Advent of Computing - Usagi Electric and the LGP-21 Restoration

Episode Date: September 13, 2026

Who can't resist Librascope hardware? Well, me, for one. And David from Usagi Electric is in the same boat. In this episode we sit down to talk about his LGP-21, how he got it, how he got it working, and what the machine can do! Subscribe to Usagi Electric to see the full project: https://www.youtube.com/@UsagiElectric

Welcome back to Advent of Computing. I'm your host, Sean Hass, and as you're listening to this, I'm probably on a very, very long plane ride. I've been out of town for a little bit, but before I left the office, I was very privileged to get a chance to sit down with friend of the show, David, from Usagi Electric. For the last few years, David's been working on the restoration of a beautiful LGP21. computer. Now, if you know me, you know I'm a big fan of any Libroscope hardware, the 21 especially since it's an odd cost-reduced LGP 30. In this episode, we sit down to talk

about the trials and tribulations of old hardware and also all the difficulties in getting an old machine to run. So pull up a chair, I'll dust off the tape of the interview, and we can get started. David, welcome to the show. Thank you for having me. This is really exciting. We should just start out by explaining that we're friends because of the LGP21, oddly enough. Exactly.

Actually, that was how we met. So we met at VCF SoCal year one, right? Was it year one? It was year one back. I think there was a SoCal like a decade ago. and then they went on hiatus, and I think it was the first year or the second year back? I think it was the first year when you came out to pick up the LGP21. Yeah, all right, so that's actually a funny story.

Do you want to just get into it? I want the story. Yeah, it's a great story. So the story actually starts in Dubai, of all places. I didn't know that. I didn't know the Dubai connection. So I was in Dubai for work. at the time, not Utagi Electric work, for my old job.

And I had just landed and I had wicked jet lag, as you do. And my phone starts blowing up at like 2.30 in the morning. And because I can't sleep anyways, I'm like, all right, whatever. I'll just get up and see what my phone's going nuts about. And about 15 people had tagged me or sent me private messages saying, hey, you need to check this out and get in touch with this guy. And there was a post on Mastodon or whatever the Twitter alternative was at the time from Tom Jennings. And he had posted up some pictures of his LGBT21 and he had said he's looking to donate it to a museum or to somebody who can get it going.

And a couple of people had mentioned me in response to that and a lot of people had sent me messages. So at three in the morning in a hotel in Dubai, I make an account so that I can post on his post all of my information, who I am, you know, the channel that I run. Give me the computer. I need it. Yeah. And I tried to be as not forthcoming as possible. I wanted him to make the decision. I didn't want to push him into it. You want to be jet lag diplomatic, if at all possible. Yeah, yeah, as diplomatic as possible. And so I made what I thought was a nice diplomatic post. And then I closed the laptop and tried to get back to sleep, finished out my week of work there.

And I get back and I ultimately get an email from Tom and he says, hey, let's have a phone call to chat about this. And so we get on the phone and we start talking about it. And about 10 minutes into it, I'm still trying to play the diplomat, trying to convince him to give it to me without saying, hey, I want it. and he goes, let me stop you right there. I've already decided that you're the one getting it. And I was like, oh, oh, all right. We're good. It's done.

Yeah, we talked to Tom. Tom Jennings is just one of the most lovely people I've ever had the chance to talk to. And we talked for probably two hours on the phone, just about anything and everything. And that was when we made the decision that I, you know, I'm going to drive out to California to pick it up. And it turns out he lives north of L.A. And so I was looking at the schedule and I was like, well, VcF SoCal is happening in February. I was like, can you sit on it for about four months until VcF SoCal comes up?

And then I can pick it up, put it on display at SoCal, and then haul it all the way back down here to Texas. And he said, sure, no problem. It's amazing that you can't just ship that, right? Yeah, I mean, it would be terrifying to try and ship. I mean, I've seen how FedEx spikes my packages when they deliver them. Well, I mean, it's big, it's fragile. it's maybe one of a kind at this point. Yeah, it's very one of a kind.

I don't know how many were produced in total, but of known surviving ones, currently SystemSource Museum has just the CPU boxed one. And then I think Computer History Museum has a complete one, or at least mostly complete. Looking through their online archives, they have the CRT, the CPU, potentially a flexo writer and I think the reader.

But, you know, if they have that much, they likely have the entire thing. They just haven't updated their website fully. Well, I guess context for anyone listening at home, it doesn't know what we're talking about. The LGP 21 is a cost-reduced version of the LGP 30. So we're already talking about an obscure version of an obscure computer. So it's not exactly like sliced bread. Yeah, very low production numbers. I forget what serial number mine is, but I think it's in the hundreds.

But my guess would be that they probably made less than a thousand of these things. I can totally buy that. And of the known surviving ones, I mean, there's those three CPU units I mentioned. And then there's a licensed version of the LGBT21 that was built by a company in Germany. And there's one of those that still exists. And like, that's it. That's all we know of. I'm sure there's more out there, but that's all that we've uncovered.

Yeah. So you get out to California. Yeah, so I do a... With a FedEx truck, right? No, no, I drove out in an SUV. It turns out the thing is compact enough that you can kind of squeeze it all into an SUV. And so I rodeo ride out to California. We hang out with Tom all day long.

