Vacuum Tube Computer P.20 - IF-THEN Hardware 

Thank you so much! This is a project that I'm having a ton of fun making, so it means a lot to hear that others are enjoying the journey along with me! I hadn't heard of PBASIC used on the BASIC stamp before, but that's really cool! I had no idea there was a microcontroller out there that had a variant of BASIC on it, I may have to hunt one of those down someday and give it a play!

It's getting there, I'm starting to get super amped up as this almost year long project draws nearer to (1/4) completion! Memory, Program Control and I/O are all sorts of new an interesting problems, but with processor working, it'll feel doable. I've pretty much put all other projects on hold until I get the tube processor fire up and running!

Thank you so much, it means a ton to hear you're enjoying the episodes!

I'll put some heater glow shots in the next video for you!

Thank you so much for the kind words! And I want to thank you for being super interactive and participating on the videos, it means a lot to hear from viewers and see that they're enjoying the videos!

Thank you so much! All credit to OCRemix for the music, there's an amazing collection of tunes over there!

Thank you! And Chrono Trigger music is always a safe bet!

Thank you so much! I'm super amped up about it being so close to mostly complete (though this is just the beginning, memory is going to be a proper nightmare)!

@@UsagiElectric I'm especially interested in how this will be done I can't wait 😊.

Thank you so much! When I first started learning about the MC14500, I totally didn't get the IEN and OEN functions. But, after starting to dive into it in a bit more detail, it's a little crazy just how powerful they can be! It can also get even more impressive if the designer builds additional jump circuitry separate to the MC14500 working on the JMP flag. That way you can use IEN/OEN to execute conditional IF-THEN-ELSE statements and have the ICU conditionally toggle the JMP flag. Thank you again for the wonderful comments on the projects! It means a ton to hear that you're enjoying the content and getting something out of it!

Thank you so much! The completed CPU will have, interestingly, exactly 190 tubes on it. Each tube heater is rated at 300mA at 6.3V, but I'm undervolting them by running them in series off of 24V, so they're actually closer to about 280mA each. That comes out to about 1.7W to 1.8W per tube, and with 190 tubes, that's a massive 340W of power! Fortunately, that's split between the two power rails, with the 24V rail powering 156 tubes and the -12V rail powering 34 tubes, which comes out to 24V @ 12A and -12V @ 5A. It's a power hungry beast! You can definitely feel the temperature in the room creeping up the longer its on, haha. But, the processor is just the tip of the iceberg. I reckon I have another 500 tubes to go before the entire computer is complete!

I think the next module you build will need to be the air conditioning module...

Thank you so much!

Thank you so much!

Hah, that's just the bag that's open, I've got six more ziploc bags filled full of 6AU6s! If you want, come join us over on the Discord chat server, I'd love to try to help you work on the design in Logisim and share files! ( discord.gg/p7UsfHD )

@@UsagiElectric Thanks alot! That's very kind! I'll check it out (never used discord before) :D.

I'm getting pretty excited about how it's coming together!

Thank you so much! Using little plug in modules like the IBM604 used could really make a huge difference! Especially if I made them just a little bit taller and managed to squeeze two NOR gates per module. That means I could build an entire D Flip Flop with just three modules, which would be an insane footprint reduction. Currently, one of my D-Flip Flop boards that uses 6 tubes (like for OEN, IEN and Instruction Registers), is 150mm x 60mm x 55mm (LxWxH). That gives a total area of about 0.5L. The plug in module the 604 uses measures in at 25mm x 25mm x 145mm (LxWxH). It's quite a lot taller, but the footprint is minuscule. If we bump that up to 150mm height to stuff to NOR gates in there, that gives us a total area of about 0.1L per unit, and with three units that's a total area of 0.3L, so a pretty notable reduction in volume. But the real kicker is the reduction in footprint. With the way I have it laid out, height is totally inconsequential. Whether I'm 55mm or 145mm makes no difference at all, but the footprint drops from 150x60 to 75x25, which is insane. That means I could very nearly fit four D Flip Flops in the same footprint as my original one D Flip Flop. That's a massive reduction in size! Interfacing with I/O is going to be a serious challenge! Magnetic tape should be relatively easy since the amplifiers to read it can all be run at 24V, but interfacing with an output device like my Teletype is going to be quite difficult. That thing needs a 120V DC 60mA current loop, which would be a mess, so I need to build some kind of tube based opto-isolator to interface with that. Then of course, if I want to interface with something newer over RS232, I need to go the other way, which is going to be equally as interesting. But, that's all for future Me to worry about, haha. The 24V supply is still totally fine, but the 12V supply just isn't keeping up. Ultimately, I'll need about 6A on the -12V rail, and the little wall wart in my current supply can only supply up to 3A. So, I'm working on building a new supply that has a beefier 12V on it. I've got it mostly built, I just need to wire it all up now!

