First Minis header3

Rev. 1.6

If you go looking for the first minicomputer, you can easily find it—several of them in fact! Join me as we dig into the facts to find out which minis were actually first.

I was reading Murray’s book, “The Supermen” (the story of Seymour Cray and supercomputers)33 and decided to lookup the CDC-1604 mentioned there. That included a mention of the CDC-160 minicomputer (hereafter “mini”). The CDC-160 was built as an afterthought, to handle input/output tasks for the 1604. Following that trail, I found a site that suggested that it was the first minicomputer, which surprised me. Reading the Wikipedia article on minis, I was even more surprised to find a statement that the USSR created the first mini, the UM-1NKh, in 1963!39 [Since changed by Tom94022.] Well, now I just had to find out for sure, who DID build the first minicomputers!

However, we should recognize that some scholars feel that trying to determine such a thing as the first mini, is an unproductive exercise.35 There is a lot to be said for that view: Such a study is heavily dependent on how you define minicomputer (in this case). Also, history shows that such complex inventions are rarely created suddenly. Typically, they are part of a broad progression of ideas and are built incrementally with many steps and many contributors. Trying to determine exactly when a thing arose and who created it is often nebulous, uncertain and frustrating.

—Well, yes. But what do you say to somebody who asks the simple question? I was that somebody and I wanted an answer. I was prepared for dealing with definitions, nebulosity and uncertainty. (All in a day’s work for an electronics engineer :)

What Do You Mean by Minicomputer?

Back in the day, (the early seventies) it seemed pretty clear what a mini was—there was a PDP-8/e sitting on a table in the lab. Of course, by the time you had a nice system, with a couple RK05 disk drives and DECTape drives, it took up two racks. Then there was the rack for the FPP-12 floating point processor add-on. It was still a mini, though!

In sorting through the first-mini candidates, it quickly became apparent that drawing the line between which computers were minis and which weren’t would be a challenging part of this task. Perhaps we should begin by naming particular examples which we do or do not speak of as minis. Consider these early, small computers:

  • DEC PDP-8 pictured above is taken by many as the quintessential mini, so surely it should be part of our paradigm. Shipped in 1965, its combination of decent performance and $18K price tag made it just right for many lab and industrial applications.
  • Data General Nova at right was DG’s first product in 1969. This successful 16-bit mini challenged DEC to introduce the PDP-11 line.
  • DEC PDP-1 below right was DEC’s first computer. It occupied four racks, weighed 1350-lbs, used 2160VA of power, cost $120K in 1960 and had an 18-bit word. It seems pretty tough to consider this as a mini, even though it was very important in the evolution of Digital Equipment Corp.
dg-nova 350
PDP-1 350
  • SDS-910 at left was produced by Scientific Data Systems in 1962. Occupying only one rack, it cost $41K,  weighed in at about 900-lbs and had a 24-bit word. The PDP-8 occupied about half a rack when packaged that way and each ad­di­tional 4K of memory required another 10-inches of rack space. So occupying a single rack doesn’t seem unreasonable for a mini.

Notions of a Mini

The lexical definitions for a minicomputer that we find in dictionaries are vague and not very useful for our task here. In a 1970 survey, a prominent news organization suggested a consensus definition of a minicomputer as a machine costing less than $25K, with an input-output device such as a teleprinter, with at least 4K words of memory and that is capable of running programs in a higher level language, such as Fortran or Basic. The suggestion of a $25K price tag doesn’t work well for the early minis, which were necessarily more expensive. Besides, even a PDP-8/e, with drives, memory and peripherals could easily have exceeded that figure. Fact was, it took $50-100K to get a nice mini, before say, 1965. In this study, we were hoping to avoid such arbitrary and absolute critieria. Rather, it seemed to me that we could examine the myriad of machines that existed and find well-fitting classifications for them. From that, the answer could emerge.

