5 ms·

A Vacuum Tube Single Board 1-Bit Computer [video]

76 points by zdw 3y ago

Animats 3y ago
That's cute. What's the memory? It ought to be a delay line, for traditional reasons.
- FullyFunctional 3y ago
  This is a demonstration board for the bigger brother, UE-1. UE-1 will be driven by paper tape, though core memory keeps being discussed. I don't think they are considering delay lines -- might be too challenging/fiddly.
  - codedokode 3y ago
    By the way there are cheap and easy to use quartz delay lines used in old TVs.
    - orbital-decay 3y ago
      Those provide 64 microseconds delays, which is probably too short to store data for a slow machine like this. Delay lines used in storage typically provided single-digit ms.
      - Animats 3y ago
        A few of those in series, with a regenerator stage after each one, would work. It's hard to find non-IC delay lines today. "Analog" delay lines in guitar pedals now use a charge-coupled bucket brigade IC. They're analog in amplitude, but discrete in time. I was expecting that someone would still be manufacturing some kind of acoustic delay line product, but there's not much out there. The EDSAC rebuild people built a delay line memory.[1] Not too complicated, but a pain to get it to work. You need a piezoelectric or magnetostrictive transducer driving a metal rod, with a receiving sensor at the other end. Liquid level sensors which work that way are common. A metal rod dips into the liquid, and a transducer at the top sends a ping down the rod. The impedance change where the rod enters the liquid causes a reflection, which is sensed by the transducer that sent the pulse. Converting one of those into a delay line might be possible. You need some length. The speed of sound in steel is around 5,000 meters per second, so you'll need 5 meters for a millisecond delay. This can be a coil; it doesn't have to be straight. A longer delay line can store more bits, but the cycle time is slower. These things are temperature sensitive, so they either have to be self-clocking or temperature controlled. [1] https://www.youtube.com/watch?v=9BA4AyvlKnM https://www.youtube.com/watch?v=9BA4AyvlKnM
        orbital-decay 3y ago
        If I remember correctly some old textbooks I've read, these lines were usually made of invar or a similar alloy to avoid the thermal expansion, and the torsional wave spreads slower (around 3km/s). Reflections were dampened using rubber pads.
  - orbital-decay 3y ago
    Torsion wire delay lines seem pretty DIY friendly, and resistant to interference. They've been used in low-end computers for a reason. (in contrast to the core memory)
    - FullyFunctional 3y ago
      Delay lines are fascinating, but I don't see why this is easier than core memory; seems like you are trading an electronics problem for a mechanical one. Regardless, at the speeds they are targeting, I think delay lines might be difficult. Another interesting option (for ROM) is core rope memory. CuriousMarc has the ultimate explanation: https://www.youtube.com/watch?v=hckwxq8rnr0 https://www.youtube.com/watch?v=hckwxq8rnr0
      - Animats 3y ago
        Because, if you're using tubes, you don't want a huge number of them. Core memories need many drivers and sense amplifiers. Delay lines are inherently 1-bit devices.
      - orbital-decay 3y ago
