ACM ByteCast - Robert Metcalfe - Episode 40

Episode Date: July 6, 2023

In this episode of ACM ByteCast, our special guest host Scott Hanselman (of The Hanselminutes Podcast) welcomes 2022 ACM A.M. Turing Award Laureate Robert Metcalfe, Emeritus Professor of Electrical and Computer Engineering at The University of Texas at Austin and Research Affiliate in Computational Engineering at the Massachusetts Institute of Technology (MIT) Computer Science & Artificial Intelligence Laboratory (CSAIL). Metcalfe received his Turing Award for the invention, standardization, and commercialization of Ethernet, the foundational technology of the Internet, which supports more than 5 billion users and enables much of modern life. His other honors include the National Medal of Technology, IEEE Medal of Honor, Marconi Prize, Japan Computer & Communications Prize, ACM Grace Murray Hopper Award, and IEEE Alexander Graham Bell Medal. He is a Fellow of the US National Academy of Engineering, the American Academy of Arts and Sciences, and the National Inventors, Consumer Electronics, and Internet Halls of Fame. In a wide-ranging interview, Bob reflects on his “Ethernet paper” with David Boggs from 1976, and how the interoperability and backward compatibility baked into the Ethernet allows the technology to hold up today, in the age of Netflix and Zoom. Bob also describes his most recent project, modeling geothermal wells as a computational engineer at MIT, with the aim of harnessing geothermal energy as an alternative to fossil fuels. Along the way, they touch on “stretch goals,” GPUs, and how far down “the stack” one needs to go to fully appreciate and understand a piece of technology.  Link: "Ethernet: Distributed Packet Switching for Local Computer Networks" (Metcalfe and Boggs' classic 1976 article in Communications of the ACM)

This is ACM ByteCast, a podcast series from the Association for Computing Machinery, the world's largest education and scientific computing society. We talk to researchers, practitioners, and innovators who are at the intersection of computing research and practice. They share their experiences, the lessons they've learned, and their own visions for the future of computing. I'm your host today, Scott Hanselman. Hi, I'm Scott Hanselman. This is another episode of Hansel Minutes on. Also, we're combining this

with the ACM ByteCast. So you may be hearing this as a part of the ACM ByteCast podcast series, or as a part of Hansel Minutes. Depending on where you're listening to it, I encourage you to listen to both podcasts because they're wonderful. We are talking today with Dr. Bob Metcalf, graduated from MIT in 1969 and is the co-inventor of Ethernet and founded 3Com. How are you, sir? Great. How are you? I am very thrilled to be chatting with you because we wouldn't be chatting with you had

you not done the things that you've done. I don't know how we would be talking over the internet. Maybe we'd be using Token Ring at this point. No, I don't. But you're right. Were it not for Ethernet, I wouldn't be here today. So I'm curious. We could do what people typically do at these points in the podcast, and we could start at the beginning. But I kind of like to work backwards. What are you working on on a day-to-day basis right now? What are you doing now? Well, it's complicated. I've taken a job to start my new career. My new career is a computational engineer, and I'm affiliated with MIT, so I'm a research affiliate at my alma mater, MIT.

And my first engagement as a computational engineer is to model geothermal wells. I'm interested in how we can use geothermal energy instead of fossil fuels. And the reason it's complicated is that I've just barely started and I have a long way to go. Doing a new career is not easy. For example, here I am modeling geothermal wells, and you got to know thermodynamics to do that. And I don't know thermodynamics, so I have this big digression to go learn thermals. I'm at a point in my career where I think I'm thinking about retiring. And I was on vacation recently for a week, and I felt like that must be like practice retirement. And then I went back to work, and I realized that it's not about retirement, which is doing nothing. It's about retiring and

then doing what you want to do. What made you want to do this? I mean, you don't have to work, but you thought geothermal was exciting because it could change the world and help us with our power issues. I've tried retiring before several times and I see you tried it for a week. And that's interesting because after a week, you still think it's cool. It's two or three weeks where it begins to get boring. So you better extend your tests a little or you'll get the wrong impression. So having attempted retirement before, I realized I like being on the steep part of the learning curve. It's just more fun. And what I like about the geothermal problem is that it's a very big thing.

