The DEC PDP-8 StoryPart II: the minicomputer revolution
by
Douglas W. Jones
|
The DEC PDP-8 StoryPart II: the minicomputer revolution
by
Douglas W. Jones
|
Originally published in Historically Brewed, 8. This copy is an HTML version of the text file I submitted for publication.
In the first part of this series, the focus was on the technological innovations that allowed Digital Equipment Corporation to introduce the revolutionary PDP-8 family of computers. In this installment, the focus is on the competition and how DEC responded to maintain their leadership in the turbulent new minicomputer market of the late 1960's.
When the first PDP-8 was delivered, in 1965, Digital Equipment Corporation was not alone in the small computer market. Control Data Corporation had originally opened the market with its CDC 160, and two other young companies were ready to meet the challenge posed by the PDP-8.
Computer Control Corporation of Framingham Mass., known by their trademark, 3C, was a successful maker of medium sized computers with a 24 bit word. Within 14 months of DEC's announcement of the PDP-5, in October 1964, 3C announced a far larger machine, the DDP 116. This was the first 16 bit minicomputer, and like the PDP-5, it was available packaged in a single 6 foot high EIA standard 19 inch mounting rack; with a list price of $28,500, it cost only $1,500 more than the 12 bit PDP-5.
The minimal DDP-116 configuration had 4K 16 bit words of main memory, and like the PDP-8, this could be expanded to 32K words. The DDP-116 was a single accumulator machine, but unlike the -8, it had an index register. Furthermore, the 16 bit word was large enough that, unlike the 12 bit PDP-8, no difficult to use segmenting scheme was required to address a full 32K words of memory.
Another established maker of 24 bit machines, Scientific Data Systems, also recognized the potential market for small machines. The SDS 92, announced in the summer of 1964, was a direct competitor with the PDP-5, with 12 bits per word, a minimum of 4K words, and a maximum of 32K words.
The primary innovation of the SDS 92 was not architectural, but electronic. The SDS 92 was the first commercially available computer to use monolithic integrated circuits instead of the discrete transistors used by most manufacturers or the hybrid integrated circuits used by DEC and IBM.
Neither machine was as fast as the PDP-8, confirming the fact that the PDP-8, was, for its time, about as fast as the technology allowed without resorting to parallel architectures. Where the PDP-8 had a 1.5 microsecond memory cycle, the DDP-116 cycle was 1.7 microseconds, and the SDS 92 was 1.75 microseconds.
In May of 1965, Systems Engineering Laboratories jumped into the market with their SEL 816, a machine with a 16 bit word, available with between 4K and 32K words of 1.75 microsecond core memory. This machine was packaged in a desk-sized unit and priced at about $26,500.
In early 1966, within a year of DEC's announcement of the PDP-8, Control Data Corporation responded to the growing competition by announcing the CDC 1700. This was a replacement for the CDC 160 priced at under $30,000 and incorporating the same logic modules and packaging technology as was used in the CDC 6600 supercomputer.
All of these machines were priced to compete with the PDP-5, but the PDP-8 undercut all of them by a wide margin while outperforming them by many measures. Furthermore, neither SDS nor 3C was able to deliver until May of 1965, and CDC was even later with its machine. By the end of 1966, the figures for total machines installed confirmed DEC's market leadership.
| May 65 | July 66 | Dec 66 | |||||||
|---|---|---|---|---|---|---|---|---|---|
| SDS 92 | 25 | 42 | 64 | ||||||
| DDP 116 | 30 | 55 | 130 | ||||||
| SEL 816 | 0 | 19 | 28 | ||||||
| CDC 1700 | 0 | 0 | 30 | ||||||
| PDP 8 | 75 | 250 | 490 | ||||||
Of the competing machines, the 3C DDP 116 was clearly the strongest, clearly demonstrating the value of 16 bit architectures. The strong sales figures for this machine appear to be the reason Honeywell bought 3C in 1966. SEL, although starting out at a decided disadvantage, managed to continue growing through the 1960's, further confirming the value of using a 16 bit word.