And my favorite story about Tom, I'm a lovely guy. It was so much fun hanging out and chatting with him. But we're sitting there, and after about, you know, four hours of chatting, and he's starting to get comfortable with me. And he goes, oh, oh, you'll like this. And he starts digging through his drawers. And I'm like, oh, what are you looking for? And he's like, I'm looking for my uranium.

And I'm like, wait, what? Yes. And he's like, I've got some uranium here. You'll dig it. And then he's trying to find it. Hey, kid, you want to feel something real warm, real quick. Yeah. And he's like, well, I can't find it.

I lost my uranium. But I was still. Oh, no. Yeah. But that was, that was, it was so much fun hanging out with Tom. and I would happily go spend another full day just hanging out with him. But we loaded it up. Now, quick question, though, why did he have an LGP-21?

Because there's some provenance there, isn't there? There's some interesting history in the machine, if you want to get into it. The machine started life at Los Alamos. That's right. Let's come to that later. Yeah, we'll circle back to that. But Los Alamos sold it as a surplus to a friend of Tom's, who had it for many years, and then Tom got it from him. Gotcha.

And then I got it from Tom. So I'm technically the fourth owner on the machine. And so as the fourth owner... Is that a higher low owner account for a computer of that era? I don't know. It's like, it's 63 years old or something at this point, right? So that's four owners in 63 years is not bad. good. Right.

So we're up to you have physical ownership of the machine. That's right. And I drive from North L.A. down to wherever the, I forget where it was in L.A. where the... It was in Orange. Yeah, there you go, in orange. So we drive down to orange.

And I had let the organizers know ahead of time that I was going to be picking up this LGBT-21 and I wanted it to be on display. And so they set up a spot front and center right where the doors open. So it was the first thing that everybody saw.

And we got it set up there. And I had learned, I had not learned of advent of computing until about a month before VCF SoCal. And so on the entire drive from Texas to SoCal, I was just pure podcast the whole way. That's the part that's just too weird to me is you find something that's baked for me specifically and you're like, oh, I've been enjoying this podcast lately. Anyway, no connection at all. Yeah.

That nothing weird could come of this. And then, of course, you show up and see the LGBT and you're like, oh, my God, an LGBT. Yeah. Does that work? Show me the disc. Instant friends. Instant best friends. Like I had known you my whole.

life. We bonded over the LGP21 on display. And then I think the VCF SoCal was one of my favorite VCs that I've been to. It wasn't as big as Southwest, so it felt a little more intimate. And the atmosphere was just so friendly. Everybody was super nice. And I have really fond of memories of that event. It was great.

And then, of course, I loaded it up and went down and visited. I visited Shelby at Tech Tangents on the way back and visited my buddy in New Mexico who had a teletype that I picked up, which I somehow managed to squeeze into the SUV, and then got it all the way back here, and we started the restoration process in earnest. The restoration process is something I really want to talk about because, and this is something we've discussed before off mic, but as a software guy, hardware restoration to me is really interesting because there's so many choices and challenges you have to face. Right? Like, how do you preserve what's essentially an artifact in a way where it's useful?

Because computers, like a lot of other artifacts, or unlike a lot of other artifacts, are totally experiential. Right? If you can't use it, you miss like almost 50% maybe more of the point of the artifact. I have, I don't want to throw shade here, but I have a really hard time with a computer that is just on display and never turned on. Yeah. Hey, I'm with you. I'm completely with throwing shade. Like there are logistical issues, right? Of course. You're not going to be able to get any act to work ever again. in, but with a production computer, that's a lot different. And once you get up into the 1950s and 1960s, I think it is actually attainable to get almost anything going again.

It's just a question of time and effort, really. As you know, and as some of your listeners may know, this is not my first 60-plus-year-old computer to try and restore. Not your first rodeo as it were? As it were, yeah. I kind of cut my teeth on very old machines with a Bindex G15 from 1956. Now that's a full vacuum tube computer, but that's a little misleading to say it's a full vacuum tube computer because a lot of the logic is implemented with germanium diodes.

So you can make and gates and you can make ore gates out of germanium diodes. and you can do a lot of really complex things with just and and or, and then you slip a tube in whatever you need an inversion. And so by doing that, in the G15, they have around 3,400 germanium diodes and about 460 dual triode vacuum tubes. And they're able to do all sorts of really amazing complex stuff because they've got a big rotating drum memory in there that does a lot of the heavy lifting.

And while we were restoring that, I came across five bad vacuum tubes, which is like an incredibly good rate. But we came across... Out of how many? Out of 460. So five, two... Oh, wow. Okay, now that's great. That's fantastic.

Incredible. And it was pretty easy to track them down. And of those five, three or four of them had bad filaments. That was all. And so you could visually tell that tube was no good because only half. of it was illuminating. The fifth one just had bad emission on a pre-amp. So we weren't getting, we weren't able to read the data coming off of one of the lines. And the diagnostic software told us which line it was. So that was a really pretty easy one to track down too.

That's super cool. The bigger issue was that we had 60 bad diodes. Now... That would be harder to track down. Yeah. As to reiterate, there's 3,400. germanium diodes in the thing. So 60 out of 3,400 is an incredible visually inspect those. No, you can't visually inspect them. But thankfully, Bendix makes a diode tester where you plug the individual diode packs into it. You can test each individual diode. So using this pack tester, I tested all 3,400 of the diodes. It took two days, which is still really quick for 3,400. And I didn't replace them

germaniums, I used a modern shot key that I slipped in there. But it had very similar characteristics and it didn't affect the operation in any way. So why would you use a modern diode over a more traditional germanium? So this is kind of... Just availability? No, it's more spiteful than that. Yeah. Yeah, so within the Bindex G15 community, because there is a Bindex G15 community.