If successful, it will. You don't hear about geothermal that much. In fact, it accounts for 0.4% of our energy, but there's a billion years of energy sitting several kilometers down, and we just need to know how to get it out of the earth economically, and life will change dramatically. I'm heading to South Africa for a month, and they're having, of course, a big power issue there with what's called load shedding, where they turn off the power grid on a neighborhood by neighborhood basis. And I'm taking some solar panels with me, pocket solar panels. And I've been testing these just here in Oregon. And I know it's obvious. I know that there are those who are excited about solar and alternative energies. But the idea that I took a big 40,000 milliamp hour battery and I stuck it in my front porch and I put out a

little suitcase version of a solar panel. And I came back later and I had free power. And I'm like, now I can't stop walking around looking at all the wasted power that's hitting the earth from above. Like, why have we not blanketed all of Utah, all of Arizona, all of these places with solar panels, you know, of course the minerals and whatnot. But now when you think about geothermal, and I did some research into this, you're right. It's just, we're sitting on this giant battery of energy and it's just churning around in there but we have to dig deep enough to get to it when you learned about geothermal did that just kind of overwhelm you with like this potential is there and we have to get to it well i was dragged in kicking and screaming by a geothermal enthusiast named jamie beard and she captivated me into the topic. And the nature of

the problem is interesting. And so you say the energy is right there. It's right beneath our feet. The trouble is that drilling is too expensive. We're all working on getting drilling to be cheap, and then suddenly we will unleash this enormous CO2-free, low-cost energy source. Jamie Beard, of course, is the founder and executive director of Project Inner Space, which is a nonprofit focused on geothermal. And she's been kind of working with the Geothermal Entrepreneurship Organization. Why? Is it just because it's too hard to use energy to dig deep?

I mean, we've gone several miles down, haven't we, before? Is it just beneath our ability to drill or drill safely? I have to get realistic. A mistake that a lot of energy hippies make is they think the energy just has to be clean. There's a lot of them in Portland and, you know, oh, clean energy is all we need. Well, no. No, it has to be cheap. You won't get massive adoption until it's cheap. So it has to be cheap and clean, not merely clean. And so there's a thing called the levelized cost of energy, the LCOE.

And we're all fighting to get it. It varies. It's $0.10 per kilowatt hour or $0.20. We'd like to get it to $0.01. My goal is $'s 10 cents per kilowatt hour or 20 cents. We'd like to get it to one cent. My goal is one cent per kilowatt hour. And that's viewed as very unrealistic. But I think when we achieve the drilling breakthrough, one cent LCOE is going to surprise us. It's going to come suddenly. Really? There's a lot of stuff that's been coming suddenly lately, unprecedented stuff. It feels like we could use a win right now, and that would be a pretty wonderful win. Yes, it would. Well, connectivity comes suddenly. And the internet, for example, is 54 years old and signed up two-thirds of the human race, and that's 5 billion people.

You've heard that expression, we tend to overestimate what we can do in the short term and underestimate. We're entering the long term of a lot of things, and they seem suddenly surprising. AI is now surprising. The internet is now surprising. We have so much connectivity now, we don't know what to do with it all. And there's more of it coming. The next generation internet is on the verge. It's happening now. And so a lot of the pathologies of connectivity are attributable to the suddenness of how quickly we became connected.

