Half a croissant, on a plate, with a sign in front of it saying '50c'
h a l f b a k e r y
Why not imagine it in a way that works?

idea: add, search, annotate, link, view, overview, recent, by name, random

meta: news, help, about, links, report a problem

account: browse anonymously, or get an account and write.



Seriously Serial CPU

One valve performs all functions.
  [vote for,

Pocket calculators used to cut down on transistor count by performing arithmetic operations serially. I know this thing but not the details. And of course there are such things as bit-slice CPUs.

Taking this from the larger to the smaller scale, this is what we have:

1. At the largest scale, there's a CPU which appears to have a certain word length - say 64 bits. However, it in fact consists logically of an array of 32 two-bit processors. Physically, however, it consists of a single processor used thirty-two times as a separate processor, with the inputs and outputs temporarily stored so as to look to the outside as if it's a sixty-four bit processor. I would say this could be dealt with by some kind of shift register type thingy, but that would involve lots of logic gates and I want to avoid that, because that means more switching elements.

2. At a smaller scale, the ALU and registers, such as they are (probably just a stack pointer and program counter) are also serial. As with the PDP-11, memory-to-memory addressing is used and any other registers are in memory, including flags. Instructions are also decoded in series.

3. At a smaller scale still, each logic gate is only virtually realised. Whereas there may appear to be a two-input binary adder consisting of a few logic gates, again they are working in serial with delays, and in fact consist of the same gates used repeatedly, pretending to be other gates.

4. Still smaller, they are of course realised using NAND or NOR gates in various configurations, and naturally these are all the same logic gate reused.

5. Below the scale even of logic gates, the switching element itself is just a valve (or transistor if you insist) reused in various ways in order to realise the logical equivalent of an NAND or NOR gate, depending on which way you want to go.

Ultimately, the CPU consists of a single valve plugged into a socket which connects with itself in various ways using various other components arranged very cleverly, probably mainly capacitors and resistors. But to a programmer, it just looks like a normal 64-bit CPU like any other.

Why a valve? Well, obviously retro styling, but also it means you can just plug it in and let it glow. When it finally blows because of all the turning on and off, you get another one.

It would probably be a good idea to stock up in advance with a couple of million or so valves so it lasts more than about a minute. Maybe there should be an automated VLSI-based system for replacing them.

nineteenthly, Jan 31 2018

Colossus computer https://en.wikipedi...i/Colossus_computer
"The first Mark 2 Colossus, containing 2400 valves ..." [8th of 7, Jan 31 2018]


       Switching speeds of valves are vastly lower than switching speeds of solid-state electronics. I expect the valve to generally last quite a bit longer than a minute. However, the real question is, with that slow switching speed in mind, and given how much work it has to do to emulate a 64-bit processor, can it last long enough to get any particular job finished?   

       (For those who don't know, likely Americans unaccustomed to this particular Britishism, a valve is a vacuum tube.)
Vernon, Jan 31 2018

       Even with a complex thyratron/dekatron valve, there's no way that a single tube can ever approach the necessary complexity.   

       It might be possible - in theory - to build a "computational element" using thermionic emission in a large vacuum chamber (glass, or stainless steel) then pump it down to a level where it would work - but one failure would mean back to square one. Plus, even using a digital regime, isolation between elements would be a serious problem.   

       To build a basic flip-flop, you essentally need two valves, altho there are a few specialist "logic" types.   

       Building any sort of computational device using a single off-the-shelf valve is impractical. And very halfbaked.
8th of 7, Jan 31 2018

       How would you add with such a device?
RayfordSteele, Jan 31 2018

       Some sort of really fragile abacus ?
8th of 7, Jan 31 2018

       The problem is memory, if the output of the gate in its previous state has to be fed into the next state, it needs to be saved somewhere, and memory normally means flip-flops, which are made from gates. And if you have memory, you already have enough to build logic, since a single unit of memory can function as a lookup table. This is the basis of FPGAs.   

       I like the idea of memory as logic, an n-bit memory with an m-bit address space can function as an n by m lookup table. And all combinational logic can be represented by a lookup table mapping inputs to outputs.   

       Instead of wasting logic elements on building memory, why not go Seriously Serial and fit a delay-line memory? This would need at least one tube to amplify and retransmit the pulses, but if the delay line was long enough you could store a fairly substantial lookup table. You can build, for instance, an ALU using just combinational logic, and perhaps with multiple delay lines feeding the same tube you could store input data and opcodes, I don't know. You would also need some magical gate-free way of starting the system up to initially load all the data.
mitxela, Jan 31 2018

       Colossus used paper tape. The rebuild at Bletchley Park demonstrates it perfectly.   

       But there are a LOT of valves ...   

8th of 7, Jan 31 2018

       You don't need flip-flops to store things. Semiconductor- based dynamic RAM, for example, consists of capacitors and other techniques are available. For instance, you could twang a wire, use CRTs with metal plates in front of them, use rings on wires or a mercury delay line. There's no need to squander logic gates on making memory.
nineteenthly, Feb 01 2018

       That correlates with my experience here in fact. I often feel I'm discovering ideas rather than inventing them, and I kind of feel all invention is in fact discovery. Ideally, then, a computer is a truth finder. Makes me think of quantum computers somehow.
nineteenthly, Feb 01 2018

       So a newly commisioned build, the quantum computer named "The Eye of Serialon".
wjt, Feb 02 2018

       [IT], that's either a deeply profound insight into the nature of the Universe, or specious rubbish.   

       If only it were possible to determine which ....
8th of 7, Feb 02 2018

       Just ask it. If it answers, then it's probably true, or possibly enjoying a good laugh at your expense.
RayfordSteele, Feb 02 2018

       //the truth doesn’t need storing, merely accessing// This is fine as soon as you've built a device to scan the entire universe in real time. Don't forget transactional integrity. Again, this is easy, once you've persuaded the universe to slow down and wait for you.
pertinax, Feb 03 2018

       Why do it all?   

       The computation only needs the prime factors and processes to the E-collar that our simple focus is interested in. A bit of extrapolation and modelling and the stating of truth will be like magic.
wjt, Feb 03 2018

       [Ian], the reason for my silence on this page is that you've triggered an interest in functional programming which is taking up my time. It occurs to me that just as there was once an object-oriented CPU, maybe there could be, or perhaps already are, *purely functional* circuits. But it also occurs to me that there might be some kind of signal processing device which is effectively already this, or that it's only a particular perceptual filter which prevents me from seeing that they are already functional.   

       My only experience of functional programming is APL. I can see that as a kind of over-featured calculator, and speculate as to the hardware of my imaginary device.
nineteenthly, Feb 03 2018

       Toggle realys make very nice memory modules A bit slow but giving the single tube computation that should be fine.
dev45, Feb 03 2018

       That interests me rather too much Ian. I'm very easily distracted, and now I know there's something interesting called Elm out there it may dissipate my energies rather. But thanks anyway.   

       I currently have an image of data shooting through boxes without touching the sides and I'm kind of thinking everything could stream, but there could be major eddies in that stream.
nineteenthly, Feb 03 2018


back: main index

business  computer  culture  fashion  food  halfbakery  home  other  product  public  science  sport  vehicle