Of course. A lot of people have simulated the circuits in the G15 in Spice. And the general consensus was that it had to be a germanium diode, otherwise it would not work. Because germanium diodes have a certain amount of reverse leakage. And the flip-flops had kind of taken a little bit of that into account. And so, but I'm looking at it going. digital logic is digital logic. It shouldn't matter whether it's germanyum or shakier.

A diode is a diode. And multiple people were like, no, it's not going to work. You can't run a shot key in the G15. It just flat won't work. And I was like, well, has anybody actually tried? And they're like, well, of course not. There's no running G15s out there. Well, and also there was no G15

to test it on. And so I was like, well, all right, why don't we give it a shot? And so, Yeah, we can just do it. Yeah, the worst it could happen is you have to replace another diode. Yeah, and I just go back in and remove the shotkey. But I put 60 shot keys in there because if anything was going to make it not fail, 60 of them would definitely bring up all sorts of problems. And it worked flawlessly.

That's funny. Yeah. But the benefit of the G15 was that all of the diode cards were easily removable and had a dedicated diode pack tester. So going into the LGP21, I was a little nervous because the boards on the LGP21 are much larger. And it's all germanium diodes and germanium transistors. Those are a lot harder to test. And I was looking at a 2% failure rate on the G15.

So I was terrified that if I had a 2% transistor failure rate or 2% germanium diode failure rate on the LGP21, it would be... Yeah, that's going to be a lot of work. an immense amount of work. We would have to write, but we would have to get it operational enough to load test programs in and then run test programs to exercise every facet of the machine and trace it out with, I mean, it was going to be massive. And the truth of the matter is,

is that we plugged it and turned it on and the whole thing flat worked. I think that's like the best testament to 1960s engineering, that it's just been stable for so long. Engineering and manufacturing. Okay, fair. Very fair. The quality of the germanium diodes between, just in 10 years, between 1955 and 1965.

And actually, it's a little less than that. We're talking 56 to 63. But just in that 10, less than 10 years, the quality of the diodes went from a 2% failure rate to a 0% failure rate, right? That's wild. Right. It's crazy.

And, well, that's the kind of stuff that you don't read about in trade journals from the day unless you get like really, really deep. Oh, yeah. Yeah. And it's one of those things that you would only discover by restoring multiple machines from the 50s and 60s. Everyone has some time to do that, right?

You can just do that on the weekend. Easy. Now, that's not to say that the LGBT, I say it flat worked. The electronics flat worked. There were other things that gave us. headaches on the LGBTQ21. Like what? What could possibly go wrong with a computer of this vintage?

Well, the single component that tripped me up for at least a month is one of your favorite items. Are we talking the flexo? We are talking the flexo. We're talking the Freed and Flexo writer. The Freed and Flexo writer is just a fascinating, like insane. device to me. I should do another episode on it.

There's just so much. It's just weird. There's some interesting things about the freedom flexo writer. Correct. Truer words have never been spoken. Hang on, let me rephrase that. There are some interesting things about the LGP21 specific freedom flexo writer. Because it is not the same as any other words.

Any other flexo writer out there. Well, yeah, they were all custom. That's like the wild thing. It's like, oh, you want a terminal for your computer? Call the manufacturer. We'll work out something special for you. Yes. That's horrific.

I hate that. So the flexo writer, with a little bit of effort, got to where it was typing. And I could use it as a typewriter. That part was pretty solid. I had to clean a lot and free up. up a lot of gummy stuff. But that part was not the hard part. The hard part was LGP and their infinite wisdom decided to offload... In their infinite precision, please. Yes, there we go. They decided to offload some of the I.O processing oomph to the flexo. Oh yeah, that's right.

are you talking about con stop specifically? Maybe. I'm not sure what con stop means. So con stop is the character that's rendered as a single quote that you have to do at the end of buffer input that like signals a trap on the CPU side that you can go back to processing.

Okay, that is also a giant pain in the butt and we butt our, we hit our heads against the wall with that too. but that's not the specific one that I'm referring to. All right. What's the issue? So we first got it up and going. We had the FlexoWriter plugged in. The CPU looks like it's doing the things that it's supposed to do.

I can type on the FlexoWriter and see the data being shifted into the input register. We can push a button on the LGP21 to copy that input register over into the instruction register, and then we can push a different button on the LGB21 to execute that instruction. So I can now use the FlexoWriter to input an instruction, move it to the instruction register, and execute it. And so we input an instruction to tell the CPU to print the letter D on the Flexo rider just once. Hard coding. Yes. Of course.

So we type in the instruction, we move it into the instruction register. I hit execute and the FlexoWriter types D a million times. Well, yeah, you told it to print. Why would it ever stop printing? No, we only told it to print once. That was the instruction. Was it print just one time? It should have printed the D one time, but instead it printed it on repeat.

And so we started digging, we thought, oh my gosh, do we have a logic fault? Do we have a bad transistor? Do we have something? So we dig into the schematics and we start trying to wrap our head around how I.O. is working on this machine. And it turns out that all of the I.O. timing is done on the flexo rider itself. The CPU... Wait, so the timing specifically. Yes.

