform independent, it doesn't need any hardware, you can just float out there - you know, you can mirror it all over the place and the code can be mirrored around the web to so - Anyway in another year we are choosing another machine because were basically finished with this one.

Guest - so this is on your web site right now?

Yes, we keep really good notes, you know, lab notebook, really well done, documented, photographed, everyone signs off on this stuff. No one is allowed to come in here by himself and do something. So there is always a buddy system so that you don't go doing little mods on your own (we say so that you don't electrocute yourself :-) This helps assure that it is documented and save weeks to unravel.

There is kind of an internal squabble in Museum community. You really have 2 major choices of what to do with an artifact

  1. - no touch - don't change a thing, and that model works in an archeological context. Say you have a bone 3 million years old, you don't take out the Pinesol and start cleaning it. I totally agree with that.
  2. - make it work - This was machinery that was made to function. We could do a punch card installation if we had the space. It all still works. The IBM stuff is really well made.
When we get our new building, we may build a raised floor, and restore a main frame computer system

Guest - How many of these 1620's were made?

Over 2,000, and we know of about 20. There is a problem with IBM stuff, originally it was all leased - and when a machine was returned from lease, IBM scraped it (some say dumped into ocean). There is a place in Yorktown, N.Y. that has some IBM equipment - but they don't have one of everything. We basically have machines that were available after the IBM Consent Decree where they were available for sale.

We also like to collect prototypes, like a wire-wrapped prototype Apple-1 board, which we have.

This is the PIXAR image computer that Toy Story was done on. It is a special, they hooked up works to it. It is a high performance, graphics, highly parallel kind of work station.

There are some prime number sieves here that work with no electronic parts. You can actually find a 15 digit prime number with this machine. Does anyone know the Sieve of Eratosthenes? http://www.utm.edu/research/primes/programs/Eratosthenes/

I don't know anything about it so I'm not going to say any more. Its been a long time.

Later-Cray, Cray 1, 2, 3

Seymour Cray - being a loner guy, hated being a manager, he worked at Control Data - the company got so large, in part because of his own success. He left Control Data in 1976, - back to his home town Chippewa Falls, Wisconsin - created his own company which he called Cray Research.

-----------------------------------------------
   er - Jet Lag??
    "In 1972, Seymour Cray founded Cray Research ..."
   http://www.sgi.com/products/seymour.html
---------------------------------------------------

He left Control Data, on good terms - William Norris - CEO of Control Data - even invested in Cray's new company. Two years later, Cray came out with this computer which was called the Cray 1. Fastest computer in the world, as all of Cray's machines were.

Thomas Watson, Jr., head of IBM, wrote a memo that said "How come one guy in the middle of nowhere with 34 employees - including the janitor - can beat IBM with its 300,000 employees and $100 billion annual sales?" And Seymour Cray said "I think Mr. Watson answered his own question." And the reason is that small is beautiful when you are trying to create something of this complexity.


CRAY-1
And when it came out it cost eight to ten million dollars. It was called the world's most expensive love seat. Under the seats are the power supplies.

Just to give you some astounding technical facts

  • it was ECL, bi-polar logic, uses 3 voltages
  • the + 5 volt supply came in at 12,000 amps
  • there were two motor generator sets in the basement of NASA building 33 there. You can still see them there, you build the building around the motor-generator sets. You put the slab down, then the motor generators, then the building.

So what you don't see about this machine is the vast infrastructure underneath it of air conditioning and motor generator sets - which is a motor connected to a generator. It runs on 400 hertz, not 60 hertz to keep the power supplies - the inductors - small.

It became an icon because of it's unusual shape. It had a "C" shape, does anyone know why?

Guest - "C" for Cray?

Not really - it was to keep the wires short. If you fold it out like that, all of a sudden your wires get way longer. But like all good symbols it has multiple meanings - and in fact if you look down on it forms a "C" for Cray. It was all wired by hand, by those same Scandinavian ladies, with the needle point experience. About a hundred miles of wiring, all done by hand, you can see it in here.

Its beautifully done. there are incredibly subtle architectural details to this if you read the manual, which is very readable by the way. The memory is in the outer banks, and the CPU is in the middle, and it is beautifully timed so that all the signals arrive just in time. When you have a central clock source its a real problem to distribute the clock to all points of the circuitry.

Guest - who donated this?

It's from Livermore, serial # 6.

I can show you over here a typical circuit board. Seymour Cray was very conservative at the component level - he did not go for really esoteric technologies until right near the very end of his life.