        Core memory needs a lot of manual wiring, and you need two demultiplexers to address anything in it. Torsion wire delay line has no demultiplexer and is much simpler; it's basically a coil of invar wire on a spacer, with some rubber dampeners, an actuator, and a sensor. And it's inherently serial, many early calculators had serial architectures built around acoustic or IC delay lines. > at the speeds they are targeting, I think delay lines might be difficult Speed of the torsional wave is around 3000-3500m/s depending on the alloy you use. You can make the coil about several dozen meters long before running into issues (practical torsion wire delay lines were 20 to 80m long). That would give 6 to 27ms of delay, which can keep 600 to 2700 bits in the loop at 100kHz, or 10 times more at 1MHz. The same amount of cores would be far more labor intensive to make.
- arethuza 3y ago
  I'd have thought a Williams tube would be ideal? https://en.wikipedia.org/wiki/Williams_tube https://en.wikipedia.org/wiki/Williams_tube
- calamari4065 3y ago
  Depends on your definition of memory. There is no RAM, just a single 1 bit accumulator. It's made of vacuum tubes configured as NOR gates.
osigurdson 3y ago
I always thought it would be fun to create a vacuum tube computer in which the primary computation is performed by a couple of vacuum tubes (i.e. forming a single NAND gate). All of the intermediate results, would be handled by a regular microprocessor but the vacuum tubes are employed to perform the computation. It would glow and look cool (maybe).
- jazzyjackson 3y ago
  trouble with tubes is they look the same whether theyre idle or busy. I like relay compututers, since the clacking really is the computation, they're a step above LED indicating the state of transistors, since compute would happen with or without their indicating.
- JKCalhoun 3y ago
  Interesting. Also, maybe: build some op-amp analog adders, multipliers, and wire them into a digital computer via ADCs and DACs for your math functions. (Could also do log, antilog, maybe other functions as well.)
  - osigurdson 3y ago
    That would be cool as well as the results would have some inherent noise.
kazinator 3y ago
Why would you use tubes if you're not making guitar distortion? Do you really need pentodes at the speeds involved? It looks like the guy is toggling a switch by hand to generate a clock signal. Pentodes are devices for RF amplification. Those 6AU6 tubes are 7 pin units, but they contain only one pentode element. Two pins for the heater, then five pins to hook up the pentode. There exist common 8 or 9 pin tubes that contain two triodes, with which you could cut the tube count in half. (You wouldn't save any money because those tubes cost about twice as much. But less board area and fewer tube sockets counts for something.
- ganzsz 3y ago
  IIRC he has a few thousand of these 6AU6 tubes that he got in bulk. So he has designed most of his tube applications around these tubes. About why to use tubes? I don't know, entertainment purpuoses I guess. His ventures into computing using tubes entertain some at least[0]. [0](n=1)
- asddubs 3y ago
  Why not tubes? Anyone building a computer this primitive out of discrete components isn't doing it because of its utility anyway
  - j_not_j 3y ago
    Tubes. And diodes (see the BOM).
- dharma1 3y ago
  Warmer
- BoxOfRain 3y ago
  > Why would you use tubes if you're not making guitar distortion? I'm planning on building a stereo Mullard 5-10 next year, I know it will be objectively worse than a modern HiFi amp or even a more recent valve-based design at actually being an audio amplifier but I think it would be a fun anachronism to use a homebuilt 1950s amp with modern equipment and there's a lot to be said for the aesthetics of valve amps as well.