So the CPU... That's nonsense. The CPU throws a value onto this parallel input that's going into the flexo rider. And the CPU just keeps puking that value onto that parallel input until the flexor rider sends back a signal that output has been finished. At which point the CPU goes back to what it's doing. So I have a question. Yes. So timing on the LGP30, I know, works off a track on the drum.

Does timing on the 21 work off a track on the disc? Yes. So we're talking about timing in two different aspects here. So you're thinking of timing for the CPU. Yeah, well, that's what I'm getting to. Does the flexo writer have like a crystal clock inside it that does its own timing? It's so much worse than that. Yes.

Let's go. It has a camshaft. Is it another spinning disc? Yes! Oh no! It has a four-lobe camshaft that makes and breaks contacts at specific rotational points that- Oh, my God. Defines the I.O. timing for the entire LGP21.

So while the CPU is timed off of the disc, when it's doing I.O., it's waiting for specific I.O. timing signals from the LGP. And of course, that camshaft is buried very deep. So I had to take most of the LGF, I had to take most of the Flexo writer apart to get to the camshaft, and it was all gummed up

and gross and sticky, and I had to break a bunch of stuff loose and get oil into all the right places. And then, after that, it worked just fine. I mean, I can see how I can see how Frieden doing their own thing and making a very, like that's the thing with the flexor writer, right?

Is it's not a teletype. It's not a terminal in any way that a human would recognize a terminal. It's a mechanical, a completely electromechanical typewriter that you can feed data into kind of. Yes. Yeah. So I can see why it would be its own mechanical thing because I was about to say, oh, it could just take a clock signal from the computer, but that, no, that would not work. Nah, it's just a CEMSAT.

And actually, in the manual, it's hilarious. When you go through the manual, in the manual, they have some diagrams that talk about timing on the flexo rider. And these diagrams have little drawings that show at what rotational degree the contact will be made or broken. So it's like, I hate that so much. It's the most cursed thing ever. So that means that, well, this is ringing a lot of bells because when I made my LGP30 emulator, I had so many issues figuring out how to have, like, do FlexoWriter emulation because I wanted to have it as a separate module. And I ended up giving up on that idea because that's garbage, during the timing synchronization, like messed up so much stuff. So I'm glad to know that in reality, the timing synchronization was garbage.

Oh, yeah. It's cursed. That's great. It's incredibly cursed. Just wrong. Oh, my goodness. Yeah, so the flexo writer tripped us up for at least a month. But once we got the flexo writer going, and it was working pretty reliably.

It turns out that the rotating disc memory on this is non-volatile. just like the LGBT 30. As with all, all disc memory is non-volatile. Although, except on... There's a caveat. If it's magnetic.

Okay, what's the caveat? The caveat is that the Bendix G-15 is volatile. How in the world does that... No, I don't believe you. That's a lie. How is it volatile? It's magnetic, right? It's a magnetic drum.

Yes, it's a magnetic drum. They put a permanent erase magnet. Oh, that's... all of the tracks. So every track is erased, every rotation. They're using it as if though it is a delay line memory. That, you know, I wish you hadn't said that. That's awful.

Yeah. But. But the LGB21 rotating disk memory is non-volatile. It's normal. It's correct. Yes. So we poked around a little bit to pull some values out of memory. And sure enough, there was stuff in memory.

So now we're thinking I remember when you When I heard you say that messaging you and being like You have to dump the disc the disc must be dumped Yes But how do you dump the disc in a way that makes sense There's 4,000 words of memory on this thing by the way You just take

You don't even use an oscilloscope You just get one of those little blinker flash bulb things And you spin the disc and you have it flash on the readhead and you just read it really quick. No, that's, that's, boy, that's a lot of work. Remember, there's 4,000 words of memory on this thing. Each word is 31 bits long, right? Yeah, it's a lot.

So we were thinking, and we, at this point, we have a functioning CPU, so let's utilize the computer itself to dump the memory. And so, you know, the punch, we haven't gotten the punches to work yet, but even if we did, and we punched out the memory on paper two, tape, it would be like 300 yards of paper tape. It's a lot. And so what we did instead was we just did a hex dump on paper using

the flexo writer. So every single word was punched out in hexadecimal on paper, and it just did that for every single track. Well, you had like a whole track that was

empty to work with, right? If I remember correctly. Yes and no. We thought that we had track zero and track one empty so that we could load our dump routine into that, into those tracks and dump the rest of the But it wasn't. But they weren't empty. They were

broken. Oh. We couldn't We couldn't read or write anything to those two tracks. So, okay, we've got bad heads or something. But instead of trying to troubleshoot that, let's

dump the rest of the disc while we can. Because we're in a position where we can get 96% of all data off of this thing. So let's do that. So we went to the last track. We had dumped the last track before because that was where we were loading stuff into. Oh, no, no, no. My mistake.

We went to track two. So zero and one were bad. Track two was where we had pulled data off of track two because that's where we were kind of experimenting and playing the stuff. So we loaded our dump program into track two and off to the races. we dumped all of the data. Turns out every other track worked perfectly fine. Zero problems.

Just those first two. It was just those first two. And we'll come back to those in a minute because we didn't give up on them. Well, I mean, it goes around. We'll always come back to track. Yeah, there you go. Oh, man, that's a bad one. That's really bad.

That stinks. So we get the rest, but now we've got it out, printed out on paper from a flexo writer with a failing ribbon. But I scan it in and over the next week we do OCR with manual confirmation

of OCR. Oh, gross. Yeah. It took a week but we got it done. And so we have a full digitized text file that's shared in multiple places of the disc. So fully safely back done. And then we go back and we start trying to figure out what happened with track zero and track one.

And the heads are quite interesting. They have, it's just a coil of wire with three taps. So you have a tap on either end and a center tap. And all three of those taps come through the casing to lugs on the top that you then solder everything too. So that way you can have your read-right head accessible that way. So one half is read, one-half is right, or however they've got it set up. and if I measure from the end of the head, using the lugs on the top side of the case, on the outside of it,

the head measures fine. But if I measure from the center tap to either side, the head measures bad. So what's happened on track zero and track one is that center tap has broken on the inside. Huh. But magnetics are fun because if you're spinning the disc

it's going to be inducing a current into the head regardless, right? So we spun the disc up. I put the ground probe of an oscilloscope on one end of the head, the oscilloscope probe

on the other end of the head, and we can now see the flux transitions on my fancy oscilloscope of the disc. So you just rewire it an extra leg and you're done? No, no, no, no. Because at this point, we can only read, right? We can't

write anything down. Oh, gotcha. We're making that center tap. So this is how we extract the data off of it. I just use my oscilloscope, take a scope shot, and then we go through and pulse view and we extracted the individual bits and figured out those tracks. So we were able to rescue the entire disc this way. Now,

LGP knew that you would have a head fail at some point. So they gave you four spare tracks and four spare heads. That's just value. You don't even get a spare bit these days. So all I did was desodder the wire to track from, you know, the track zero head and

soldered it onto a spare. Same for the track, one head. And we were off to the... I wonder if there was a persistent manufacturing issue or something that made them do that or if that was just like standard procedure. Standard procedure. Bendix gives you a full spare track plus a spare clock track. Gotcha.

That was just kind of, that was what they expected you to do. It was kind of the wild weather. That is, that is neat. Stuff is going to break, yo. Might as well give you spares. Yeah. So, but here's where it starts getting exciting because we've got the data from this disc. We know that the machine came, spent the first years of its life at Los Alamos because we still have.

Doing who knows. Yeah, who knows, right? Because all of the, uh, Individual components have the Los Alamos scientific laboratories asset tags on them still. So we know that the provenance is real, right? It did actually come from Los Alamos. And so we start digging through the data. And there's, you know, there's instructions.

And you can always tell an instruction versus data versus text based on the first three digits, three hexadecimal digits of the word. So immediately looking at it, we can go, okay, this. is this is an instruction. Oh, this looks like data. Oh, this looks like maybe Asky text or something like that. And so we kind of do... Well, Asky.

Flexo E text. Yeah, yeah. It's Flexo Asky, which is whatever... Flexi. It's some weird six-bit monstrosity. And because it's... I think it's actually based off telegraph encoding. If I remember...

Yeah, I think so. It might be. I don't know. But because it's... It doesn't matter. And each word is 31 bits. They stuff multiple letters in a single word. Oh, yeah, that's right.

So you have the, I remember that now. The print packing for like printing strings is horrific. Oh, yeah. No, it's, it's a- It's a nightmare. It's a nightmare. It's a nightmare. It is a,

and like an hour-long conversation with this guy met also at VCS SoCal a while back that made an implementation of fourth for the LGP 30. Okay. And I've been wanting to do an implementation of Colossal Cave Adventure because I think that would be funny. You could do like an infinite loop tape to load the text. It would be an art piece, right? And so I've always been trying to think about like string print, string compare. And so I was asking him, well, how do you do print and compare for your fourth?

And he just like got really quiet. He's like, well, it took some time. Yeah, something that we take for granted today, which is just like printing text to the user, was... Non-trivial. Nightmarries being difficult with a flexo writer. But we started... So you find packed text. Yeah.

We started pulling strings out of this thing. You know, we wrote some utilities to go through and just grab anything that looked like actual letters. And there wasn't much. There wasn't much text data on the disc. It was mostly like actual data. But the text data that we did see came up with the first thing that we that we decoded started with M-O-L-T-E-N. I was like, oh, molten. Okay, that's pretty interesting.

Yeah. And then that was the next word was molten sodium. And we were like, okay, this is getting excited. The next word was P-U. This is getting I-TAR-E. Yeah. Oh. P-U for plutonium.

And then what capped that one off was experiment. So it was the molten sodium plutonium loop experiment. That's cool. That's a really cool thing to find on an ancient magnetic disk. Yeah. And then we dig a little deeper and we find references to U-H-T-R-E-X, which is the ultra-high-temperature reactor experiment. This is a reactor that uses enriched uranium that is unclad. So everything gets contaminated, but it allows the reactor to be much more compact.

And then you have a heat exchanger with a contaminated cooling loop and a non-contaminated cooling loop. And so we found a bunch of information about UHT-R-E-X. We found information about T-N-T debris, all sorts of interesting stuff. And I was a little nervous. So I sent this to my buddy who actually currently works at Los Alamos. And he made it. You're just like, hey, pal, there's a leak. There's a leak, buddy.

It's me. He had it make the rounds of all of his buddies. And they all said, yeah, this is, all of this stuff has been declassified and is no longer. Well, that's good. That's a plus. So the men in black are not going to come knocking my door down anytime. soon. They're just going to carefully remove the magnetic plate and the text files.