@@UsagiElectric Cool, thank you for the detailed analysis. Very interesting to see this idea in actual numbers ... and fascinating, that you already did all the math. I'm really looking forward to the next part. :)

Fast relays were used for TTY interfacing back in the day.

That's actually a topic I really want to make a video on in the near future! It'll be a few episodes before I get to it though, but in the meantime, I did go into a bit more detail about how I mill my PCBs in a very, very early episode on this channel here: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-F2FRN5z2S78.html

It's actually a cover of a piece from Chrono Trigger, which is totally from the same era, so you're not far off!

The diodes totally can be replaced by tubes! I did some testing using the 6AL5 dual diode and it works brilliantly, even at low voltages. Check out my 1-minute video on making an OR/NOR gate using the 6AL5 and 6AU6 here: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-5YrcGr04kKk.html

Thank you! I think there's a couple different reasons 9-pin and 7-pin tubes were primarily used for the majority of vacuum tube computers, but the main reason I'm guessing is timing. The large majority of vacuum tube computers were built between 1950 and 1959 - there are only about 10 computer built after 1959 that used tubes, and a lot of these were actually hybrid computers using a mixture of tubes and transistors. So, with regards to the Nuvistor, that wasn't announced and put into production until 1959, so it was a bit too late to the party. For sub-miniature tubes, like the CK502AX, I think there's actually a couple of reasons they didn't use these often. These were actually available in the early 50's and you could find them in some of the very first electronic hearing aids. But, I don't think they were being produced in the large numbers that would have been required for vacuum tube computers. For example, the IBM 650 uses approximately 2,000 tubes and there were around 2,000 of the machines built, meaning for the 650 model alone, IBM needed around 4 to 5 million tubes. Also, tubes are tough as hell, but with 2,000 tubes (and the 650 was a small computer), tube failures are inevitable. 7-pin and 9-pin tubes can be unplugged and a new tube plugged in really quickly, keeping downtime to a minimum. Submini tubes though don't have a socket and must be soldered in place, which would take much longer to replace. Finally, I think there might have been some actual design reasons why subminiature tubes weren't preferred. Subminis, to the best of my knowledge, don't actually have a cathode, they're directly heated tubes, utilizing the filament as the cathode. But, for tube computers, sometimes you need the cathodes to sit at weird levels, which would have been quite difficult to design with directly heated tubes. The engineers certainly could have worked around this, but it might have been more trouble than it was worth. Of course, that's all speculation on my part, so hopefully I'm not too far off the mark!

@@UsagiElectric Well, you're thorough....Thanks for the detailed response. I guess my point is if they made millions of tiny specialized vacuum tubes for Pozit shells for example, they could have made smaller, less power hungry vacuum tubes for computers--putting more than one device in each bulb, for example. The building and AC was a major expense for computer "installations" back in the day (Sage). When I see the very large logic circuits compared to Pozit and other miniaturization techniques they knew about, it seems like they missed the boat on that. I get why Tommy Flowers had no choice, but for production, they could have developed specialized "micro miniature" parts IMO. Which is why. your reconstruction looms large. Really nice job BTW. In 2005 (100th anniversary of the valve). I made some DIY triodes. Not very good ones. But similar to the gassy Audions DeForrest first made. We made transistors in HS with a kit from Bell Telephone Labs. The kit contained asbestos and hydrofluoric acid. Today, they'd be locking down the school.