Before we get into the horse trading of which minis had the best claim on being first, we need to set a few ground rules for computers we will consider:

SDS-910 200 nb

No guidance compsGeneral Purpose Computer
It must have been a general purpose electronic computer. By that, we mean that it has RAM for storing programs and data, and interfaces to human-readable input and output devices. There is also some expectation of a minimum amount of capacity and speed. Generally, if it could run a high level language, that would be sufficient. Notice however, that we have already excluded a broad swath of military and control computers, which might not have been designed with standard interfaces and might never have been intended to run something like a compiler. Examples would be industrial controllers and ICBM missile guidance computers (right). While those might have been substantial small computers with important missions, they simply aren’t what is generally meant by a classic minicomputer.

Why is this distinction important? The early minis were often used interactively. They introduced people to the personal computing experience for the very first time. Back in the day, I remember fantasizing about what it would be like, if I could have one of my very own! As a result, the mini paved the way for the personal computer revolution. If it hadn’t had the human interface or hadn’t been able to run a high level language, there would have been no such feeling. As an aside, one might ask whether timeshare terminals could have provided this same experience. Perhaps in principle they could but not generally in practice. In the corporate environment, third-party timeshare was quite expensive. At my first engineering job in 1975, we were required to punch our programs off-line on paper tape, so they could be rapidly uploaded, minimizing online time. No way could you play around or even think about editing documents. With a mini, you were much more likely to have fun. Often, you were given your own reel of DECTape on which to keep your files. Man, that gave you your own private space—your own stake in the computer!

A Real Product
PDP8 assembly lineIt must have been manufactured in quantity. One might very well be interested in finding the very first laboratory prototype of a small computer but as a practical matter, it would be difficult to prove exactly when such machines were finished and that you had found the very first one. There were simply too many secret military research programs and other R&D operations which we may or may not be privy to. Moreover, I submit that it is only the machines which were made in quantity that could be considered available to small organizations. A one-off lab curiosity might be an interesting historical footnote but it doesn’t deserve to unseat something like the PDP-8 from being the first successful mini. (Just taking a hypothetical example!)

Real Ship Date
For priority

Defining Cost, Size and Weight

For our comparisons we will define cost in terms of a minimal proc­es­sor con­fig­uration. Naturally, a practical system would typically cost much more. Un­for­tunately, the obtainable minimum proc­es­sor prices often include a Teletype or a paper tape reader/­punch. That shouldn’t make a great deal of difference for our purposes, though. Similarly, size and weight are for the minimal proc­es­sor. In some cases, that includes such input/output devices if they were integrated.

date, we will consider only the date that the first production unit was shipped. No vaporware need apply. Mar­ket­ing generated “introduction” or “announcement” dates don’t count. Dates that a prototype first worked in the lab are interesting but that certainly wasn’t a machine which was available to small organizations.

Criteria to Narrow the Field

So what is meant by the term, “minicomputer”? First, it’s important to recognize that this is not about when and how that term came into existence. Like all words, it is defined by the people who use it and definitions evolve over time. We do in fact use the term and others seem to know what we mean. Our task is to clarify what people mean by it today. After that, we can search for the first machines which meet the definition.

Perhaps the most important thing about the mini is that it was simply available to small organizations. For it to be available, it had to differ from the classic mainframes in cost, size, weight, and power requirements. The biggest problem with trying to find the first minicomputer is that there was a continuous stream of smaller computers being delivered in the critical years of say, 1959 to 1965. These machines spanned a wide range of prices, sizes, markets and applications. Depending on what criteria you choose to define a minicomputer, you get different answers as to which was first.

LINC computer2 350This explains in part, why different sources give different answers for the first minicomputer. Some of the nominees that I’ve seen for this or similar designations include: PDP-8, PDP-1, CDC-160, Soviet UM-1NKh, IBM-1130, IBM-1401, and LINC (shown at left). The LINC is unique, in that it came out of the Whirlwind project at the MIT Lincoln Lab. About 50-units were made, all built from standard DEC logic modules. Some were built at MIT and later, DEC built about 20.

Before we get close to looking for winners, we need to narrow the field of smaller computers to develop a list of potential candidates. Let’s look at some of the criteria and see if we can establish guidelines for that. Keep in mind that these guidelines are broadly chosen at this point and may include some machines which we may later decide are not minicomputers. The criteria we are after therefore, are necessary but perhaps not sufficient conditions to be considered a mini.