In contrast, the production of the SDS 92 never exceeded a few machines per month, and by 1966, the lesson was clear to the people at SDS. In that year, SDS released a 16 bit minicomputer, the SDS Sigma 2.
Formally speaking, the smaller IBM System 360 systems could also be classed as 16 bit minicomputers, but the smallest of these, the IBM 360/20, was ten times the price of comparable minicomputers. As a result, IBM was never a significant player in the minicomputer market of the 1960's.
From the start, DEC recognized that software was essential to selling computers. FORTRAN II and a macro assembly language were both available on the PDP-5, ready to go to market with the PDP-8 along with a suite of utilities such as an on-line paper-tape based text editor and a symbolic debugger. Furthermore, the DEC User's Society, DECUS, maintained a rapidly growing program library that already included 16 PDP-5 programs by the time the PDP-8 was released.
All of this software was provided in paper-tape form for those with small PDP-8 systems; for larger systems with DECtape drives, DEC offered the DECtape library system, a rudimentary interactive operating system with a command language interpreter that could be used to load and run programs from DECtape.
At the time, few computer users expected to program computers using on-line development tools. Users of IBM equipment expected to use punched cards for program development, while users of most smaller computers universally expected to use punched paper tape. Seen in this context, the DECtape library system, along with the suite of software tools DEC provided was actually somewhat ahead of the market.
The first wave of competition for the DEC PDP-8 was priced to compete with the PDP-5, but the second wave was right on target.
In early 1966, Scientific Control Systems announced the SCS 650, priced at $14,800. With with 4K 12 bit words of memory, expandable to 32K. Little information is available about this machine; it may have been a clone of the PDP-8, although its 2 microsecond memory cycle made it significantly slower.
Honeywell announced the Honeywell 20, with 16 bits per word, and in 1966, Data Machines announced the Data 620, another 16 bit machine, priced at $13,900 with 4K of memory. The latter was a rack mounted machine, only about 15 inches high, and it was clearly a significant threat to the PDP-8. 35 of these were sold by January 1967, by which time Data Machines had been acquired by Varian.
Hewlett Packard was initially a large user of DEC equipment, but by 1967, they recognized that they would be better off building their own machines, and in September, they entered the fray with full page ads proclaiming that "Hewlett Packard is now in the Computer Business." By November, 56 HP2116A machines had been sold, these were 16 bit dual accumulator minicomputers, and within a year, they would be followed by the compatible HP2114A selling for only $9,950.
It would have been natural for DEC to counter with a 16 bit minicomputer, and some people within DEC began to argue for this. By 1966, the sales figures for the DDP 116 clearly demonstrated that there was a large and growing market for 16 bit machines, but DEC avoided this move because it would have undercut their successful line of medium sized 18 bit machines.
DEC eventually did bring out a 16 bit machine in 1970, the immensely successful PDP-11. Until then, they concentrated on enlarging the PDP-8 market by introducing ever less expensive models. The effect of this strategy was the rapid opening of an immense market for small computers.
The first DEC machine along the lines of this strategy was the PDP-8/S. This was the first general purpose computer to sell for $10,000, including the cost of the console teletype and paper-tape based software, and its introduction at this price on August 23, 1966 at WESCON had an immediate impact.
With the exception of the core memory, this machine was based on exactly the same hardware technology as the original PDP-8. The same semi-automated wire-wrap technology was used to produce the backplane and the same discrete-transistor ``flip chip'' modules carried out the logic, yet it was only 28 inches deep, 19 inches wide, and 10 inches high, and it weighed only 75 pounds (1/4 the weight of the original PDP-8).
The key to the economy of this machine was that most processing and data transfers were done in bit-serial fashion between shift registers. There was quite a performance penalty for this! The add operation that took 3 microseconds on a PDP-8 took 36 microseconds on a PDP-8/S, and the increment operation that took 1.5 microseconds on the PDP-8 took 28 microseconds on a PDP-8/S.
Despite the incredibly low speed of this machine, it was an immediate success in the marketplace. in just five months, over 400 machines were sold, and over 400 more were sold in the next year before DEC undercut the PDP-8/S with a faster and less expensive model. By the end of the production run, 1024 PDP-8/S machines were built.
