22

Prompted by this question querying the prevalence of byte-addressable memory on machines with 32 bit registers: Why is every address in a micro-controller only 8 bits in size?

I'm familiar with the concept that the mass-market microprocessors from the 1970's were quite resource constrained with typically 8 or 16 bit registers, and maybe before then computers were closer to one-off designs rather than based on standardised architectures. Was there any 'typical' register size for early electronic computers, and what were the extremes of this range?

I'm assuming here that early computers did actually have identifiable general purpose registers as we recognise them today...

34

It varied immensely, and many early computers have unique architectures that might make answering the question difficult. However the common practice in the earliest days seems to have been large word sizes, suiting their use as pure number crunchers, and their low clock speeds. Most of them were word-addressed, so you could say they had a "byte" equal to their word size. A smattering of examples of pre-1960 machines:

ENIAC (1946): 20 decimal digits.

Manchester "Baby" (1948): 32 bits.

Manchester Mark 1 (1949): 40 bits.

EDSAC (1949): 18-bit "word size", but available registers were the accumulator (71 bits) and the multiplier (35 bits), and values in memory were either 17 or 35 bits (in all cases, one bit of the 1/2/4 word unit is unusable).

EDVAC (1949): 44 bits.

Pilot ACE (1950): 32 bits.

IBM 701 (1952): 36-bit words; 38-bit accumulator including overflow bits.

UNIVAC (1952): 11 decimal digits plus sign.

IBM 650 (1953): 10 decimal digits plus sign.

PDP-1 (1959): 18 bits.

  • The "baby" had 32-bit words, but could only operate on one bit at a time. Relatively little logic was needed to make each instruction operate upon 32 bits in sequence, and the presence of that logic reduced the number of instructions required to perform tasks. – supercat Apr 19 '17 at 16:40
12

Early computers used a variety of register sizes, depending on the types/sizes of problems they were intended to solve. For example, early Digital Equipment Corporation (DEC) computers were 12-bit, 18-bit and 36-bit. The extremely popular PDP-11 was a 16-bit machine (byte-addressed), and eventually led to the 32-bit VAX computer (also byte-addressed).

Larger registers allowed calculations to be done more quickly (more bits processed per clock cycle), but were also more expensive to build and operate. Smaller word sizes allowed a computer to be cheaper to manufacture, and physically smaller as well, so that they could be used in environments where a larger computer was not possible or practical (e.g. in a laboratory). Since computers at that time were still very expensive, it made sense to offer computers with multiple word sizes at multiple price points.

10

Typical register sizes varied wildly among the early computers. For example, ENIAC's early design had nothing but registers, and these were ten decimal digits wide. The Harvard Mark I was also decimal, with 23-digit registers. The Zuse Z3 used binary, with a 22-bit accumulator.

The first computer with registers other than accumulators (the Manchester Mark 1) used 40-bit registers, while its commercial variant, the Ferranti Mark 1, used a mix of 20-bit, 40-bit, and 80-bit registers.

5

I'll keep my answer short, since there is an excellent list of word sizes on Wikipedia.

You mention the 70's, and obviously by then most new chips were standardized on numbers of bits - typically powers of 2 - so 4, 8, 16, 32 and 64 were the most common. But some earlier machines used a length of digits insteads of words, and a few even had variable length words with a terminator.

Here are some notes on some older register architectures, but Wikipedia's list is hard to beat.

  • 1
    It would be good if you included some of those notes in the question; the non-link content you have at the moment is quite vague. – wizzwizz4 Oct 29 '16 at 10:00
  • 1
    The wikipedia link is nice, with instruction size and address resolution. I guess this is a problem with history questions, the good answers are written already on other sites. – Sean Houlihane Oct 29 '16 at 10:13
  • 1
    Yeah. I will try when I have time, but it will be a lot of work to summarize that Wikipedia content. My knowledge in this area is mostly limited to a computer architecture paper 27 years ago, – Nick Westgate Oct 29 '16 at 10:24
  • I have taken the opportunity to add Titan to the Wikipedia page; it had a word length of 48 bits. – Michael Kay Jun 8 '17 at 16:37
  • @NickWestgate Have you time now? :-) – wizzwizz4 Aug 16 '17 at 13:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.