5

Considering the PDP-8, PDP-7, PDP-9, PDP-1, even though they are completely different architectures (from a programmer's or compiler's point of view at least), they have some remarkable similarities in some details.

For example, the microcoded instructions on both the PDP-8 and PDP-7 have cla, cll, cma, cml, and single or double rotations, which conveniently build up small constants in the accumulator.

The same two computers also have exactly the same skip instructions, and a bit in the instruction to indicate an indirect memory access. The same operations are also available in the PDP-1 instruction set, as are, obviously, things like add, and load.

The same two computers also have an optional EAE, which provides another register MQ and some mathematical operations, though the size of MQ obviously varies between 12 and 18 bits.

So straight away, my mind is supposing that there's some kind of commonality in the implementation as well.

  • Maybe there's even a single part, which implements all these operations on a bit or group of bits, and which is installed in all these computers to implement a register and ALU.
  • Maybe there's some part which takes a bitfield from an instruction, and determines which of these operations need to take place (and is thus part of the instruction decoder) and is installed in at least some of these computers.

Are there large subsystems in common between the implementations of both 12-bit and 18-bit PDPs, which are responsible for the patterns and similarities between these two groups of computers?

2
  • 1
    DEC was -- at the time -- a small company known for making standardized Flip-Chips. I wouldn't be surprised if they reused a lot of ideas and parts to save engineering time/costs.
    – RonJohn
    Commented Sep 25, 2018 at 15:53
  • Toyota builds a lot of cars with the same engine as well. Commented Oct 12, 2018 at 14:22

4 Answers 4

9

Yes, at a physical level there was some commonality, but not at as high a level the question references.

In the mid 1960s, DEC sold a line of simple logic modules, known as flip chips. The PDP-7 and the original PDP-8 were built (at least partially) out of the R-series modules, though there were some custom modules created just for them (e.g. a W130 specific to the PDP-8. These were interconnected through a wire-wrap backplane, as illustrated

These modules preformed fairly simple logic function that were highly reusable, for example a R107 module implemented seven inverters (NOT gates) using discrete transistors. You can find the datasheets and documentation for the flip-chip modules in the Digital Logic Handbook.enter image description here

Later versions of DEC processors were more integrated, with more functionality on one custom board, and as integrated circuit technology developed, a SSI digital logic chips could do everything a flip-chip could, in much less space. For comparison, a SN7404 IC implemented a six bit inverter function similar to the R107 flip-chip.

By 1969, new DEC designs began using integrated circuits rather than the discrete transistors of the older flip chip modules. The modules however remained in use for some time after this. If memory serves, the early PDP-10 and PDP-11 models also used the red-handled R-series modules as well. Those machines were in use well into the 1970s and early 1980s, so spare modules for repair were common.

Sometimes just reseating or cleaning the contacts on a module's card edge connector would fix a problem, the modules were fairly robust, and a Pink Pearl brand pencil eraser was commonly used for cleaning (with modules removed from the backplane.)

It may also be of interest to note that in the 1970s, Texas Instruments ICs such as the 74181 ALU bit-slice were used in processors produced by several different vendors, including DEC. But these did not influence the instruction sets much either.

10
  • Out of curiosity, what was the purpose of the flip chips? What sort of things, other than computers, were the found in? Commented Oct 15, 2018 at 18:22
  • FLIP-CHIP modules were probably used almost exclusively in building computer-related applications. For example, peripheral device controllers (e.g. tape and disk drives.) Commented Oct 15, 2018 at 19:12
  • 2
    Many modular technologies like that offering low-scale integration of logic likely found use in dedicated path computational devices before and in addition to being used in stored-program computers. For example I once took apart an electrical load demand controller that used digital logic and BCD math implemented in 44-pin plug-in circuit modules - they were programmed with a wire-wrap gun to create data paths performing certain computations, rather than having a sequential program stored in a memory. Most of the modules had discrete components, a few had early ceramic ICs. Commented Oct 19, 2018 at 16:59
  • 1
    FlipChip was a DEC trademark and so only used for their products: computers, device controllers, etc. There were also sold to customers building their own interfaces or other digital systems. They had a iconic plastic handle with the color indicating the series (speed, circuitry type, function) and specific dimensions. They were made possible by Sylvania sockets and Gardner-Denver wire-wrap machines. The IBM SMS line of modules, used in their computers, used the same sockets and wire-wrap technology, but AFAIK were not sold individually. They had their own specific dimensions. Commented Apr 29 at 16:45
  • 2
    @dave: Indeed, at least in the early PDP era, DEC didn't make computers. They made "Programmable Data Processors".
    – supercat
    Commented Apr 29 at 17:17
9

There were obviously no same parts that were doing any of those operations.

The DEC hardware evolved from transistor technology (PDP-1, PDP-8, etc.) to medium-scale IC logic like standard TTL (PDP-8/E, early PDP-11) and eventually to large-scale ASIC CPUs like LSI-11 or 6100 PDP-8 clone.