76 points by zdw 3y ago

Animats 3y ago

That's cute. What's the memory? It ought to be a delay line, for traditional reasons.

FullyFunctional 3y ago
This is a demonstration board for the bigger brother, UE-1. UE-1 will be driven by paper tape, though core memory keeps being discussed. I don't think they are considering delay lines -- might be too challenging/fiddly.
- codedokode 3y ago
  By the way there are cheap and easy to use quartz delay lines used in old TVs.
  - orbital-decay 3y ago
    Those provide 64 microseconds delays, which is probably too short to store data for a slow machine like this. Delay lines used in storage typically provided single-digit ms.
    - Animats 3y ago
      A few of those in series, with a regenerator stage after each one, would work. It's hard to find non-IC delay lines today. "Analog" delay lines in guitar pedals now use a charge-coupled bucket brigade IC. They're analog in amplitude, but discrete in time. I was expecting that someone would still be manufacturing some kind of acoustic delay line product, but there's not much out there. The EDSAC rebuild people built a delay line memory.[1] Not too complicated, but a pain to get it to work. You need a piezoelectric or magnetostrictive transducer driving a metal rod, with a receiving sensor at the other end. Liquid level sensors which work that way are common. A metal rod dips into the liquid, and a transducer at the top sends a ping down the rod. The impedance change where the rod enters the liquid causes a reflection, which is sensed by the transducer that sent the pulse. Converting one of those into a delay line might be possible. You need some length. The speed of sound in steel is around 5,000 meters per second, so you'll need 5 meters for a millisecond delay. This can be a coil; it doesn't have to be straight. A longer delay line can store more bits, but the cycle time is slower. These things are temperature sensitive, so they either have to be self-clocking or temperature controlled. [1] https://www.youtube.com/watch?v=9BA4AyvlKnM https://www.youtube.com/watch?v=9BA4AyvlKnM
      - orbital-decay 3y ago
        If I remember correctly some old textbooks I've read, these lines were usually made of invar or a similar alloy to avoid the thermal expansion, and the torsional wave spreads slower (around 3km/s). Reflections were dampened using rubber pads.
- orbital-decay 3y ago
  Torsion wire delay lines seem pretty DIY friendly, and resistant to interference. They've been used in low-end computers for a reason. (in contrast to the core memory)
  - FullyFunctional 3y ago
    Delay lines are fascinating, but I don't see why this is easier than core memory; seems like you are trading an electronics problem for a mechanical one. Regardless, at the speeds they are targeting, I think delay lines might be difficult. Another interesting option (for ROM) is core rope memory. CuriousMarc has the ultimate explanation: https://www.youtube.com/watch?v=hckwxq8rnr0 https://www.youtube.com/watch?v=hckwxq8rnr0
    - Animats 3y ago
      Because, if you're using tubes, you don't want a huge number of them. Core memories need many drivers and sense amplifiers. Delay lines are inherently 1-bit devices.
    - orbital-decay 3y ago
      Core memory needs a lot of manual wiring, and you need two demultiplexers to address anything in it. Torsion wire delay line has no demultiplexer and is much simpler; it's basically a coil of invar wire on a spacer, with some rubber dampeners, an actuator, and a sensor. And it's inherently serial, many early calculators had serial architectures built around acoustic or IC delay lines. > at the speeds they are targeting, I think delay lines might be difficult Speed of the torsional wave is around 3000-3500m/s depending on the alloy you use. You can make the coil about several dozen meters long before running into issues (practical torsion wire delay lines were 20 to 80m long). That would give 6 to 27ms of delay, which can keep 600 to 2700 bits in the loop at 100kHz, or 10 times more at 1MHz. The same amount of cores would be far more labor intensive to make.
arethuza 3y ago
I'd have thought a Williams tube would be ideal? https://en.wikipedia.org/wiki/Williams_tube https://en.wikipedia.org/wiki/Williams_tube
calamari4065 3y ago
Depends on your definition of memory. There is no RAM, just a single 1 bit accumulator. It's made of vacuum tubes configured as NOR gates.

osigurdson 3y ago

I always thought it would be fun to create a vacuum tube computer in which the primary computation is performed by a couple of vacuum tubes (i.e. forming a single NAND gate). All of the intermediate results, would be handled by a regular microprocessor but the vacuum tubes are employed to perform the computation. It would glow and look cool (maybe).

jazzyjackson 3y ago
trouble with tubes is they look the same whether theyre idle or busy. I like relay compututers, since the clacking really is the computation, they're a step above LED indicating the state of transistors, since compute would happen with or without their indicating.
JKCalhoun 3y ago
Interesting. Also, maybe: build some op-amp analog adders, multipliers, and wire them into a digital computer via ADCs and DACs for your math functions. (Could also do log, antilog, maybe other functions as well.)
- osigurdson 3y ago
  That would be cool as well as the results would have some inherent noise.

kazinator 3y ago

Why would you use tubes if you're not making guitar distortion? Do you really need pentodes at the speeds involved? It looks like the guy is toggling a switch by hand to generate a clock signal. Pentodes are devices for RF amplification. Those 6AU6 tubes are 7 pin units, but they contain only one pentode element. Two pins for the heater, then five pins to hook up the pentode. There exist common 8 or 9 pin tubes that contain two triodes, with which you could cut the tube count in half. (You wouldn't save any money because those tubes cost about twice as much. But less board area and fewer tube sockets counts for something.

ganzsz 3y ago
IIRC he has a few thousand of these 6AU6 tubes that he got in bulk. So he has designed most of his tube applications around these tubes. About why to use tubes? I don't know, entertainment purpuoses I guess. His ventures into computing using tubes entertain some at least[0]. [0](n=1)
asddubs 3y ago
Why not tubes? Anyone building a computer this primitive out of discrete components isn't doing it because of its utility anyway
- j_not_j 3y ago
  Tubes. And diodes (see the BOM).
dharma1 3y ago
Warmer
BoxOfRain 3y ago
> Why would you use tubes if you're not making guitar distortion? I'm planning on building a stereo Mullard 5-10 next year, I know it will be objectively worse than a modern HiFi amp or even a more recent valve-based design at actually being an audio amplifier but I think it would be a fun anachronism to use a homebuilt 1950s amp with modern equipment and there's a lot to be said for the aesthetics of valve amps as well.