Cost
The news organization definition of mini mentioned above used price as the primary criterion. Certainly price is a critical parameter in determining how available a product is to a small office. No doubt, the figure picked bears some relation to the original $18K price of the PDP-8. That machine is considered by many to be the quintessential minicomputer, if not the first. However, what we find is that in the years leading up to the 1965 introduction of the PDP-8, the cost of small computers was dropping rapidly. Setting a price threshold would be a lot like setting an arbitrary cutoff date.

While the price point of the PDP-8 was surely important to the market, should we draw an arbitrary line just before it? What about a machine which was very much like the PDP-8 but cost $27K? (That’s the PDP-5.) Then there’s one at $41K (SDS-910). Where do you cut it off? Instead of drawing a hard line, perhaps we should consider price along with other parameters in making a judgment on a case-by-case basis, whether a machine belongs in the minicomputer class. As far as screening candidates, the most expensive machine fitting the other criteria turned out to cost $120K. That was still affordable to a large number of organizations and it isn’t out of line with the next most expensive candidate.

Size and Weight
Some people might feel that a minicomputer should be something that could fit on a desktop. The PDP-8 could sit on a table but at 250-lbs, it had to be a pretty sturdy one. As mentioned above, the machine was also sold in rack form. If you added memory, tape drives and disk drives, it could easily occupy multiple racks. Some small computers were built into a desk-shaped form factor. Aren’t those effectively small enough? After all, one could argue that they’re actually smaller than a PDP-8 sitting on a desk!

But keeping our eye on ball, we’re trying to define “minicomputer” here, so perhaps when the CPU itself requires two or more racks, it’s getting outside the reservation. The thing about the minicomputer was that, you could put it in a single office. Though it might not have been typical, that let every engineer and scientist fantasize that perhaps someday, he could have one in his office! Let’s leave size, like price, somewhat flexible but bear in mind that it should fit in an office, in principle. I suppose that would mean up to one rack and perhaps a TTY or tabletop console. Along those lines, it would have to be light enough that it wouldn’t need special floor support, could be transported in a standard elevator and could be installed without a great deal of fuss. That would mean that it could weigh no more than 1000-2000-lbs.36,37

Power Requirements
In a 1965 ad, DEC was able to say that you could just wheel the PDP-8 into your office and plug it into the wall. Before that time, people thought of computers as needing lots of power and extensive electrical support for installation, so that was a big deal. Running on 115VAC would limit power to about 2300VA. If we allow 220V operation, it could be about 4400VA. Special air conditioning arrangements might be needed at the high end of that range though. Certainly, we don’t associate special air conditioning with minicomputers. So let’s just say that it must run on 115V or 220V and machines requiring more than 2300VA get demerits. Three-phase power machines are out.

LGP-30 350

Copyright Computermuseum Stuttgart

What about Vacuum Tube and Drum Machines?

This is kind of a tough one. Trouble is, there were some prominent, very cleverly designed small computers in the mid-to-late 50’s, which used drums instead of core memory. They were very slow but could execute a stored program. The slow drum dovetailed into a a simple bit-serial CPU, allowing a relatively low-cost vacuum tube computer to be built. With no minimum speed requirement, it would be hard to exclude some of these machines. Indeed, the Librascope LGP-30 (right) was just 21-cuft, 800-lbs, 1500VA. It stored 4K x 32b. Its electronics used 113-tubes and 1450-diodes. With that, it executed a “high level” language called ACT-III.

LGP-30 screen 350It would seem difficult to exclude this masterfully clever design from our competition, on the surface of it. On the other hand, there were just 16-instructions in its machine language. A multiply took 17ms. Instructions required 2.2ms. Perhaps worst of all, it could take 15ms to access a location on the drum, due to its rotation. This meant that programmers of drum machines had to resort to complex schemes to interleave data and instructions, to minimize drum latency. I suspect that had a chilling effect on writing sophisticated software, as programming effort was sapped in dealing with machine foibles. Watching a remarkable video of the LGP-30 in operation, I was fascinated but a little taken aback. At left, we see that instead of lights or numbers to indicate register bits, it had a built-in oscilloscope screen. Individual bit values are read by position along the marked graticule.  Brazil movie computer 200The video reminded me of the steampunk-esque technology depicted in the movie, Brazil (right :)  It was noisy, slow and limited but it was a small, affordable, stored program computer in 1956. ACT-III seems to be a rather arcane language. Bendix G-15 150