And they all (ARCH64, x64) did away with this

Whoa, that's really interesting! I'm really unfamiliar with the ARM architecture in general, but it's really cool that they (at least initially) had the capability to do conditional execution. I can definitely see how it would be preferable for really simple conditional statements, but I can also see that there are some serious limitations to it as well. Having both conditional execution and the ability to jump to different spots in the program counter would be extremely useful though, I feel!

@@UsagiElectric conditional execution of a JMP or JSR is only some special case. They had 32 bit in their instruction word a looked for application. MIPS use those bits to address more registers instead. The older MIPS won -- but only when combined with branch prediction. Branch prediction killed the branch delay slot, which is evil anyway.

Yup! Although, it was really common for tube computers in the late 50's to use germanium diodes, very similar to what I'm using here. The IBM 604 and 650 both used quite a few silicon diodes in places where space was a priority (though from the documentation, it seems IBM still preferred tube diodes where possible). But, the 6AL5 is a brilliant little dual diode tube that I've actually used to build up some test pieces of the logic gates I use, and it works brilliantly! The only downside is the number of tubes, and subsequently tube heaters, would double!

I'm happy to see this as I didn't see that part of the video and I was mortified to realize it looked like this was all being built without sockets! I was having a mini panic attacking wondering how to replace one of those tubes and what the tube life expectancy would be like if they had all been soldered to their legs.

So far they seem to work really well! A regular tube socket is obviously a better solution as they allow for some movement due to thermal expansion. These sockets in the PCB could cause some stress on the glass base under extended use, but so far I've had absolutely no problems with them. I haven't done any testing to see how the spring tension holds up to high multiples of plugging and unplugging, but I do have some that I've pulled the tubes out of at least 10 times and the spring tension still feels totally normal!

@@RN1441 That's probably the number one thing I get comments on! Since the tubes are right on the base of the PCB, it really looks like I soldered them in. But, these little 1mm pcb pin headers seem to work really well, aside from giving an interesting appearance that is, haha.

@@UsagiElectric Interesting, good to hear you're having good experiences with them. I'd wager a guess why companies don't use it is as you say, over a few thousand thermal cycles the joints may crack. Not to mention it's a lot more time consuming than standard sockets

Solid state diodes are a 19th century invention, "modern" junction diodes are a product of 1950s. Before the 1N34 launched in 1946 vacuum tube diodes would probably be considered more reliable .

Thank you so much! I do kind of have it planned out quite early. The PCB design for this last section I finished about three months ago, but only properly finalized it about two weeks ago (the work on memory required some rework on the processor). But, milling out the PCBs and soldering them up happens pretty much just a week before an episode goes live, so the physical state of the processor is pretty much exactly as it is in this video right now. The PCB design for the memory is now about 90% complete as well. I'm quite excited to finish up the processor and start with testing of memory elements and then starting construction of that. The memory is kind of a double-edged sword. It's a fairly simple design, but there's a massive amount of repetition, so there may be some big gaps in videos since most of the memory will just be making the same boards over and over again. Still, it should be quite exciting to see it start to come together! Program Control and I/O are still just thoughts brewing in the old brain case. Once memory gets underway, I'll start diving much more heavily in program control. Germanium diodes were quite commonly used in tube computers in the late 50s, but the 6AL5 is also a really brilliant little dual diode that would work perfectly for the type of logic gates I'm building here. I've actually built test pieces using the 6AL5 and they do work super well, the only problem is the number of tubes would be doubled!

@@UsagiElectric need to get this series on Hacker News and EEVblog and such; you deserve so much more visibility to this work!

It depends a lot on the specific code! The MC14500 needs a lot of programming gymnastics to get some stuff done that the UE14500 can do in just one or two instructions. And conversely, the MC14500 can do some trick stuff with some of its logic instructions that the UE14500 can't do. But, with an intimate knowledge of both, you could totally port programs over!

these tubes are suited for digital. Early ones were not and failed fast

Depends on a lot of factors, sometimes the filaments break and the tube dies immediately, sometimes a short can develop and it starts acting erratic, and sometimes they just slowly lose emissions over time.