These ECL chips, there are less than 10 different chip types, and there are about a dozen different board types - That's It. That's really good-

So he had these through hole components on both sides, ECL - they only have 5 or 6 gates in them each, mounted on a copper heat spreader, here, on a copper substrate. They took the heat up and away and through these channels here - and in-between here you have freon pumped through these aluminum channels that are extruded. So super-computers are all about plumbing and packaging. Certainly at this scale.

So his refrigeration engineer, who came from Amana, they make refrigerators right, was his right hand man. With out that guy, these machines would not have existed. Cooling was always the major major problem with these systems.

The Cray-1 cost $100,000 a month to operate:

$50,000 a month in power and air conditioning
$50,000 a month in service contract

You had an office for the Cray Customer Service to read magazines in until the machine broke.


CRAY-2
Have a look at this machine here, the red and black one. This is a Cray-2, it came out in 1985. This costs $19 million (nineteen million dollars). And the neat thing about this was immersion cooled. So he is no longer using freon, he is actually dunking all this circuitry into this fluid - fluorinert - here is a jug of it. It weighs 44 pounds when to is full so it is a very dense fluid. fluorinert is a really weird fluid - it is used in surgery - in operating rooms as a blood plasma substitute. If the surgeon can't type your blood in time and you are going down, he will pump you full of this to keep your blood pressure up because it has very good oxygen carrying capacity. Also it is non-conductive so that makes it pretty unique to - there are not a lot of non-conductive fluids around.

So these boards all sit in this, there is about a 10 degree rise from the bottom to the top so the fluorinert is just circulated up through these pipes here. then it cascades down here forming a really neat looking water fall.

Seymour Cray always had an esthetic component to his machines - he these folks are laying out this kind of serious money, lets make it look good as well. He really said that, he did not just want to make another box that had nothing neat about it.

Guest - the places that bought these things usually made a special room to show them off.

This machine came from Lawrence Livermore, the Magnetic Fusion Energy Lab.

This is the only 8 processor Cray-2 ever made. You could trade off number of processors for memory so there are 128 mega words of memory so you can have more memory and fewer processors.

That points out something about super-computers. They were semi-custom in that you looked at your own problem space and decided - literally down at the level of code and algorithms, where are the loops that are taking up the time, and how can we get hardware to do that. So Cray would work with you to suggest [a configuration].

Guest - Do you know how many they made?

No, I don't

Guest - How did they work on the boards?

When a board blows out, you have to drain all the fluid out. The second part of this waterfall is actually a tank, so you can drain out all the fluorinert, then you can go in there and fix it at the module level.

Have a look at this. This is a Cray-2 board. Nothing you say can do justice to how complex and tightly packed the components are in here. The density of components is incredible.

Does anyone know who is the largest consumer of computers in the world?

The answer is NSA, the "No Such Agency" agency. They were always big customers - cryptography.

Does anyone want to guess what a Cray-2 costs?

The answer is $19,000,000.


CRAY-3
One more thing This is the second to the last thing that Seymour Cray designed - the Cray-3 - and I encourage you all to lift this. This is the CPU and memory. It is a 3/4 million dollar part. All wired by hand 2 nano-second gallium arsenide logic 500 megahertz basically, and the way it is made is unbelievable. All built robotically ... and then immersed in fluorinert.

HyperCubes and other SuperComputers

These machines over here are Hyper-cubes, based on work basically out of Cal Tech and Intel (which just got out of super-computers last year, they donated a whole bunch of stuff to us.) They are all hyper-cube based on commodity microprocessors, the TouchStone Delta, and the Paragon. These were the fastest machines in the world 5 and 7 years ago They had more cabinets - like 5 more TouchStone and 7 more of these -

That is just the end of the road - Intel got out of it. Because to Intel what's 10,000 Pentiums - who cares - that's like 10 minutes of sales. So they were just loosing money and not getting what they wanted out of the super-computer industry.

Again, when you see a company like Intel get out of super-computing you can see that it extremely risky business - most people just don't hack it.

There is no money in hardware, Control Data just does software, Thinking Machines just does software, like data mining The unit sales are so low there is no money in it, even though the prices are high.

There is a good example - this machine here is called an ETA-10 - it was the only liquid nitrogen cooled CPU. This is the liquid nitrogen vessel here.

When Cray left Control Data, Control Data completely freaked out because he was the key man in the whole company. So they spun off a bunch of engineers, who used to work for Cray, into ETA Systems and tried to make a go of this new technology and basically the reason they failed was more of a marketing problem. They did not see that UNIX was going to catch on - so they did not offer it. All it takes when you haven't sold any product, is for one or two key customers to say "NO" and you are like dead in the water. So that's what happened to these guys. They went through hundreds of millions of dollars burned trough it.