In early 1967, the sales of the PDP-8/S were so strong that DEC was moved to break new ground in the computer marketplace by offering the PDP-8/S for retail sale on a cash-and-carry basis, available at every DEC field office in the United States. This was the first computer ever sold this way, complete with advertising suggesting that you drive away with a complete working computer system in the trunk of your car. As a result, some argue that the PDP-8/S was the first personal computer.
The PDP-8/S was not the only serial machine on the market. For example, Interdata took this approach to produce an ultra-low cost 16 bit minicomputer. In early 1967, less than a year after the PDP-8/S came on the market, they offered a machine for $7000 that took 45 microseconds for a 16 bit add instruction, and by the end of 1967, they had sold over 24 of these machines!
The big step that allowed Interdata to undercut DEC's price was the use of TTL integrated circuits, and the same move allowed them to advertise a parallel version of the same machine, the model 4, for $10,000 in December 1967.
The term minicomputer was apparently coined by the head of DEC's English operations, John Leng. He wrote a sales report that began something like "Here is the latest on minicomputer activity from the land of miniskirts and the Austin mini". Although the term quickly became part of DEC's technical jargon, the first widely read public use of the term appears to have been in an ad for the Interdata 4 in May of 1968.
DEC began selling M-Series logic modules, the TTL version of their Flip Chip modules, in 1967, and by December 1968, they had announced the PDP-8/I, a TTL implementation of the PDP-8. As with the original PDP-8, this machine was made using semi-automatic wire-wrapped backplanes, but the use of integrated circuits allowed approximately twice the logic to be packed onto each circuit card.
The result of this technical advance was a backplane almost as small as that of the PDP-8/S but able to operate at the same speed as the original PDP-8. Furthermore, there was room on the backplane for 8K of memory and a host of slots on the backplane came prewired for popular peripheral interfaces and CPU options.
Unfortunately, DEC had problems obtaining core memories that would operate reliably at the intended speed, so early PDP-8/I systems were run at a lower speed to compensate for this. As a result, the classic PDP-8 remained in production longer than might otherwise have been expected in order to support customers who demanded full speed operation.
The PDP-8/I was offered in two configurations, a sleek table-top model with a separate power supply that stood nearby as a floor mounted pedestal, and a rack mounted version. Few table-top models appear to have been made, but by the end of 1968, over 400 PDP-8/I systems had been delivered; by the end of the production run, over 3500 were in place.
In parallel with the PDP-8/I, DEC developed another version of the PDP-8, the 8/L; this machine was designed using exactly the same technology as the PDP-8/I, but by leaving out various options, the price of the machine was cut and the size was reduced to a single 10 inch high rack mountable box almost exactly the same size of the PDP-8/S.
As an example of options that were eliminated, the PDP-8/L did not support the optional extended arithmetic units offered with the PDP/8 and the PDP-8/I, nor was there space in the basic model to expand the memory beyond 4K. DEC did offer an optional memory expansion box that held an additional 4K of memory, along with prewired I/O slots for devices that were standard on the PDP-8/L, but they resisted allowing the PDP-8/L to be expanded to the full 32K supported by other PDP-8 models.
The PDP-8/L sold for $8,500 with 4K of memory and a teletype, and the memory cycle time was 1.6 microseconds (only 100 nanoseconds slower than the original PDP-8). When it was announced in the the summer of 1968, new orders for the PDP-8/S virtually came to a halt, and demand was so great that DEC delivered over 1000 by the end of 1969. Overall, over 3000 PDP-8/L systems were built.
In March 1968, the Israeli ELBIT 100 was announced at a price of $5000, with discounts for bulk purchases. This may have been a PDP-8 clone, but little information is available; it had a 12 bit word, 1K words of 2 microsecond core, and 256 words of ROM, a single accumulator and a carry bit. The 4K configuration cost $7000, and by the end of 1968, 35 machines had been delivered.
This machine illustrates the extent to which integrated circuit technology had simplified entry into the small computer marketplace. Another illustrations of this is the General Automation SPC-8, an 8 bit processor announced in October of 1968, and the MDP-1000, another 8 bit processor introduced in late 1968 by Motorola.
Ed deCastro, a long-time member of the PDP-8 design team, was one of the people within DEC who wanted the company to enter the 16 bit market. He and others within DEC produced a design known as the PDP-X.
DEC chose not to sell the PDP-X, and in November, 1968, Ed deCastro's new company, Data General, announced their first machine, the Nova. This 16 bit minicomputer became a serious contender in the minicomputer marketplace, and many people view it as the one of the first true RISC architectures.
On the software front, DEC was clearly far in front of the competition. While most minicomputers were offered as bare-bones systems, both DEC and the PDP-8 user community had been busily building a base of generally usable applications.
While most people at the time still viewed timesharing systems and interactive programming languages such as BASIC as being the province of huge mainframes, enterprising programmers discovered that the rudimentary memory management hardware of the PDP-8 was sufficient to protect an operating system from multiple users, and the result, in 1968, was the TSS-8 operating system for the PDP-8/I. TSS-8 could support as many as 32 users, and users were free to program in assembly code or FORTRAN.
At about the same time, Richard Merrill developed the FOCAL language, a small interpreted programming language very similar to early versions of BASIC. This was able to run on a PDP-8 with only 4K of memory, and it became a popular language, both on large TSS-8 systems and on small stand-alone machines.
The combination of low cost hardware, FOCAL and timesharing allowed DEC to enter the educational computing market, selling complete computer systems to a significant number of high schools and small colleges that had previously had no direct access to computers.
Timesharing technology was also applied to the emergent word-processing market. From the start, one market for PDP-8 systems had been the control of automated typesetting systems, and with timeshared operation, this was naturally expanded to include on-line editing of the text to be typeset. The result was a boom in sales to newspapers and publishing houses.
Despite the obvious success of the PDP-8/I and PDP-8/L, these machines were too expensive, and too much of their success was due to name recognition and the established customer base. The root of this problem was the small size of the printed circuit boards DEC was using. This format had been appropriate when each circuit module had a few tens of transistors, but it wasn't adequate when each board could hold ten or more ICs.
The solution was to use larger boards, but this limited flexibility. In solving this problem, DEC arrived at a new solution that laid the groundwork for the next 20 years of small computer designs, the unified bus architecture. The first machine clearly designed in this style was the PDP-8/E, announced in 1970 at a base price of $7,390.
The PDP-8/E was based on the OMNIBUS, a backplane where all the slots were wired in parallel by soldering to a single motherboard. This eliminated the expense of the semi-automatic wire-wrap machines that were required for the backplane wiring on the new PDP-11 as well as the earlier generation of DEC machines, and it reduced the "main frame" of the machine to a box with a power supply and a 20 slot backplane, expandable to 40 slots (of which 2 slots were taken by the jumper between backplane segments).
Everything else in a PDP-8/E, from the front panel to the core memory and input/output options, was manufactured on printed circuit cards with identical 144 pin card edge connectors to plug into the OMNIBUS. Where subsystems were too large to fit on one board, these were packaged as multiple boards connected by jumpers on the back edges, where needed.
The basic CPU consisted of 4 circuit boards, a master clock board, a register board, a control-unit board, and a bus terminator board. The optional extended arithmetic element consisted of 2 more boards; the basic 4K core memory module was 3 boards, and the memory management unit was one board, as were many common I/O interfaces.
Where sales of a few hundred machines had been a reasonable goal in the computer industry, and where, over the 6 year life of the PDP-8 family, only 10,000 machines had been sold, of all models, the initial year of PDP-8/E production delivered over 2000 machines, and within the decade, DEC would be able to build 1000 machines per month.
The outstanding performance for the PDP-8/E was in the face of tremendous competition from 16 bit minicomputers. Not only had DEC introduced its own machine in this class, the PDP-11, but the Data General Nova was selling well, the HP 2100 family was widely accepted, and Honeywell had brought out a series of very well designed successors for the DDP-116.
In the final installment of this series, we will will look at the final years of the PDP-8 family, following it through the microcomputer revolution until its demise in 1990.