As Ignacy said, it depends on a lot of stuff, so all I can really offer is anecdotal evidence. In my experience, the way in which I'm using the tubes is very understressed (even the filaments are under-volted) and their total uptime per day is very low, so once I know I have a good tube, I fully expect them to last the life time of the machine. The primary factor that might contribute to tube death in my application would most likely be my clumsy, dumb self knocking into them and breaking them physically. There's a lot of electrical factors that can contribute to a tube losing performance. Things like cathode poisoning or stripping, flakes of the anode causing a short, the filament breaking, etc. Shorts and filament death are hard to predict, but cathode issues present themselves in weaker and weaker emission. On my low voltage tube tester I made a few episodes back, that would show up with the tube not getting all the way to saturation. Any 6AU6 that only makes it down to about 8V to 6V is tired, and any 6AU6 that doesn't make it below 10V is so tired it gets replaced.

@@UsagiElectric I was thinking that when you run your arduino test for example, that its not a binary yes/no, maybe there woudl be some way to predict .. "this one will go bad sometime, vbut is okay for a few months" but that sounds out of scope for hobby work :)

Something like that could be possible, but it doesn't work well with the current planned direction for program control. My current plan is to use magnetic tape, and have the clock come directly from the tape. The instruction word will be read serially from the tape and shifted into a shift register on the UE14500, then a parallel read of that shift register and global clock pulse will execute the instruction. Unfortunately, with a ROM, I would have to do a ton of work to either get it to output data in a manner the UE14500 can read, or I would need to change the UE14500 design, so it's just not really a viable method of reading really complex programs. But, magnetic tape is just an audio signal. And it's an audio signal I will generate on my PC initially, and then record to audio tape. What this means though, is that really complex or massively long programs absolutely can be saved as MP3s. Then, with just a really minor amount of circuitry, something like a cellphone could be plugged into an AUX jack on the UE14500 and the MP3 played back! That's actually going to be a pretty essential part of the whole development process anyways. It'll be way easier to write and test programs with an audio file stored on an old cell phone than trying to record and lace up magnetic tape everytime. Once I've got some programs that are good demonstrations, I'll record them to audio tape, loop it up, and sit back and let it run on its own.

Don't sell yourself short, working on the old HP3000 computers is absolutely amazing! That requires a level of skill I certainly don't posses, though it would totally not stop me from poking around in side them, haha. I would absolutely love to get my hands on an early HP Minicomputer someday. I'm a bit of an HP fan anyways (especially of their really early tube stuff), so even seeing an HP3000 in person would be really special!

or tape of the non-magnetic variety *fingers crossed*

I've got a few ideas percolating in the old brain case! The primary goal is to minimize the circuitry necessary on the UE14500 itself, but that's going to come at the expense of speed, which is a sacrifice I'm alright with making. Once I get further along with memory, I'll start testing different ideas to see how viable they're going to be! And, it's definitely not lightweight! Though, it doesn't weigh nearly as much as that HP oscilloscope!

@@zbradbell Unfortunately, I don't think paper tape is in the cards. Magnetic tape I think will be a much more manageable path, but we're talking pretty far in the future. My plans two months ago for memory were dramatically different than my plans for memory today, so things tend to change pretty rapidly on this build!

@@UsagiElectric With regards to tape, I was thinking one channel could be for the clock, and the other for data. (is it still one bit ? I can't remember.) It might get heavier, It has no transformer or rectifier. (yet. LOL)

higher=faster like transistors

This is actually a test I want to do sometime in the future! There's a lot of factors to consider though. In this application, I'm pushing the tubes to full saturation and full cutoff. As the frequency gets higher, the tube has a harder time getting all the way down to saturation. More voltage helps accelerate the electrons faster, but even that has a limit. However, with careful design, a tube doesn't have to go all the way to full saturation. Someone much smarter than me could design logic high and logic low to be right in the middle part of the tubes curve, which means you can push them a whole lot faster before you start to run into problems. It also depends really heavily on the type of tube. Triodes are terrible at getting to saturation because the more electrons that hit the plate/anode, the lower the plate's potential is, which has less of an attraction for the electrons. That means that the closer it gets to saturation, the harder it is for it to get there. Pentodes on the other hand have a screen grid that acts as a sort of electron accelerator, keeping that attraction for the electrons regardless of the plate's potential. So, max frequency on a pentode can be quite different than that of a triode. I think that could definitely make an interesting video in the future! I just need to get my HV power supply up and going to test them properly!