RAID

Behind you here, just to jump in time, is a RAID prototype, done by Patterson and Katz at Berkeley. This is their second prototype - this is the origin of the RAID systems out.

I'm doing a paper with them on the history of RAID. RAID is a concept that has been around for a long time, but they established this taxonomy - basically what is RAID 0, 1, 2, 3, 4, 5, what does that mean, and then manufacturers totally lost it and kind of inventing their own like "RAID 72" , sort of confusing the issue.

This is it, you just build a huge storage array based on commodity disk drives that trade off speed or reliability, depending on how you set it up. Its all done by grad students, Its really well made. This cost about $200,000 to make, 10,000 hours of student labor, and is now a $10 billion/year industry because of this prototype. And the professors made no money out of it. Poor guys.

Let me show you one thing

Xerox PARC, Altos

These are Xerox Altos, 1972, done at Xerox PARC, This is one of the strong candidates for a machine we want to get running, a couple of them actually, because there were network games that you could play on them.

So, in 1972 it had a bit mapped display, page white, a desk top metaphor, windows based operating system (windows without the "R" in the circle [copyright]), pull down menus, mouse driven, word processor, e-mail, hard drive, Ethernet (there is an Ethernet cable hanging from the ceiling there - thick wire coax) This was in 1972 and the thing underneath the cable is the world's first laser printer - it looks like a photocopier - You can open that and see the laser engine - its really neat - we had Gary Starkweather talk about him inventing that. Here is the story, he couldn't get funding for a software guy for this project - the software guys were in another building across the highway. So he went to Edmond Scientific and bought 2 lasers, and connected the two building by laser. And he got the software guys in the other building to send the test code over the laser - over the high way to Gary's machine. He totally undermined the management. They did not want to give him room, did not want to bring programmers over. - there is a transcript is on our web site - He got the software guys in the other building interested and got them anyway.

And this is what the legend goes about what Steve Jobs looked at bla-bla-bla - Not totally true, there were people from PARC already working at Apple who swear they spent a year trying to get Jobs discover the Alto. Which is probably closer to the truth.

The start of a lot of interesting things. Doug Englebart who invented the mouse, but doesn't want to be known for just that, came up with a number of interesting things - you have a knee mouse, you clamp it on your knee and move it around like that - that he invented. This is a chord set - it is called - you have 5 switches so you have 32 combinations and you can actually enter letters. His idea was you get rid of the keyboard, and you just use this and the mouse to type.

And Larry Tessler can get about 30 words a minute on that which is pretty good. But if you know the debates over the Dvorak keyboard - like if that's not going to make it - there is no way this will . [Much laughter from the guests]

ILLIAC-IV

Has anyone heard of the ILLIAC-4? It was a massively parallel super-computer that was kind of a white elephant. It was supposed to cost $16,000,000, it eventually cost $31,000,000 and was a quarter of the size the it was supposed to be. It was supposed to have 256 processing elements, it ended up with 64. You know the government had to be mixed up in there somewhere.

This is a hard disk from a computer called the ILLIAC-4. I invite you all to lift if you like, its pretty beefy. And these counter weights here help balance it, like the weights on your car tires to help keep is from wobbling and killing someone. This is the drive enclosures here - there are three here. I found two of these in a NASA building in 1999. No body knew what they were. They were in the basement beside these old motor-generator sets that don't do anything anymore, but can't be moved.

Does anyone want to guess how much this hard disks holds?

This disk, which is about 3 feet in diameter, contains 10 megabytes.

Never the less, it had some pretty neat features. These hard drives are interesting. See these are the recording heads, it has 64 of them, 64 fixed heads, they don't move. There is no head seeking. The through put of this is 800,000,000 bits a second, which is the Ultra-2 SCSI rate today. So you can sort of laugh at it, but you have to tip your hat off to these guys because this is 1974.

This machine started life at the University of Illinois, but after some anti-war riots and damage to other computers, it was moved to NASA AMES - in a building within easy walking distance of here. NASA built an annex to a building to house this computer. The annex is still there, the machine is gone, the motor-generators are still there.

It came on line in 1974, and it never really worked well. They had a lot of technical problems. For example, these boards were 10 or 12 layer boards, in 1970. It must have been agonizing to work on this because they had so many delays - and by the time they would come to another mile stone, the circuitry they had been working on was totally obsolete, but they had sunk so much money into it that they couldn't go back and make it modern.

Yes - this is an extender board that you can use to help trouble shoot boards

And this was the first machine to use semi-conductor memory. You can see it right there - "Fairchild Bipolar Memory". Its a 2 K word memory, 64 bit word length.


If you have comments or suggestions, Send e-mail to Ed Thelen

Go to Antique Computer home page
Return to top

Updated March 25, 2006