Vacuum Tube Computer P.29 - Addressing and Fanout Problems

[@e7yu]

Yes, more vacuum tube computer content. 😸

[@UsagiElectric]

Winter is coming, so I need something to heat the room!

[@凱北田]

やっぱり、うさぎ電気さんみたいなKZbinr増えてほしい

[@JCWren]

This is too funny. I just started watching your channel recently, and hadn't seen any of the VTC videos yet. Watching this one, and I see the instruction set flash by at 3:45, and immediately went "Hey, that's the same as the MC14500B!" I had a mentor that helped me out building a MC14500B system out of the handbook back in the late 70's (probably 1979, specifically).

[@UsagiElectric]

Nice catch! It's not actually the same instruction set as the MC14500 though. Initially, I was building the processor as a tube replica of the MC14500, but mid-way through I got upset with the fact that Motorola didn't include a proper ALU, so I did a bit of rework and slipped in a carry register and some proper arithmetic functions all within the original 4-bit instruction style, and now I've been dubbing it the UE14500. I love the little MC14500 chip, I built that same ICU system from the handbook not long ago on a breadboard! (kzbin.info/www/bejne/pYGkaZd-qcuSbK8)

[@herbertsusmann986]

This brings to mind the famous quote from Mr. Spock: "I am attempting to build a mnemonic memory device using stone knives and bear skins"

[@cseeley]

I was thinking exactly the same thing :) kzbin.info/www/bejne/a6aQmH-IaMqngrs

[@UsagiElectric]

Spock needed way more tubes than he actually had in that scene!

[@ducksonplays4190]

@@UsagiElectric I have a theory that Spock was the rest of the tubes. (using his brain as most of the computer)

[@metatechnologist]

Spock knew things that we don't. Fwiw, I was thinking about that very device and how we need to interface our cell phone with this tube computer!!

[@GeomancerHT]

Thank a million for the Celsius temps!

[@NightRidersUrbex]

He had me already jumping joyfully after millimeters ;-)...

[@odinthorson5751]

6cg7 is a really good tube. Works great as a phase inverter/splitter or gain stage for audio. Nice to see them in your design

[@LoneWolf0648]

i was wondering when you were going to get back to this. its amazing how much space a little memory can take up

[@jeromethiel4323]

Indeed. Memory was the constraining factor for quite a while. Which is why when you go back, there were some really odd memory devices, because memory was critical, and making it with gates tended to be space and power hungry.

[@the123king]

@@jeromethiel4323 Memory devices are often some of the weirdest and novel of devices in a pre-IC computer. Delay-line, drum, core, Williams Tube etc for RAM. Core rope, transformer, diode matrix for ROM... Memory devices are often very strange.

[@UsagiElectric]

It's good to get back to work on it! Memory is easily one of the hardest aspects of computing to build. Most homebrews, even wild discrete transistor ones, use a chip for memory because it's just so painful to build otherwise. The memory I'm building may be easier to think about as being six 8-bit registers that the processor can store information in as needed. My construction method scales really poorly, which is why it's tube heavy, but if I wanted more RAM storage, I would need to start looking into stuff that scales really well like core memory. But, you're right, most CPU registers are built using similar logic setups and it's wild how intense that construction style can be when building with tubes!

[@jeromethiel4323]

@@UsagiElectric Indeed! And i don't think anybody would fault you for using even modest amount of semiconductor memory. I would not recommend going down the rabbit hole or delay line memory or magnetic core memory. Although those are cool retro memory types, the amount of circuitry required with vacuum tubes would be insane. Your vacuum tube computer is more in line with early computer types, and i think you should stick with what YOU want to accomplish with this thing you have made. It is magnificent in it's own way. And kudos for you for even attempting it, much less succeeding so far. And i have no doubt you will end up with something you can be proud of, and may even end up in a museum at some point.

[@KingSlimjeezy]

somebody mentioned you in the comments of Dave's Garage and I had to comment Usagi is the type of guy I want to be like when i am older :)

[@georgegonzalez2476]

Cool project! I wonder why you didn't go with the 6DJ8 instead of the 6CG7 as the cathode-follower. The 6DJ8 has like twice the transconductance which might be helpful. Another possibility might be to go to much higher resistors to the grids of the 6AU6's. If the tubes are not gassy you can probably go up to like 2.2 megohms or so, lowering the DC loading. The slew rate will be slower but you probably are not seeking very high clock speeds. FYI if you think you have logic swing issues, Seymour Cray's mainframe computers had to work with like only 300 millivolts of logic swing, and they worked surprisingly well with that. It required the use of around 35,000 miniature coax cables between the gates, all of them 6 feet long. Regards!

[@UsagiElectric]

Excellent question! And the answer is... I already had a bunch of 6CG7s from random lots, haha. The 6DJ8 is a brilliant tube and even at low voltages it's one of the best out there, but they're also jolly expensive and I only have a few of them in my collection. However, the 6CG7 is pin compatible with the 6DJ8, so if I find I need a little more oomph, I can always swap out later. Interestingly, I did try going higher resistance on the grids when I first started building my tube circuits. And I don't know why exactly, but with a 1M/1M split, I couldn't get the tube to go into cutoff or saturation. My current working theory is that I actually need to overdrive the grid pretty heavily to get the tube all the way into saturation with only 24V as B+, and the 220k/220k split seemed to be the best compromise for that. Seymour Cray was an absolute legend! 300 mV of swing is ridiculous, but that dude not only nailed it, he got it working so well he sold the systems to customers who pushed that thing hard. What an absolute beast of machine!

[@adamfisher8769]

aAa1

[@xephorce]

I hope to see this computer communicate with the centurion someday.

[@kevinmerrell9952]

YES! Transmit "HELLORD!"

[@UsagiElectric]

That is a long term goal actually! There are four boards planned for the tube computer: processor (done), memory (in-progress), program control (in-planning), and serial input/output. One of the main goal with that last board is to have it transmit and receive 7-bit ascii at 300 baud, which means we could totally send commands to the Centurion!

[@xephorce]

@@UsagiElectric I cant wait to see that day. I love seeing old tech given a new life. Wonder what kind of custom cards could be created for the Centurion? I wonder what capabilities could be added. I don't expected miracles but would be cool to see something like Pacman or a space invaders type game. I know its not meant for that but hey I am a gamer at heart. Thank you for the great entertainment with your vids

[@StevenIngram]

Wow, that's going to take a lot of work. Lol

[@thebiggerbyte5991]

Vintage tech, new projects and animals - a great recipe for me.

[@adrianotrindade7925]

If it's hot, you can mill at night, no? Start milling and go to sleep. Next day it's done. Love your videos on this vacuum tube computer.

[@Blacklab412294]

Q: about power? > have you thought about how you are going to drive your memory by? Your 1-bit computer is already demanding quit a lot of current, are you going to drive every thing by "one" (1- 12v & 1-24v) Set of power supplies? Or are you going to try to break it up in to one set (1-12v & 1-24v) for the CPU and one set (1-12v & 1-24v) for the memory? have you also thought about how you are going to reset MEM to a known value (for a Reset)? and have you thought about how you are going to make the counter and how you're going to set the counter (for things like jumps, and sub Routines) ? Meaning are you thinking of using a main counter, and a sub routine counter?) Or just one counter ? You already have cut down on how much memory you're going to need by putting in a full ALU unit in the CPU, BUt by having a separate main MEM counter from a Sub routine counter would that not also cut down for the need of a bigger memory? Right now every time you address memory that will take 6 bits to have to store in the counter. Also down the line... Are you going to have to make a 6 bit serial to parallel converter (to load the counters) ? In my ignoance, I am still thinking your going to need more memory, but while I am thinking about that I'm still wondering if Modulizing your memory to smaller packets might be faster? Can your router be run at night or does it need to be Supervised?

[@Blacklab412294]

Running it at night would be a lot cooler. And having some fans burning in cooler air into the building would help cool thins down too. We had to do that for work here in the NW when it got over 80 sets outside, and it was over 90+ degrees F inside.

[@Kikay0n]

This is so great! This is literally how I found you channel.

[@UsagiElectric]

Thank you so much!

[@StevenIngram]

I was thinking about this build up to RAM and wondered if you'd ever looked at IBM's "pluggable units?"

[@jrkorman]

Love it when troubleshooting is that easy! Roughly where in Texas are you? We live about an hour north of Abilene. I agree it has been brutal this past summer and this fall isn't looking much better as far as rainfall.

[@UsagiElectric]

Troubleshooting almost never goes that easy, so it was a huge relief that it only took a few minutes to figure out! We live about an hour and change south of Fort Worth, and it's been dry as all get out this year. We have three ponds on the property and one has gone completely dry, the second is not far from it and the third is still hanging on, but if we don't get rain soon, it's going to be tough.

[@GodmanchesterGoblin]

Just a comment about cathode followers... They have great rise time, but the fall time is dependent on the load at the cathode, which in your case could be the capacitance of a large number of inputs in parallel with a relatively high cathode resistor. Before you go too far with the next stage I recommend trying the circuit with a clock signal and adding a bunch of capacitance to ground - maybe a few hundred pico-farads. If you have the voltage headroom (you don't) then a mu-follower with some DC level adjustment would possibly be a better choice.

[@UsagiElectric]

That's actually really excellent advice! I have tested this cathode follower design at different speeds before, but for this particular tube computer build it's a non-issue because the clock speed is ultimately going to be mind-numbingly slow. Which is due to how I'll be building out program control - in short, I'll be executing the program directly from tape, so I'm aiming for a global clock speed of 100Hz max.

[@GodmanchesterGoblin]

@@UsagiElectric Cool! Probably you'll be fine, but just something to keep an eye on.

[@eDoc2020]

I don't think the fan-out is as bad as you are expecting. When more than one input to your NOR gate is high it will effectively draw half the current from each.

[@metatechnologist]

Adding that buffer is gonna slow down your computer!! 🤣🤣

[@PeetHobby]

Put fan on heat sink of you mill and all other places, heatsink is outside the encloser so that is an easy job. :D

[@RinoaL]

It's been the kind of summer where you come into the shop to find your solder melted on the floor. Or so it seems haha

[@UsagiElectric]

It has been brutal this year!

[@reinoud6377]

Is aggressive isolation an idea? Like a grass roof and extra reflecting coatings/isolation?

[@nickolaswilliams935]

If you want the ultimate in current and low plate resistance for your cathode follower, grab a 6080.

[@bobsbits5357]

hi my dad worked for the met office in bracknell uk you lot better get used to the hot weather it's going to get a lot more heat going buy the files he was working on there was alot of paper work back in the days and could see what is going on computers came alot and all the eye's could not see it any more love te video's

[@wtechboy18]

vacuums for days, nice. Is it concerning at all that the yellow LEDs don't fully power down when you flip the corresponding switches? it's like your "off" state isn't entirely "off"

[@p_mouse8676]

Instead of the 6cg7 it's also possible to use a 12au7 instead. Which works fine on 6V heater. (Heaters in parallel) It's a much smaller tube and much easier to get by. What frequency is the clock running again?

[@UsagiElectric]

It totally could have been possible to use a 12AU7 instead, and the 12AU7 is a lovely little tube. The reason I chose 6CG7s is that I already had a lot in my collection from random tube lots over the years. Unfortunately, with how I'm running the heater lines (four tubes in series), I can't swap in the 12AU7 now without having to redesign the PCB. I can however switch to any other pin compatible dual triode (6DJ8, 6BZ7, etc.). The clock speed hasn't been decided yet because the plan is to actually read the program directly off magnetic tape and execute it there without reading it into memory somewhere. So, the clock speed will be dictated by how quickly we can read the program off the tape, which means it'll be properly slow.

[@neilbarnes3557]

Is there scope for using a double triode as two in-series active inverters? Does this provide sufficient improved drive current for the fanout? Or is it too long since I did my electronics training (yes, it was pre-microprocessor days, and did include various valve calculations, but it was a long time ago in a far country etc).

[@UsagiElectric]

Being in series wouldn't help with fanout, but it would reinvert the signal so if I needed to amplify an input signal without inverting it, that's a possibility. There's a lot of ways to tackle fanout that I probably could have done better (like running a smaller plate resistor, or larger grid resistors), but anything worth doing is worth overdoing. I didn't just want to skirt by, I wanted my buffers to be able to supply enough current for at least double the amount they needed. Totally unnecessary, but now I know they aren't the weak point!

[@neilbarnes3557]

@@UsagiElectric Yes, I guess the problem with using valves and the implicit anode resistor is that the resistor is always going to limit the current unless you can use a cathode follower, as you have... for size comparison, I've just put a replica of the 8080 ALU (largely following the original design as revealed by Ken Shirriff) using 74HC in SOIC packages onto a board about five inches by four! :D

[@ropersonline]

22:38: Do you expect me to meow? _No, Mister Neko, I expect you to lie._

[@edgeeffect]

WOW! That schematic of the whole memory is FEROCIOUS! Is that drawn by hand or with computer assistance? Either way, it looks just a bit more than a 10 minute job! ;) Before watching this episode I sort-of more-or-less understood how transistor inverters worked but after watching this episode I REALLY get how transistor inverters work. It's funny how I keep learning "transistor stuff" from your videos, y'know considering............

[@senilyDeluxe]

now if he made core memory for that computer, that would be FERROCIOUS (notice the two Rs lol). (sorry, none of my puns are any good)

[@Flying0Dismount]

Get a swivel knife blade for your CNC and cut vinyl masks and then etch your PCBs... Or do toner transfer or pretty much ANY other method of PCB manufacture vs direct CNC milling and you'll cut down your production time to less than an hour...

[@mfbfreak]

If you want to make boards in summer time, could it be an option to fit the CNC machine with a circuit board marker or other etch resist applier, and etching the board? Just draw it on there like if it's a particularly expensive plotter, then etch it. Those temperatures are ideal for etching boards. Don't even need to heat the bath.

[@StevenIngram]

That is a sweet looking board.

[@UsagiElectric]

Thanks!

[@stinchjack]

Yes, more quality cat content

[@UsagiElectric]

They're not kittens anymore, but they're arguably even softer and more adorable now!

[@davemould4638]

Instinctively I'm pretty sure the logic could be implemented with fewer tubes. Also the 220k grid pulldown to -ve bias seems very low and is unnecessarily loading the previous output, reducing the fanout. Grid leak resistors are typically 1M to 10M, and with such a high bias voltage (-12V) I should think you could use a 10M resistor. Vacuum tubes are voltage operated devices and should not need any significant current on their inputs.

[@yorgle]

It's always a great day when you post a video! :D

[@UsagiElectric]

Thank you!

[@ChefHeisca]

I ask, you deliver! Sir you are a champ.

[@rickhole]

I am just getting hooked by your VTC computer series. My training 1968-72 was all solid state, TTL. We had RS flip-flops, you have SR. Did the terminology change from vacuum tube days? Obviously the logic function is the same, just the initials in the opposite order. Your choice of tubes, my, they are all my good friends from youth!

[@jeromethiel4323]

Quite often with SR or RS flip flops, the order is which takes precedence. For example, if you turn on both inputs to a RS FF, the output should be logical 0, as the reset is the dominant input. The reverse would be true for a SR FF. Although this convention was not always followed, so your mileage may vary.

[@rickhole]

@@jeromethiel4323 Thanks. I had never used the R=1 S=1 "indeterminant state so it never came up. So both SR and RS are valid, and which one is determined mainly by whether the circuit is implemented by NAND or NOR gates.

[@jeromethiel4323]

@@rickhole If you are using an IC based FF, then reading the data sheet is your friend. There are conditions where the S and R terminals may be simultaneously active, and you need to know how the device is going to respond. I have seen some (i believe it was 74 series TTL chips) that the output state was listed as "indeterminate" when that condition existed. But good designs are usually made so that the output is always deterministic. I, for one, never want to roll the dice on how a chip is going to respond to weird input states.

[@NielsNL68]

I'm sure you thought about this already, could you not use one register divider and hook that output trough the 4.7K registers on pin 1 of the tubes?

[@UsagiElectric]

I'm not quite sure I'm fully understanding your idea here. If you could into a bit more detail, I'd be happy to either give it a try or let you know how it works!

[@NielsNL68]

@@UsagiElectric i was looking to the image at #9:58 where you added so many register dividers in a row. I was thinking when you added the divider before all the switches and connect them to the middle point of the divider, what would be happen. But looking again trough your video again, i can't find how that image is related to the real schematic. Also, please take my suggestion with a little salt, i'm in no where experienced as you are with this.

[@kitsuneneko2567]

Whereabouts in Texas are you?

[@winstonsmith478]

Are ganged triodes ever used for audio circuits to allow smaller tubes to provide more output power?

[@UsagiElectric]

That... I don't actually know. I'm a total noob when it comes to audio circuits, so hopefully someone else can chime in. Excellent question though!

[@eDoc2020]

Probably yes but paralleled pentodes are more common.

[@TeslaTales59]

Nice work sir! - what do you do for a living?

[@DAVIDGREGORYKERR]

What about water cooling the CNC System unit so that it can be used even when it is very hot.

[@rickhole]

After days of binging I have seen episodes 1-29. I read the code for the assembler and want to get more involved. The assembler thus far generates something like micro-code, that is, just the instruction codes. I am interested in going farther with assembler and emulator that will work with the 12-bit instructions so that executable code can be written and tested, ready for UEVTC. In other words, taking the UE14500 assembler and emulator to the UEVTC assembler and emulator. I have joined the discord but don't see an appropriate place to begin this conversation.

[@danducky4741]

I LOVE THESE VIDEOS NEVER STOP

[@UsagiElectric]

Thank you so much!

[@der.Schtefan]

IMPORTANT: It looks like your leftmost LED is not really switching cleanly between on and off. It seems the yellow one never really goes off while the red one is fainter than the rest. Are you sure the levels are clean and there is no leakage/small short?

[@OscarSommerbo]

It is most likely due to the logic low signal voltage is above zero. I think he said around 6 volts for logic low.

[@andydelle4509]

On the boards where you need a number of the same type, have you considered one of the Chinese board services? Low cost and good quality. I Use PCB Way.

[@nonoyorbusness]

Bletchley Park must be informed about this, it could win WW2.

[@simona625]

Don't know if you'll get this before a tweet reply I sent today. But I sent you a valve circuit I would love to see built. It may be good for one bit memory. It uses a dual triode, but I'm sure you could make it easily with 6AU6s

[@mikegLXIVMM]

Great video! Would it be hard to do core memory?

[@UsagiElectric]

Thank you! I actually talk about this a bit more in a previous video (kzbin.info/www/bejne/sKvLlGp8gryAe6s), but core memory is very, very difficult to build, however it scales wonderfully well. So, for a small amount of memory it's really not a great idea, but if I were going for 1k or more it would be perfect. For this project it's not really in the cards as that starts to fall outside my initial design goals, but for a future tube computer project, I would absolutely love to do core memory!

[@bobsbits5357]

you know the old star trek one the one with all the tubes in it all a round the room funny in away how small things are now ic's came in

[@janikarkkainen3904]

"...and then he did a backflip, snapped the bad guys neck and saved the day"

[@spokehedz]

Hello, late to the party so please forgive me if this has been answered before: But that mill is gigantic? And, also, since you are going to be making so many PCB why not have a dedicated PCB cutting machine?

[@rickhole]

Love those old Bridgeport mills. I worked for Velocity aircraft and used one there occasionally but without the CNC mod.

[@grenouille-de-geek]

the memory look like a bus based factory in factorio

[@douro20]

Four hours! What names did you give your cats? Just don't name one Neko because that would be silly...it would be like naming a rabbit Usagi or a dog Inu (there's a pit bull on KZbin named that)

[@UsagiElectric]

It was a long cut and a long solder job! The cats are name Jun, Ten and Sue. Sue is the brown one and when we got them, he was so little and frail we didn't think he was going to make it, so I named him Sue after "A boy named Sue" saying that he needed "get tough or die," and he got tough! Jun is short for Junko, a Japanese girls name that I quite like. And Ten is just Japanese for "spots" since Ten is the spotted one.

[@douro20]

@@UsagiElectric I thought you said four hours just to cut it.

[@acoustic61]

Great stuff! I'm wondering if you can you build a vacuum tube computer for something like a cell phone with 8 cores and 4GB of RAM?

[@tissuepaper9962]

The RAM board he showed in this video is not even one byte, 4GB would take up a small nation's worth of space

[@TheGunnarRoxen]

Good job!

[@EdgarsLS]

why use pentodes? you could've cut the tube count in half by using dual-triodes, and you'd also have lower output impedance to cause less problems. there are also soviet micro-tubes which are really cheap, and a bit smaller, if you can get ahold of them.

[@diobrando2160]

dual triodes cost a lot

[@Spookieham]

Always end with kittens👍

[@SirHackaL0t.]

Gotta love that ‘it’s not global warming’ doom and gloom.

[@Orxenhorf]

You might want to consider using one of the hobbyist PCB manufacturing companies out there. Just for the time savings.

[@MarcelHuguenin]

This is why his boards are so unique, they have a very pleasing aesthetic quality. It also fits the retro look and feel better compared to modern boards.

[@frankowalker4662]

That would defeat the purpose of compleatly home built and ruin the looks of the magnificent beast. :)

[@sashakoshka]

wow, what a massive eight bits. thats not a byte thats a whole meal

[@jasongrim2027]

Hi dude

[@christiantreldal1585]

Thanks for a lot of good videos. Today I vas a litle disapointed only to see a timelapse of your mill. It could be very interesting to have a video about your mill and the tooling you are using. I am tinking about using a mill for pcb prototyping, so it would be very nice to learn from your experiences. Best whishes, and keep up the good work. Chris OZ1GNN

[@FrancSchiphorst]

wow wow wow! ;)

[@bvanb2011]

Hey, you're leaving everyone hanging, waiting for the last wow! :-D

[@FrancSchiphorst]

@@bvanb2011 wow!

[@kitsuneneko2567]

兎はどこですか？

[@belstar1128]

When i am 50 i want to be like you.

[@frankowalker4662]

Don't you just love multiplexing ? LOL. Daft kitties. (as bad as mine. Ha ha)

[@68hoffman]

kool

[@bzuidgeest]

Why not have the PCB made by a fab? You already proved you can do it yourselves. A PCB house is a lot faster then quitting for the summer because the texan legislature hasn't discovered climate change yet.

[@crasbee]

I think he explained it in one of the previous episodes, but having all those PCBs made would be incredibly expensive. The PCBs are also part of the design, since the look of the computer is a more important aspect than the density.

[@bzuidgeest]

@@crasbee all true, but he complained about the wait and with the pro option you don't have to wait. And you do get cool options for silk screen. So you can make instructions and annotations on the PCB 's. I don't think expense is a factor. You don't build tube computers if you worry about cost.

[@EdgarsLS]

I have to say those PCB layouts are really bad, you could've made them much more compact and save manufacture time. I'd suggest looking at hundreds of tube appliance designs and learning the good features of each, in my opinion that's the best way to learn good layout practices!

[@turbinegraphics16]

First 🤣

[@nmosfet5797]

Are you trying to build a vacuum tube based computer, or a transistor based computer just with transistors replaced with vacuum tubes? For example, the british managed to build a vacuum tube based memory just after WW2, with 2560 bits of memory in a single vacuum tube. en.wikipedia.org/wiki/Williams_tube