So we did actually write a nuke program that would nuke the disc, wipe it clean in case somebody told us to. But so far, nobody's actually told us to do that. So I think we're in the clear. But yeah, so we found some really exciting stuff on the disc. But once we had it all backed up, it meant that we could. lay down new things. And so we wrote a couple of new programs.

Notably, the big one that I really wanted to run was a version of Conway's game of life that is one-dimensional. And this is actually John Millen's interpretation of one-dimensional life, not Stephen Wolframs. Everybody thinks it's Wolframt's

20 life, but it's John. Now, okay, maybe I'm outing myself as a loser. I was going to ask if it's related to Nils A. Baricelli's one-dimensional finite automata, but I guess that's too much of a deep castle. It's deeper than I got. You got me lost on that one. I'll send you the paper. It's fascinating. Okay. I mean, I find Sether automata absolutely fascinating in general, and I think Conway's Game of Life should be run on every

single computer ever. But if you don't have a CRT, how do you do two-dimensional game of life? Not easily. Or very slowly. Yeah, it turns out that one-dimensional was the answer here. And so you do dimension,

so one line is one generation. So you randomly fill that line and then based upon your rules. And then you do the next and next. You do the next generation, the next generation, the next generation, next generation. And since you print it out, you get this really cool printout of the evolution of that life.

with its history embedded on the paper. And so we wrote that. I say we. I didn't write it. Reese Weatherly. All credit goes to Reese Weatherly. He did some incredible work on that. And Bill Cookear as well.

So he did some incredible programming work on the LGBTQ21. And we wrote a lot of really cool programs and got it to VCF Southwest just this year, where it was on display, and it ran for three days without a single failure. That's fantastic. Yeah. The only thing it did was it chewed through a typewriter ribbon in the process.

That was it. You know, I'm amazed how much this isn't a restoration project besides for the flex. So it just kind of works. We got incredibly lucky. But also, I think it's the kiss principle showing why it's such a good principle to follow. You know, the keep it simple, stupid principle, right? It is so simple and has so few transistors in it that there's just not that many things that can go wrong. And so the few things that did go wrong, we were able to track down and fix relatively quickly.

Now, as a result of the Kiss principle, it is very slow. this is true I would say it's actually probably half the speed of the LGP 30 now I had read that the 30 operated at like 100 hertz and the 21 operates at 80

so it which may or may not be true it's hard it's hard to to classify that way so I've you know in this getting out the manual yeah in this

what does the spec sheet say so in the spec sheet uh the yeah in the spec sheet the clock frequency is 80 kilocycles um but that's not what makes it suck uh disc access is what makes it suck and ah yeah the the problem is is that they have interleaved two tracks per head oh shoot okay i didn't realize hang on a second Does that mean it needs a double rotation? Yeah, because I think you've got 32 heads and 64 tracks.

Oh, geez. Okay. I remember reading that it was interleaved. I didn't put together that that meant you need twice the rotation for every read. So, yeah. So the way they've got it set up is track zero, word zero, is your first sector that you come across. And then the next sector is track one. word zero zero and then the next one

here's where you're going to get real messed up is track zero oh I might have this wrong word 47 okay wait please in hex incorrect hexadecimal

that's like CW or something no that's the hex uh hex 47 oh okay all right let me see if I've got this right I might I might have had that wrong.

I think it's hex. Any rate. Either way, it's stacked. It's not sequential. It's not sequential. And it's because it has to do with them trying to line the words up in such a way that track zero word one swings across the head right as the previous instruction has finished. So that way, as a programmer. you can write your program sequentially.

And so anything like a fetch instruction or anything, you know, simple instructions, they take a certain amount of time to execute. And during that time, the disk is spinning. So the disc will have rotated a certain amount while it's executing that instruction. And so they've set up the interleaved sectors in such a way that you can write your program sequentially, but the next sector should be swinging right by the head just as your previous instruction finished, which is all well and good until you do a division or a multiply, which takes like three-quarter of a revolution, or I.O, which puts the whole thing in pause and waits

for the flexo-rider to do whatever the flexo writer is doing. So I'm wondering now, the LGP 30, I don't think, has that kind of interleave at all. No, it doesn't. So that means LGBT 30 software, yeah, so software that's been optimized for the 30s drum would probably be horrifically slow. Yes. On the 21. Yes. Huh.

That's so funny because it was sold as compatible. It's mostly compatible. There's a few things that get a little wonky when trying to port between 30 and 21. Like what? I think some of the I-O is different. I would have to pull up the 30 instruction set. The 30 instruction set is slightly different. I can buy the I.O.

With the I.O. Because things were slightly different on that one. But the total amount of word storage and the core instruction set is the same. So you should be able to port with minimal effort, anything that was written for the 30 to the 21. But the 30, I'm not as familiar with it as probably, probably you are, but I don't know if they stacked the tracks, the words in such a way that the timing of the disc would, but I don't, or the drum would roll around.

I don't know. No, the timing, if I remember correct, is all based off sequential access. So you have to do, you have to stagger and optimize very specifically. Okay. Yeah, it doesn't. there's nothing it's a oh what's the word unadulterated raw computing you don't have any niceties yeah unless you write it by hand that's very similar to the g15 so the g15 you had to you had to figure you had to know how much of a rotation your instruction would take and then you would

place your data at the correct location to grab it or your next instruction at the correct And you end up with like... Classic. Classico. The instruction one is at track zero, sector zero, but instruction two is at track zero sector 50, right? Because that's just how long it took. And then you end up, it becomes a nightmare.

