Hacker News new | past | comments | ask | show | jobs | submit login
A Vacuum Tube Single Board 1-Bit Computer [video] (youtube.com)
76 points by zdw on Dec 4, 2023 | hide | past | favorite | 41 comments



That's cute.

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


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.


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)


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


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.


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.


By the way there are cheap and easy to use quartz delay lines used in old TVs.


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.


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


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.


I'd have thought a Williams tube would be ideal?

https://en.wikipedia.org/wiki/Williams_tube


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.


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).


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.


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.)


That would be cool as well as the results would have some inherent noise.


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.


Considering the primary use case seems to be for public demonstration, the tubes are a good way to get people to look at it. I have a sneaking suspicion that the creator might also like tubes.

What's the concern with using tubes? Do you really need tubes for guitar distortion?


Vacuum tubes are his specialty area. He's building a much larger vacuum tube computer and this is supposed to be a more mobile display version he can demo at events.

He's also done quite a few videos restoring old vacuum tube based test equipment from like HP. He does some really cool work.


Well, he described the single board version as a trainer: a way to demonstrate how the larger model works. That's likely why the address/data lines are toggle switches and the close is a button. People can setup the state then execute when ready. The much larger version is the thing you see behind him on the wall (it is a work in progress). It is also intended to go to events, but it is much harder to understand ... thus the trainer.


> Do you really need tubes for guitar distortion?

Technically no. But most professional guitarists still swear by tube driven amps. Solid state is getting better every day, but I think we're still a long way out before most guitar players give up their tube amps.

I am perfectly happy with my Line6 Helix, but a few friends still turn their noses up at it.


Solid state per se isn't getting better¹. Software simulation of tubes ("software state"?) is getting better, or has been (it may have plateaued).

Tubes are still the reference model.

All the DSP cores in a modern amp are just a cover band playing tube tones.

---

1. I mean other than digital solid-state getting more highly integrated and faster, to run more complex software. The necessary analog audio paths in a DSP system aren't getting better.


Yeah, the transistors themselves aren’t changing. Just how many and how well we can throw them at the simulation.

It’s more “tube virtualization” is getting better every day.


> What's the concern with using tubes?

Reliability/longevity, power draw/dissipation, high or higher voltages.

(Speaking of the latter, I wonder what plate voltages the guy is using. I'm guessing it could be well below 100V.)


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)


Why not tubes? Anyone building a computer this primitive out of discrete components isn't doing it because of its utility anyway


Tubes. And diodes (see the BOM).


> Why would you use tubes if you're not making guitar distortion?

Why even bother building something when you can simulate it on your PC in 10th of the time?


> 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.


We could just as easily ask why would you use tubes for guitar distortion?


Mainly because the dead simple circuits (common-cathode class A amplification stages taken out of a textbook) produce the sought-after tones. Nothing needs to be contrived other than some resistor and capacitor values for a certain tonal profile you may be looking for.

I prefer to have solid state almost everything, including the power amp; but there should be a couple of tubes in there for making noise.


Hollow state devices saturate differently than solid state devices. Although FETs have similar saturation curves as tubes/valves (sometimes called "glass FETs" - both are voltage amplifiers while transistors are current amplifiers). The fuzz in most guitar amps is the hard saturation that transistors do. So I guess using tubes/valves for guitar distortion is a fashion/trend. There are a number of pre-amps for electric guitars that use tubes/valves to recreate some sounds from the 40s-60s.


> The fuzz in most guitar amps is the hard saturation that transistors do.

Most guitar amps in what sense? Most guitar amp units shipped? Probably the biggest volume there is amateurs.

> recreate some sounds from the 40s-60s.

Try 40s-90s-00s--...?

Tubes are commonly used right for tones right up to the heaviest metal. It's common for some solid state driving device to be used, like an overdrive pedal. That doesn't necessarily provide any needed gain, but rather shapes the clean tone, which has an effect on what kind of distortion you get.

(Ironically, one difference between "old school" metal and "modern" metal that amps used in modern metal have more tubes: like three 12AX7's in the preamp rather than two. Nothing needs another vacuum tube more than "modern"!)

In other words, the use of tubes in rock guitar continued to evolve well into the 1990s. So tubes are the key to authentically reproducing the sounds heard in 90s and later metal, not just 40s through 60s.

Which is not to deny that solid state devices aren't used for distortion, either as driving elements for tubes or alone.


It's an art project.


Idk. Why not just simulate it on your Ryzen 9?


- Because it's not a hardware project at all then. Using transistors versus tubes is still a hardware project on the same level of abstraction.

- You've not gotten away from hardware; it's still there.

- You're using billions of transistors to simulate the operation of dozens of transistors/tubes, which is inelegant. The chip with those transistors may be small, much smaller than your project, but, in theory, your design with the few dozen transistors could be shrunken down to a chip that would be barely visible to the naked eye.


The guy knows that tubes are not the most efficient way to implement this. He's making it with tubes because it's cool. It's not that hard to understand.


The topic has changed in this subthread; this is now why build something (with discrete components, be they transistors or tubes) rather than simulate.

The educational value of building hardware is obvious to me; it cannot be replaced with simulation, and simulation runs on hardware that someone built, who knew how.


Good to know that the topic has changed. Let me know if it changes again!


He's building a CPU from scratch. Practicality is not a concern at all.


Warmer




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: