The PDP8 was one of the first computers I had both hardware and software access to.

In 1976 I got my first real job in the computer industry. Amongst other hardware, I worked with PDP8s . They were not new then, and I think we got the 8i because it was scrapped from somewhere else! I was responsible for keeping the hardware going, and also improving the software. MSI logic made it possible for anyone with the time and money to build a minicomputer. Minicomputers were not small or cheap - they typically cost the same as a six bedroom house! Most people did not get to play with the hardware. The PDP8 I used normally ran multi-user, although, in practice, it only supported two users!

I liked the simplicity of the design of the PDP8. At the time it was designed, most machines had longer words, and 12 bits was considered quite a small word. This was because all the early computers had a fixed word length. It was not till IBM came out with the 360 architecture that anyone used variable length instructions. 12 bits is about the least you can get away with and have a usable computer. Legend has it that DEC asked MIT to design the smallest number of gates you could put together and call it a computer! The PDP8 was extremely popular as an instrument controller. One of the two I used was originally used for testing patients at University College Hospital. Its had many other uses, and ended life as the DECMate - a word processor! SOme PDP8s were still in use in 1990, and for all I know, there may be some still in use!.

Most programers dreamed of having a computer on our desk - the ultimate dream was a "Ten On A Desk" or "Toad". The "Ten" was a PDP10 - a mainframe commonly used to supply the enftre computing power of a large university. Today (2002) there is actually a company selling one now! At least two people are building PFGA based clones of the PDP10.

Lessons learned in my early years of computing

In the mid 1970s, minicomputer manufacturers were springing up on every corner. Most had no softare, although nearly all claimed ot have Basic and Fortran. It seeemed pretty obvious to me that without software, the machines were useless. This was when Intel had just released the 8080, a precursor of the 8086, and other people were announcing microprocessors. These could be used to assemble a computer for about the price of a cheap car. However, the resultant machine not only had no software, it also had no support for peripherals. One or two companies addressed this issue - the most famous being Imsai. There was a lesson here - a computer is not the CPU, its the entire system.

In 1976, I could not afford the time or money to buy a computer, not even to even build one. I planned how to make one, one day, and studied the architectures of the Nova, PDP11, and a couple of hundred others to figure out the best way to do it. Later on, working for GEC, I was on the design team of several computers. I learned that designing a computer is a lot more than designing the "programmer's model". Far more work goes into thermal management., component sourcing and a mass of other "management" issues. I was always impressed with the way the early DEC machines were designed.

The PDP8 Today

The PDP8 has had the longest lifespan of any architecture (even including the 8086) - since announcing this project, I have heard from at least one company who is still using one for real work! However, the 12 bit word is relatively rare today. I know of no processor in volume production today that has a 12 bit ALU and datapath. However, In the course of my career, I have designed two different systems which used 12 bit words, which were implemented using TTL/bit slice technology. (One was a controller for 1/2" tape, and the other a peripheral controller.) I found that, where you are not primarily handling ASCII, 12 bits was a very good size. The DECMate wordprocessor showed that even when handling ASCII, its not a bad size.

Naturally enough, most manufacturers concentrate on where the largest market is, and both 8-bit micro-controllers and 32/64 bit data processing are much bigger markets. However, there are situations, if relatively fewer, where small size and CPU cost are insignificant - considder:

• Deterministic execution is required - ie cache is not acceptable, and complex instructions are not desirable
• Access to more than 8 (but 12 or less) bits in a single cycle is required (generally true for A to D or D to A)
• Centralised control over a significant number of physically large peripherals is required - test and measurement is a prime example
• Long product life is required - the latest 8 bit micro or even 32 bit micro, is unlikely to remain in production for more than 3 years. In these markets, new developments with Moore's law make a new product more attractive. For some products a working life of 10 or more years is attractive, because the development cycle may exceed 5 years, and then there is training, etc. Hardly sensible to design in a component that has ceased production before you ship the first system!
• CPU cost is irrelevant as a component of system cost - who cares if you save $10 on the price of a CPU, when your system costs $50,000? If you have to connect 32 peripherals, then reducing the cost of the interconnect or peripherals, and ensuring reliability of interconnect, is far more important.

There IS a case for a machine with an architecture that is tried and tested, and implemented as a system and not just a CPU chip. I intend to implement this system in such a way that it could be used in an 8, 12 or 16 bit configuration. I am using a 12 bit engine as it is the simplest to implement!

I have the skill and experience to build this type of system, and I will!

© 2002, Andrew C Grillet, All rights Reserved