The Bendix G-15 at left was another successful vacuum tube, serial drum computer, introduced about the same time. Having a power consumption of 3.5kVA, it needed either a large room or special air conditioning. Though it had only 2176-words of 29-bit memory, amazingly, a compiler supporting a variant of ALGOL was developed for it.

With the tubes and all, reliability was not what we came to expect of minicomputers. The MTBF of the G-15 was reported as just 100-hours.14 Finally, the fact is, the drum machines don’t have an electronic main memory—it’s electromechanical. Since the processor isn’t electronic, it isn’t an an electronic computer as required in our definition of mini. Admittedly, the mass storage devices used with minis are electromechanical but they aren’t part of the processor itself. Rather than try to include these weird beasts in the competition, I suggest that the small vacuum tube and drum machines belong in their own special category. Indeed, it should be a place of honor!

Packard Bell 250 150Magnetostrictive Delay Line Computers

Another cyclic, early, main-memory technology was based on acoustic delay lines. Magnetostrictive transducers were used at either end of a coiled wire to transmit torsional waves. Bits transmitted from one end were delayed before being received at the other end. By regenerating and recirculating the bits, they could be stored indefinitely. The Packard Bell 250 at right is an interesting example of a computer based on this technology. Like the drum machines, it coupled a serial CPU with the serial memory, thereby greatly reducing the logic hardware, without much additional performance impact. It was a table-top solid-state model which occupied only 7cuft, weighed only 110lbs and used 110VA of power.

The 3000us delay line allowed storage of about 16000, 22-bit words. However, this meant that the average random access time was about 1500us. Like the drum machines, special programming techniques could reduce the effective access times greatly by arranging data and instructions in memory. Also, there was a small amount of faster (96us) memory, which could be used as a “scratch pad” to speed things up. Of course, that too required appropriate programming techniques. While these machines were a big improvement over their drum-based cousins, they were still burdened by very slow main memory and required the programmer to get deeply involved in tricks to minimize its impact. As a result, these belong in a special category. Since both these and the drum machines share cyclic memory and its foibles, it seems appropriate to lump them together, separate from the minis.

A Table of Potential Candidates for the First General Purpose Minicomputer

First, let’s summarize how we are defining “First General Purpose Minicomputer,” in this study. The criteria are:

  • It had to be an general purpose electronic computer with enough speed and RAM to run a high level language. It would also have an interface for human interaction and program loading, such as a serial port to a Teletype.
  • Manufactured in quantity.
  • Weighs no more than 2000-lbs.
  • CPU occupies no more than one rack or a desk-sized package.
  • Must run on 115V or 220V (single phase) and machines requiring more than 2300VA get demerits.
  • Priority is set by the date of the first production shipment.

Notice that these points are a far cry from the New York Times definition of a mini as a computer costing less than $25K (and a couple of other things). In view of the rich hoard of interesting early small computers that turned up, it looks like our criteria are preferable. I certainly don’t mean to suggest that our definition is better than all others, though. Others might reasonably choose different criteria. The main thing is that in any meaningful search for the first mini, it is critical that the criteria be clearly stated. Comments are welcome and we intend to revise this article as new information comes to light. Searching for machines meeting the criteria and doing research to find the key data resulted in the table below:

Maker

Model

Price

Date

Volume

Weight

Package

AC Power

Word Leng

RAM

Mem

Mem Cyc

CDC

160

$60K

Jun-60

~39 cuft

~800 lb

desk

1380 VA

12-bits

4K wds

core

6.4 us

DEC

PDP-1

$120K

Nov-60

94

1350

4-racks

2160

18

4-64K

core

5

IBM

1620

$64K

~Dec-60

~79

1210

desk+console

2000

12-72
(var)

10-30K (12b)

core

20

CDC

160A

$90K

Jul-61

39

810

desk

1840

12

8-32K

core

6.4

MIT/DEC

LINC

$44K

Mar-62

~49

~1070

rack+console

1000

12

1-2K

core

8

DEC

PDP-4

$65K

Jul-62

47

1030

2-racks

1125

18

1-32K

core

8

SDS

910

$41K

Aug-62

28

900

1-rack

1870

24

2-16K

core

8

DEC

PDP-5

$27K

Sep-63

24

540

1-rack

780

12

4-32K

core

6

USSR

UM1-NKh

$56K

~Dec-63

6?

180?

desktop?

100? 200?

13? 15? 20?

256-4K

ferrite plate

100?

DEC

PDP-8

$18K

Apr-65

8

250

desktop

780

12

4-32K

core

1.6

IBM

1130

$32K

Dec-65

44

760

desk

1100

16

4-32K

core

2.2-3.6


Comments on the Table

Bear in mind that these are just potential candidates, so we tried to be liberal in bending the criteria for possible entries, especially for machines proposed by others as some of the first minis. The PDP-1 has been called a mini, so it was included, if only to give explicit reasons for excluding it later.38 The table is in order of ship date. Please refer to the notes for caveats, as some dates are not precise to the month. No machines meeting our criteria were found prior to June, 1960. The CDC-160A was initially included in case the CDC-160 became excluded. For example, we didn’t know at first if the CDC-160 was in fact used as a standalone mini or whether it was solely used to process I/O for the CDC-1604. The literature seems to indicate that it did see standalone service, perhaps most easily supported by the fact that it was marketed by NCR as the NCR-310. A couple machines from SDS were left out because the 910 was obviously a stronger candidate.

UM1-NKh computer2 250Spy Thriller
We struggled to include the USSR machine (left), since it was formerly claimed as the first mini in a Wikipedia article. (No definition of mini was given with the claim and it didn’t fit the NYT definition quoted in the article.)39 Unfortunately, precious little data was to be found and what was available was inconsistent (hence the question marks and multiple choices). The search led me to read Usdin’s excellent book about early Soviet technology development called, Engineering Communism. 6 The first part reads like a spy thriller! By the way, the UM1-NKh appears to have been original work, though the principals involved were well aware of developments in the West. They were in fact, Americans who formerly spied for the Soviet Union and ended up spending their lives there after fleeing from the FBI. Fascinating story! They certainly produced a remarkable machine, though it ultimately didn’t fare well in this particular competition, as we shall see.

Coming up next:

Part II - Making the Cuts, And the Winners Are... >>>

 

Reader Comments


Posted by Steve L. January 24, 2014 - 04:01 pm
Hi Edfair, Yes indeed, the IBM 1401 was available in a configuration with just the left-half of the cabinet pictured in Part-II of this article. As you note, it had 1.4Kbytes of memory. And as you say, it still would not have qualified. However, that's an excellent suggestion, to include mention of this smaller configuration in the honorable mention. I will do so. Thank you for mentioning the credit card billing operation. That's an excellent illustration of how much they were able to do with such modest computers!

Posted by edfair January 24, 2014 - 11:54 am
You could have noted the baby 1401, half the size of the one pictured. Only saw one, Hess Oil, Atlanta, running credit card billing. 1.4 K IIRC, with 1402 Reader/Punch and 1403 printer. Still wouldn't have qualified.

Posted by Steve L. October 22, 2014 - 08:21 am
Dear Computermuseum Stuttgart, We appreciate the permission previously received to publish the photo and have added the copyright notice to the photo from there. Please note that copyrights, attribution and links are also provided for in the Acknowledgements section of this article. (The source of the LGP-30 scope screen is covered there.) Please contact me directly if there is anything else that I can help with. Warm Regards, Steve [PS: You might need to refresh your browser pages to see the changes.]

Posted by Computermuseum Stuttgart January 24, 2014 - 04:48 am
Please add copyright notices to our LGP-30 photos! These photos are our property.

Posted by Steve L. January 23, 2014 - 08:27 pm
Please feel free to add any comments or questions you might have. No sign-up is needed.

Add your comments here...


Name:


Message:

_