The similarity of instruction sets has nothing to do with the actual implementation of the CPUs. Thus, even on discrete transistors, you can have just different implementations of flip-flops or latches, different implementations of the same logic gate like NAND or NOR (for example, DTL-like logic vs ECL-like logic), etc. Level up, you can have different logic implementations of the CPU, for example, PDP-8/S is a bit-serial CPU while all other PDP-8 machines are bit-parallel. On latter bit-parallel CPUs, the execution of the command might be microcoded or hard-wired which again results in completely different logic structures of the corresponding CPUs.

The ISA (instruction set architecture) is the very top level of the CPU, and you can have two completely different CPUs executing just the same instructions. Therefore, similar or even coinciding instruction sets of two different CPUs yield exactly no information of their internal structure and how they execute some specific instruction.

9
  • 1
    Are instructions ever provided which take special advantage of the underlying architecture, or are the two designed completely independently? Commented Sep 25, 2018 at 19:28
  • I wouldn't be at all surprised if there is some commonality at the logic level, if only because there are only a few ways to implement many functions and, if the same engineers built the different computers they are naturally likely to reuse previous designs.
    – JeremyP
    Commented Sep 26, 2018 at 9:44
  • I think that instruction set architecture most often is the basis, on which the logic structure of the CPU is then built.
    – lvd
    Commented Sep 26, 2018 at 12:23
  • 2
    Many early 16-bit minicomputers, such as the HP 2100 series and the Data General Novas, borrowed a lot of concepts from the PDP-8: A bit to indicate indirect addressing, page-mode memory with a bit to indicate "current page" vs "base page", etc. I assure you that there was not a common implementation of these things among these different makers' machines! More: IC technology was nowhere near at the point where such a thing would have been feasible. A given IC (14- or 16-pin DIP) might have four two-input gates or a pair of flip-flops in it. Not an "implement indirect addressing" circuit. Commented Sep 26, 2018 at 14:43
  • 1
    And then there are form factors to consider. DEC for many years built their machines from a large number of "flip chip" modules, each implementing a small function, that were connected together by wire-wrap connections on the backplane the modules plugged into. But there were several generations of these modules, from ones using discrete transistors to later ones carrying many ICs. By contrast the HP 2100 was built on half a dozen or so PCBs each about 8 x 9 inches, plus many more for I/O interfaces. The DG Nova used a single board about 17 x 20 inches! There just wasn't much in common. Commented Sep 26, 2018 at 14:47
5

As I recall, the PDP 1 through 6 were built from system modules, and these came out before flip-chips. The 7 was like the 4, only the 7 had flip chips. The 8 was like the 5, only the 8 had flip chips. The 10 was like the 6 only the 10 had filp chips.

The 11 was a departure in architecture in several ways. Word size, instruction set, memory addressing, and so on.

5

To start with, the 18-bit TX-0 was (AFAIK) the first machine anywhere (counter-examples welcome!) to have an 'operate' instruction which used bits in the address part of the instruction to create operations using implied operands (AC, PC). This trick was carried over into all of the DEC 18-bit and 12-bit architectures because it was so useful: it reduced the number of opcode bits needed for memory-reference instructions. It was something the designers had used before, and they just kept using it.

So IMHO the important commonality here was the group of people, including Ken Olsen, Harlan Anderson, Wes Clark, Ben Gurley, Gordon Bell, Alan Kotok, Dave Gross, and many others. Some had been students at MIT, some worked there after their student days, and many of them ended up at DEC. Doing research for Project SAGE, they created the real-time interactive mode of computing (as opposed to batch data processing or scientific number crunching). The people knew each other, worked together, and knew about each others' work.

The machines built at MIT included the Whirlwind, Memory Test Computer, TX-0 ("tixo"), TX-2, L-1, LINC, and others. These machines all had a short word length, fast parallel arithmetic, and core memory. Significantly, it was easy to interface new I/O devices such as graphic displays or communications lines to them.

Just as architects might speak of the Bauhaus School of Design, I call this the MIT/DEC school of design.

The original DEC Laboratory Modules and System Modules were based on the circuitry that Ken Olsen had designed for the TX-2. These products got the company started. Ben Gurley said that his mandate for the PDP-1 was to design it from inventory -- the initial set of System Modules. Its architecture was very similar to the TX-0. Gordon Bell designed the PDP-4 as a cheaper PDP-1. Bell & Alan Kotok designed the PDP-5 architecture; the link bit in the PDP-5 came directly from the L-1 computer. Bell, Kotok, and others designed the PDP-6. Manufacturing difficulties with the PDP-6's large, hand-soldered backplane inspired the Flip Chip product line, using Gardner-Denver wire-wrap machines. The PDP-7, -8, -9, and -10 were new implementations of the earlier architectures using this newer packaging and newer circuit technology.

IBM marketing convinced the customer base that memory came in 8-bit bytes, instead of words with a multiple of 6 bits, and folks had new ideas which led to general registers and addressing modes, resulting in the PDP-11. Eventually DEC created the VAX-11 to get around the PDP-11's 16-bit address space, and then the Alpha, a 64-bit VLSI RISC architecture.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .