Vacuum Tube Computer P.18 - ALU Installed and Tested 

Thank you so much! I was thinking about using Ferrite memory, but for the relatively tiny amount of memory I'm planning on implementing here, I think the read/write circuitry for it would be a little too intense. I have a few ideas I'm working on though, and ultimately, I would be very happy with 256-bits of memory, but I think I'll probably end up around 128-bits or so. Given how long programs will have to be in order to use all of that though, I think that's probably a pretty reasonable amount!

Magnetic drum or even delay line memory may be more period-appropriate :) 64 microsecond ultrasound monolithic delay line units were used in PAL TVs and VCRs, and are still somehow available

I was thinking of a Williams tube memory :-)

@@sashimanu A drum is a challenge on its own but I would _love_ to see it. Another option would be to weave some core rope memory for the code. The working memory is the hard bit.

Thank you so much! I'm really happy with how it's coming together!

Thank you so much! If you look closely, you can see there's already a bunch of drilled holes in the wood before I start. That was my first try at trying to get the Instruction Register and Decoder boards mounted. I totally flubbed it, haha. So, I flipped the whole backboard upside down and had a good long think before I came up with the template idea. Fortunately, that worked like gangbusters!

@@UsagiElectric Ha ha, Nice one. I'm loving this series so much.

@thomas.j: agreed. Gives a whole to meaning to the term heater in the tube.

I'll have the hottest calculator in the world!

Thank you! And thanks for checking the channel out!

Thank you! And you're actually not far off on my intended path here! I actually recently purchased some nice big IV-12s for use on the memory side of things. I'm still working on how to build the memory up, but I for certain want an 8-bit output register with those big IV-12s displaying the contents. (And ultimately, that 8-bit output register will hook up to the I/O board which can then send the value in that register over a current loop to a teletype to print stuff). I still haven't figured out the best way to handle the program counter and program storage section just yet. The 1-bit architecture with IEN and OEN allows for some interesting methods to perform loops that should make building program control a lot easier, but that's something I'll tackle after I get memory all taken care of!

@@UsagiElectric I can‘t wait to see it! Thanks for building this wonderful machine and sharing it with all of us!

Thanks for checking the project out! I never received any official training or education on tubes, just what I could learn from reading a bunch of old documents and experimentation. So I probably got quite a lot of stuff wrong, but I'm sure having fun playing with them!

Thank you! And very hot indeed! Though, the 6AU6 is a pretty cool running tube, especially compared to some older octal tubes like the 6V6. But, the sheer number of them that I have going means this thing puts out some serious heat!

Haha, thank you! OCRemix is a treasure trove of awesome music!

Thank you so much! In order to keep things as simple as possible (so the entire thing doesn't end up filling up the entire room), there's not actually going to be any microinstruction stuff here. Instead, the program counter will read the instruction word into a register, and the instruction word will have the 4-bit processor instruction and a 6- or 7-bit memory address and that's it! Though, even just implementing that is going to be an intense amount of work!

@@UsagiElectric awwww..... Now, to be more realistic, consider how much memory a minimum program uses (for example, calculate fibonacci, or a division) and would set it as the minimum amount of memory. "6 or 7 bits memory address" is an order of magnitude, more or less! ... plus the jump direction register, the return register, maybe some jump decision mechanism (unconditional, carry, zero) If you want I could upload a sketch to a drive of how I have it implemented, but if you are similar to me (only in the electronic aspect, heh) I would not want influences on how to develop ... it tends to deviate and create chaos. just say. I never tire of saying it, Good job!! (sorry for my "google english")

Thank you so much! I needed a way to get a back light behind the paper, and the window seemed the easiest! Unfortunately, it was pretty overcast that day, so it was a little hard to get it just right, but in the end it all seems to have lined up just right!

Thank you! I would love to use Nixies, but I just don't have the voltage for it. My goal with this build is to keep the entire supply voltage at just +24V and -12V, which isn't quite enough to light off a Nixie tube. But, I do have some very beautiful IV-12 VFD tubes I plan on using!

Thank you! I'm really happy with how it's coming together! Time to get my butt back out to the garage and cut up some more boards!

Thank you! If you look closely, you can see a bunch of holes already in the wood before I start drilling. Those are from my first try at drilling holes, and I totally flubbed it, haha. I flipped the back board upside down and had a good long think about how best to do it, and fortunately, this method seems to work pretty great!

Thank you so much! I'm having a ton of fun building this thing up and I've learned so much about the fundamentals of computing!

Thank you so much! I'm quite happy with how it's shaping up, I'm getting really excited to finish off the processor and get into the memory!

Thank you! I'm super happy with how it's shaping up!

Thank you so much! I'm really happy with how it's coming together!

There's not many of us insane enough to try it, but boy it sure is fun! Are you working on any tube logic projects? I love checking other's work in this weird field!

@@UsagiElectric I made some 4 bit relay calculator on the similar logic architecture as your project i made a 4 bit vaccum calculator but i use double triode tube (6BQ7A) with relay memory in order to save the maximum vaccum tube, I will now work the display with nixie tubes

@@linushyky476 That's awesome! If you've got any links to pictures or anything of your projects, I'd love to see them! Relay memory is a smart choice, I'm working through memory design at the moment now and it's really hard to balance memory size with physical size.

@@UsagiElectric In order to keep in mind the relay is a good choice and easy to implement if you want to see my modest projects on Pinterest france. Thank you

Thank you so much!

Thank you!

Thank you so much!

Thank you so much!

Thank you so much!

It should work perfectly with little submini pentodes! I would have loved to build it with submini tubes, but they are really expensive comparatively, so I sacrificed physical space for affordability. Interestingly, the 24V designs I'm using here could be transitioned over to NFETs with almost no changes at all! Changing the whole design to SMD FETs and passives could really make it very compact.

Thank you very much!

Thank you so much! You're not far off on your calculation! The majority of the heaters are actually powered off the primary 24V B+ line (with four in series), so the current requirements are a bit less. But, I'm fully expecting the +24V line to pull between 5A to 10A, and the -12V line to pull around 5A to 7A, which is a lot of juice for something so simple!

It's a slow but steady march! I'm starting to get amped up now that I've managed to clear the most difficult part of the processor build. I'm headed back out to the garage tomorrow to cut some more boards!

Thank you so much! For the 24V supply, I'm using a Mean Well 24V 15A supply, and that's plenty powerful enough. For the -12V supply, I was just using a little 12V 3A wall wart that I stuffed inside my wooden box. The 12V supply is going to need an upgrade though, so I'm going to bump it up to a Mean Well supply too.

Dekatrons would be awesome! Unfortunately, one of the (totally arbitrary) design limitations I set for myself for the whole project is a low B+ voltage of just +24V, and that's just not quite enough to ignite a Dekatron. I do really want to use some Dekatrons for a future project though, because they look epic!

Thank you so much! I'm really happy with how it's coming together!

Thank you! Memory is a tough one. Because it's a full 1-bit architecture, not a serial architecture, I can only really access a single bit a time. In a traditional 8-bit CPU, one address gives you 8-bits in return, so an 8-bit address can access 2-kilobytes of memory. But, with the 1-bit architecture, one address only accesses one-bit, so an 8-bit address can only access 256-bits of memory, or 32-bytes in total. Still, given how long programs would have to be fill all of that in, that's actually a pretty substantial amount I think! So, the question is, how to build it? And I don't quite have an answer for that yet, haha. I'm thinking about 24-bits will be vacuum tube based registers, and then another 8-bits will be physical switched input, and another 4-bits will be reserved for special stuff (RR loopback, inverse RR loopback, low, high). If I want to use an 8-bit address, that leaves 220 bits left to build (if I use a 7-bit address, that knocks us down to 92-bits). I need to come up with a good solution, and right now I'm leaning towards using VFDs and LDRs, but the control for that is still giving me some grief that I have to work out. Here's a basic proof of concept though: i.imgur.com/OgxDr0M.mp4

@@UsagiElectric how about magnetic core memory?

@@primus711 On the Discord we've been talking a lot about how to take care of the memory, and the problem we're running into now is coming down to physical size. The address decoding and select circuitry for each individual bit is physically relatively large, regardless of what kind of memory cell I end up using. I need to limit the size of the memory board to 1,000mm x 700mm (that's the current size of the processor back board) to make it fit in the total design. That means I'm running into all sorts of issues before even getting to the type of memory cell I use. But, we're slowly working out the kinks and I think I'll get at least 64-bits (8 bytes) of memory, maybe more (I would like 128-bits or 16 bytes, but I don't know if that's doable). I do want to do something with core memory in the future, and I have an idea for a project that is much better suited to it!

Thank you! I'm getting really amped up about it now that it's starting to come together!

I actually discovered LSD's covers through Samuel Ascher-Weiss, I love his piano work!

I'm digging the pun! And I'm really happy with how its coming together!

I've often thought about using an Arduino to test functions out, and as it grows in complexity, that may actually be necessary, but for the time being, I actually enjoy keying things in by hand!

@@UsagiElectric I vote for a Raspberry Pi with Discord interface, so we can execute programs remotely :D A microcontroller would be a great way to check the maximum frequency too 🤔

Thank you so much!

Thank you! That's actually a really good idea! My ultimate goals with the project are to calculate the Fibonacci sequence and to print something on a Teletype terminal. A traffic light controller could be a very good demonstration tool as well though!

Thank you! Interestingly, the PCB layout design that I built on the computer (and you can see briefly at 19.45) is my phone wallpaper. Unfortunately, it's an old revision, so I know where all the mistakes are on it, haha. As the build grows in complexity, thinking of a good, quick way to ensure that all the functions are working appropriately is definitely going to become more necessary. At the moment, it's a little difficult because the only actual feedback for stuff is coming via the VFD indicators because a lot of signals are still waiting on the rest of the processor to be built. But, once I get the last part in place, I'll have a ton of pins I can connect up to on the bottom and debugging should be a lot simpler.

Haha, you got me, that -12V rail finally crept up on me. I've got a new 12V PSU sitting on the floor ready for me to build a new wood enclosure for it! Thanks for checking the videos out!

That's a lot of exclamation marks!!!!! I'm glad you're digging the videos!

Thank you very much! I actually did think about using a serial architecture like the PDP-8/S, but honestly, it came down to my ability to wrap my head around the architecture. While the MC14500 I used for inspiration is hilariously weak in comparison, it was an architecture design I could follow and visualize in my head, which made life a lot easier. As I learn more about these old computing techniques, other, more complex designs are starting to make a bit more sense though, so in the future, I would definitely like to build another tube computer that's far more capable! I actually am using little silicon diodes here! They're not exactly Germanium diodes, they're little Schottky diodes, but they perform quite similarly to the Germanium diodes that were commonly used. I actually build my NOR gates by using two diodes to make an OR gate, and then a single 6AU6 to invert and amplify it for a clean logic low and logic high. I also did a ton of testing of different tubes at super low voltages to see what would be the best choice. With smaller, cooler tubes like the 6AK5 or 9003, there just wasn't quite enough emission to get the tube working right. The hotter the tube and the larger the plate, the better the tube seems to perform with just +24V on the plate. The 6AU6 seemed to work really well, and they were an extremely common tube, so I was able to find them in large lots for really cheap!

@@UsagiElectric great that you investigated tubes at low voltages. The data sheets were never very clear what happened down there. I am a tube nut too. I have a dozen or two 28D7’s which are low voltage dual power pentodes. Totally impractical due to their heater power and wildly non-linear curves but fun to play with.

Thank you!

Yeah, sketchy would have been a much better description of the setup anyways. I'll keep that in mind for the future!

Fortunately, vacuum tubes are super resilient and can readily withstand short pulses on the order of kilo-volts without any problems!

@@UsagiElectric I think he's pulling your leg. At least, I hope so😬

@@UsagiElectric Sorry, I thought I sufficiently flagged the humorous nature of my post. Then again, they not so resilient if you drop them!

All the tubes!

Thank you! And you know what, you're awesome too!

That's no moon! imgur.com/gallery/bBO1mGu

@@UsagiElectric Reminds me of the scene at the end of 2001: A Space Odyssey with the space child and orchestra playing the intro to Also sprach Zarathustra. 👍

Haha, you're not wrong, I think you have to be a little crazy to try to build something like this!

Thank you! And she's my favorite!

Those are little Russian made IV-15s, they're a copy of the DM-160 that came out of Japan. I think they look fantastic too!

You figure out how many football fields you need to lay it out! ;

Hmm, so it's a little tough finding an exact figure for the number of transistors on an i9, but based on their other processors, I'm guessing it's between 3 billion and 8 billion, so we'll pick 5 billion as a nice easy number to get to. Now, doing things with tubes actually can yield a little better efficiency because we can use diode logic and other neat tricks to bring the tube count down. For example, the original MC14500 used around 500 transistors, but I managed to build the UE14500 which an upgraded ALU with just 180 tubes. If I were using an LU like the Motorola chip, I could get that number down to about 120 tubes. So, that's about 1/4 the number. So, let's pull out all the tricks in the book and get the number from 5 billion down to about 1.2 billion. Now, with some better designing and dual sided PCBs, I can get one tube with supporting passives down to an area of about 30mm x 20mm. That's 600mm2, or 0.0006km2. Multiply that by 1.2 billion and we need about 720,000 km2, which is a number so large I can't really imagine it, haha. That's about 180 million acres, or roughly the same amount of landmass as the state of Texas! Now, how many power plants would we need for the heaters? Each tube draws 300mA at 6.3V, so with 1.2 billion tubes, that's 360 million amps, or about 2.268 Gigawatts, which is roughly the same energy consumption as the entire country of Iceland!

@@UsagiElectric Thanks for the breakdown! lol

I'd hate to pay your electric bill.

Those were actually called Williams Tubes and I absolutely love them! Someday, I do really want to build something with Williams tube memory, but for this project, I'm trying to keep everything at really low voltages (just +24V and -12V), so unfortunately, that rules out using CRTs. Still, excellent suggestion and something I hop to get into in the future!

Thank you! I think it's pretty génial too!

Haha, you know, I never really thought about the fact that I'm interpreting a whole lot from just a few flashing lights. "You get used to it, though. Your brain does the translating. I don't even see the code. All I see is blonde, brunette, redhead. Hey uh, you want a drink?"

@@UsagiElectric I was thinking it would be helpful to post a label by each indicator. On the 20-LED board I would have arranged them in groups of 4 (nybbles) for easier comprehension. Just my preference.

An IBM PC? Like this? How many soccer pitches does he get to cover?

@@hubbsllc As many as he likes!!! Heck, I have 9 acres of land and plenty of buildings he can come over here and work on it! ;-)

@@waynethompson8416 SO I guess the ALU can be in one building, the bus interface can be in another building...

Interestingly, I have two goals with this one. I would like for it to calculate the Fibonacci sequence, and I would like for it to print something to a teletype terminal. That means I'm going to need memory, some program control, and input/output. Now, it won't even be 1/10,000 as powerful as an original IBM PC, but I think that should be pretty good for a first go at building a computer from scratch. Ultimately, I want each of the four main components (processor, memory, program control, input/output) to be contained on a single backboard the same size as the processor is now, so I can mount them like this: i.postimg.cc/QM43nPYb/DualRack.jpg Now, I've been doing a lot of thinking about program control, and depending on what day you ask me, I've got a different idea, haha. I've narrowed it down to either punch tape or audio tape. And, if you were to ask me today, I'm leaning a little more heavily towards audio tape. Traditional computers with punch card systems, read the cards into memory and then execute the program from memory. But since memory is so physically massive, I'm actually going to execute the program directly from the tape, and I think audio tape might lend itself to that a little better, particularly as the speeds pick up. Now, in the future, I really want to build another tube computer that is a lot, lot more powerful. I want to use a lot of the design methods the LGP-30 used, but expand it out to a proper 8-bit style architecture like the SAP-1 or even the 8088. For example, here's a 1-bit full adder I designed up that uses just a single tube: i.postimg.cc/02RLpVRB/1-TUBEHVFULLADDER.jpg Of course, the hardest part of that is going to be memory, so I'm going to need to do a lot more research into core-memory if I want that to even be slightly viable. Still, I think it could be very cool!

It's built from 6AU6 only. He did a video where he explained why: Low Voltage performance and low cost.

Crasbe nailed it, I'm building the entire thing out of 6AU6s because they perform really well at low voltages and they're super cheap! In the future, I would love to take a stab at another tube computer, but instead use dual triodes like the 12AU7 and higher voltages, but I think I'll go with an LGP-30 inspired design and use a lot of Diode Logic!

No worries! I'm actually using the 6AU6 pentode for everything except the two soft start relay driver tubes at the very top, which are 2D21 thyratrons. I chose the 6AU6 because it works exceptionally well at the low voltages I'm using here, and, more importantly, they're dirt cheap!

Thank you! As soon as I get off my lazy butt and order some new PCBs and test them out, I'll get the first few kits sent out to people!

I love working on tube equipment, it always feels so special!

Memory is going to be a serious undertaking! On the Discord we've been talking a lot about how to take care of the memory, and the problem is coming down to physical size. The small, 1-bit data bus means that regardless of what I use for the memory cell, the address decoding and select circuitry for each individual bit is relatively large. Regarding the size of the entire thing, I need to limit the size of the memory board to 1,000mm x 700mm (that's the current size of the processor back board). That means I'm running into all sorts of issues before even getting to the type of memory cell I use. But, we're slowly working out the kinks and I think I'll get at least 64-bits (8 bytes) of memory, maybe more (I would like 128-bits or 16 bytes, but I don't know if that's doable).

It is always a shame when a tube dies and goes in the bin, but I'm trying to save and use as many as I can! Also, the 6AU6 was made in absolutely staggering numbers - it's got to have one of the highest production numbers out of any tube. The 6136 and EF94 are equivalent tubes, and I know they were made in Russia for a very long time. So, fortunately, there's still a ton of these little guys out there.

You're not far off from my ultimate goal! I want this thing to achieve two things. I want it calculate the Fibonacci sequence up to 233 (E9h), and I want it to print "Hello World" on something. The first goal is to print that onto a Teletype using 5-bit baudot code. But, I would also love to build something that outputs it to a CRT!

@@UsagiElectric could you make the crt tube driven too? Its not that hard to build a tube TV?

@@Rainbow__cookie Initially, the plan is to send something to a data terminal, just to see if I can. But, once the processor is finished and I'm successfully sending data to terminals, then I think I might look into building up a tube based data terminal. The CRT driver circuitry isn't that difficult, but there are some serious hurdles that have to be overcome for it to be a terminal. I need some kind of character ROM as well as some RAM to store input data. It'll probably end up being as large as the entire CPU itself, but it sure would be cool!

So, lets ignore everything except the processor, and the minimum processor requirements for Crysis was a Pentium 4 at 2.8 GHz. The P4 was a single core, 32-bit processor, so we can do some comparisons! First, let's consider 15 fps to be the minimum amount of frames per second to be considered "playable". The UE14500 tube computer uses a 1-bit data bus, and the clock speed is actually going to be dependent on the speed of the program counter, which will most likely be painfully slow. We'll take an educated guess at around 10Hz. For the P4, it takes the 2.8GHz 32-bit CPU 0.06s to render one frame. For the UE14500, we need to do 32 times as many instructions, and we're running at around 0.0000035 the clock speed. So, it would take approximately, 537,600s to render one frame of data. That's 8,960 minutes, or 149.3 hours, or 6.2 days!

@@UsagiElectric Thank you. I was not expecting such an acurate answer. Did you do you all the maths using the UE14500 tube computer?