Just trying to think about drum optimization with the lobe thing on the flexo you described now is just scaring me in a way I don't want to consider. Yeah. As soon as you throw I.O. into the mix, it all, everything goes sideways. Just falls to bits. But even excluding I.O. Disc access on the LGBTQ21 is so slow.

And it is such a minimal instruction set. There's like 21 instructions in total. But I'm thumb. Yeah. Is it 21? That would, that would make me look through the, that would make me look through this. It is 23.

Sorry, 23 in total. Okay. Should I call it the LGP 23. Yeah, they should call the LGP 23. Missed opportunity. It's so slow that when we were doing the 1D life, so you know, you would have one generation that's maybe, I think we were doing like 60 characters across, right? So you have to look at a cell and then look at the two neighbors to the left and the two neighbors to the right of that cell to determine the things. to determine the fate of that cell,

store that in a new array, and then do that full calculation, then print that new array out. Oh, and with no index register at all, you have to self-modify. Yeah, yeah, yeah, everything, of course, because why would things be simple? You have literally three registers in the entire machine.

You have the accumulator, the instruction register, and the program counter, and that's it. That's all you get. And you can only touch one. And you can only touch, yeah, and you can only touch the accumulator. So, yeah, you have to work your butt off for it. And so to calculate a single generation of life takes like two minutes.

It's slow. That's great. Yeah, it's very slow. That's computing. Yeah. We did get it to do the Mandelbrot set. And so a quite small Mandelbrot, maybe three by five inches when it prints out, how many characters that comes to, takes

about 50 minutes to do. Oh my God. Yeah. It's slow. Yeah. Well, that's wild to me because the stuff that people were using the 34, most famously, it was used

to write this language called dope, which was a precursor to basics. So it was used in classrooms to teach with instant feedback, and it was used for simulations of chaos theory. Okay.

So like fractals, people would have been actually doing similar stuff where they're generating fractals and doing like generational kind of work. So the 30 was notably faster because... But still, that's like the fractal you just talked about would have taken like twice the speed 20 minutes instead of 50. Yeah. Well, and so interestingly, that's an equivalent, an equivalently something. sized mandelbrot on the Bendix G15 is between 15 and 20 minutes.

Interesting. So the G15 is about a little more than twice as fast as the LGBT21. And I think, you know, I've never been hands on with a 30. I would love to get an LGBT 30 in here and drag race it against a G15. If you can find an LGBT 30, I will, I think, have an aneurysm. That would be crazy. That's one of my bucket list machines. I will fly to Texas.

I will show up at your house. There was one at the Living Computer Museum that went up for sale on Christie's auction after Paul Allen's passing. And I threw my hat in the ring and I tapped out at $10,000 because that was getting beyond what I could afford. At that point in time,

Bob Roswell at SystemSource threw his hat in the ring and he tapped out at $20,000 and ultimately it sold for $32. I mean, it is literally a museum piece. I have only seen those behind glass. Yeah. So the 30 is, that's a bucketless machine I would love to get my hands on someday.

And it would be really fun to drag race it against the G15 because they had very similar target audiences and they had very similar performance. but achieved in very dramatically different ways. And so it's actually interesting to me that they were using the LGBT-21 at Los Alamos because it would have been kind of terrible because Los Alamos had plenty of LGBT 30s already. That was a machine that they used a lot at Los Alamos. Well, and plenty of money. And yeah, yeah, they had the game genie with the infinite money glitch. Right.

So it seems crazy to me that they would have, Los Alamos would have bought an LGBT21. All I can think of is that maybe, you know, somebody that worked there that was kind of their personal machine. They somehow convinced the beam counters to allow him to buy a machine for him alone or something or her alone. That's still wild, though, because it is such a question that I don't think I've ever been able to answer to my satisfaction is what people actually thought about performance of computers in the period. Yeah, I mean, it would have been. And it's so slow, even compared to the field. Like, what's the application for a computer that's that slow? Because you're not doing, like, commodity computing.

You're not doing computing as a service yet. Yeah. I mean, I think if it's less about like, again, I was born in 1984, so this is total wild speculation. But my theory is that it was more like just having an unpaid intern. Because I think you would just. Yeah, I could see that. You would just load up something that would take two hours to calculate and you'd hit go. And then you'd go get your cup of coffee and sit down at your desk and do something else.

And then when it was done, it would punch out the result. And the result was perfect every single time because it's a computer. And it was just something that could sit in the background and crunch numbers. Even if it was slow, you know, if it took two hours versus one hour, it doesn't matter. I mean, you're in this for the long haul. But again, this is total wild speculation. That's about the only thing that makes sense to me, really. Yeah, yeah.

I'll keep my eyes out when I'm going through accounts. Yeah, yeah. Because that is an interesting take. It would be fascinating to talk to somebody at Los Alamos in the 60s that was there in the 60s and 70s. I mean, they would be up there in years, but they would have some interesting insights into how these computers were used. Because when we think of Los Alamos, we think of all the really big, exciting stuff they were doing with nuclear material. and reactors and research that way. But we never, you know, nobody ever really talks about the processing power that was,

you know, the horsepower that was feeding that, that was powering all of that. It was pushing it along. And, I mean, even today, like, I have no idea what kind of computers Los Alamos are using or what they were using in the 90s or in the 80s or in the 70s. I'm sure they had access to some ridiculously powerful machines for each decade that they were in operation. But, I mean, it'd be really fast. fascinating to figure out the full computing history of Los Alamos.

I wonder if there's a document about that, because the NSA has multiple reports over time. They're just titled History of Computing at the NSA that go like year by year over what hardware they had. Oh. So there might be something internal or like FOIA responsive for Los Alamos. Yeah, that would be a fascinating thing to figure out because if you're not to, if you're on like the very bleeding edge, which Los Alamos was and is,

you need bleeding edge computers to help you do that. And so, which is why it's so weird that they had an LGBT 21. Because even in 1963, it was slow.

Well, I mean, 1963, people are doing time sharing by 63. Yeah, like 63, 64. Yeah, that's just odd. It's wild to consider. Yeah. So it's a really fascinating machine. We need to get back into it from the restoration perspective. The punch doesn't work yet. I haven't gotten to work on the punch.

We've gotten the tally reader working just fine. So that's how we've been loading programs in. And of course, the Flexo writer is still fighting us. We need to... It always will. It always will. It always will. Curse technology. It never will not fight us.

We need to pull the friction roller out or the power roller out as well as the platinum. And I need to send that off to a guy up north that resurfaces them because they're starting to come apart. And then the reader on it is missing a piece. It had fallen out over the years and gotten lost. So I need to build a new piece to make that work. and then the punch is fully seized on the flexo rider. So I got to figure out how to break that thing loose and get it spinning again.

But the core part of the flexo rider is mostly working. Sweet. I am, I will say, one of the things that I'm most excited about with the LGP21 work you've been doing is, I think at this point, the graph of people that can program LGP hardware, there's a big bump in the 60s, 50s and 60s, right? I think it slid down to probably nothing for decades. It's starting to go up again. Oh, yeah, yeah.

There's an odd number of people that know LGP assembly language now. Yeah, it's crazy. One of the first things we ran on it was it printing out the Fivinacci sequence in hexadecimal. That was kind of our live or dead test, which means that somebody set down and wrote a program to calculate the Fibonacci sequence. in LGP21 assembly, right? Now, I do want to just ask one question. Your buddies that are working in LGP assembly, do they have an assembler yet?

Or are they doing it raw? I think they've written their own tools to do it. Okay, good. I will say that I am a terrible programmer. and what they do is so far beyond my capability, even to understand, that I'm going to have to defer completely to them and their toolset, notably recent Bill. But there's a handful of people, I'd probably say five or maybe even 10 people

that have over on the Discord that have been experimenting with LGBT21 Assembly. and actually like getting things done and having them run in the in Paul's amazing emulator, which is like totally mind-blowing to me. But to be fair, LGP assembly is far more approachable than Bendix G15 assembly. Vindex G15 assembly is like the most cursed programming I've ever seen in my life. LGBT21 is almost feels modern. That's horrific. Because I will say, knowing LGP30 machine code,

it does not feel modern at all. So it feels ancient. Well, that's just because you're saying the bindings is worse. You've got to compare it against the bindix, man. All right. I'll go, I'll bone up on the bindix. The bindix is very, very rough. It's very, very low level.

It's so low level it makes the LGB21 feel high level. Oh, is it more like microcode almost? Kind of, it's more like you're interacting with the drum directly is what it feels like. Ah, okay. That does sound bad. It's rough. It's, uh, I've, I've spent a couple hours with Bill trying to figure it out and having him walk me through what he's learned about it. And I've talked with, uh, Lloyd who's done, who's done quite a lot of amazing BinxG-15 programming.

And I just can't keep up. I just can't wrap my head around it. Uh, but at least with LGBT1 programming, I can go, okay, that's obtuse, but I can see how it works. You can appreciate that it is software. Yes. Yeah. That's great. Well, I think that that's as good a place to call it as any. Otherwise, I think we'd both keep talking about old machines.

Oh, yeah. I mean, this podcast, this could have easily been five hours long. Ha. Well, I'm excited to see what comes next for the 21. I know you say you're taking a break from it, but very, very amped to see. more software running on the thing. Yeah, so we're only taking a break for as long as it takes me to get the new room that it's in sorted.

So we're not taking a break because I want to step away from it and do something else. We're taking a break because real life has gotten in the way. Yeah. As it tends to do. But that room is starting to take shape, and we've got the LGBTQ21 set up next to a lot of other really neat things. and we've got power run to it, so if I could, you know, move enough junk out of the way to sit down at it, it should theoretically be able to turn on and run even right now.

So we're definitely going to get back into it. I'm going to have new videos coming out on it, hopefully by the end of the year, but if not, certainly in 2027, we're going to run a whole lot more software on it and do a lot more work on it. Sweet. And where can people watch those videos if they don't already know? That would be the Usagi Electric YouTube channel. It's spelled USAGI

like United States of America government issue. Although Usagi really... I don't think that's what it means. No, that's not what it means. Usagi is the Japanese word for rabbit. But yeah, Usagi Electric

on YouTube. And that's where all of the entire G-15 restoration is. That's where the LGB-21 up to this point and where the coming restoration on the next G15 will be as well. Sweet. Cool. Well, thanks for coming on. It's been

wonderful talking with you. We need to talk more. I always have a kick whenever we connect. Well, thank you so much for having me. And, well, now that you said we need to talk more, I'm going to call you every week. Do it. See you around. Thank you so much, man.