Suddenly we have all this bandwidth, but we don't know how to manage it yet. So we get these pathologies like hacking and pornography and average odd advertising used to be viewed as a pathology of the internet until we figured out that advertising was going to pay for everything and then spam was left and we don't talk much about spam anymore and now we're on to new things like fake news and censorship and the modern pathologies of connectivity. The interesting thing about pathology, my wife is a nurse and she also had cancer, is that we get it not to zero, but we get it to an acceptable level. Spam exists, but it's largely in our spam folder. And we now have it to a level

that is manageable. I wonder if that's the thing. You introduce a new item, pathology increases, and then you push it down to an acceptable level. And then it's up to society to decide what is acceptable, whether it be inappropriate behavior on social media, or whether it be spam, or whether it be fake news. We get it down to some percentage. It's like, all right right we as a society agree that the cost of entry to the internet is this many banner ads or or this many privacy in the 1970s i worked for city bank on one of the maybe the first cash card and uh you know at gas stations and things like that and i got a i got to look at the business plan for my subsidiary of Citibank.

And there was a line item for fraud. And that offended me because I was an engineer. It was my job to eliminate, make sure fraud was impossible. But the business people realized what you just said, which is you work down these pathologies to an acceptable level, then you move on. And so they had a line item for fraud. It was a single digit percent. And we tried to make it go down and down and down, but fraud persisted. Is that a design philosophy? Because I'm calling back to your idea that you want to have a levelized cost of energy of 0.1, that's certainly aggressive.

But your goals are, isn't the goal to have an aggressive goal, right? An unwavering goal that then pushes you further than it would other people. Yeah, what do we call it in Silicon Valley? A stretch, I think it's called a stretch goal. And it's good to have those well when you made when you worked on with david boggs on the ethernet paper from 1976 was it true that you could move the bits around faster than the computers could handle them so you had more bandwidth than they couldn't be sent or received you had a bigger pipe than you needed initially. Well, right after building the coaxial cable Ethernet at 2.94 megabits per second, I went off with this other guy and we developed a fiber optic interconnect at 100 megabits per second.

And it was completely useless because there was no computer that could take that bandwidth economically and put it into memory. Remember, memory was expensive then. We were using the first DRAMs, the 1103s from Intel, and they were just too expensive to really load up. So memory bandwidth was a constraint in the design of these connectivity systems. Yeah. And now it boggles my mind that my kids, my kids are 17 now, can take for granted our gigabit network at home here.

And I actually have dual gigabit. I have both Ziply Fiber and Comcast running through a router and then 10 megabit throughout the house, which is ridiculous because all we're doing is watching Netflix. It's so interesting that the Comcast of the world try to sell gigabit speeds to people who are using maybe 20, maybe 30 megabits per second. It seems to be a bit of a scam. I wonder who needs that kind of bandwidth. Even myself, I like having it, but I don't really use it unless I'm moving a giant file around. Calling it a scam is a little over the top.

I read the other day that 82% of the traffic on the internet is video. And 82%. And the internet was never designed to carry video, but it managed to evolve. 80% of it is just Netflix, probably. So Netflix is a kind of video. It's one way. But the big, powerful new video is Zoom.

It is two-way. And then we've suddenly been introduced to two-way video, and so I expect that's going to be skyrocketing. We were suddenly introduced to it by COVID, and now we, I remember there were all these professors, I was a professor, and we were all, many of us were denying that we would ever teach a course on video because it prevented us from reaching out to our students and developing the proper rapport. And then suddenly COVID came along, and every single course at the University of Texas was taught over Zoom. So much for pressing the flesh.

Did you teach classes over video? What were you doing during COVID? What were you up to? I was a professor at the University of Texas. And you were doing Zoom classes? Yeah. My last class was 18 freshmen over Zoom on startups at the end. It was in the second semester of the year, their freshman year.

They were all freshmen. And then it became obvious that not one of them had been to Austin, Texas. They were freshmen remotely. And to me, they were just little boxes on the screen. There was one little Indian guy. He kept showing up in the lower right-hand corner, one of my best students. And I was dying to meet him. This little guy over here in the right-hand corner, he was six foot four.

When I finally got to meet him. A different presentation. I work at Microsoft in my day job and I've been here now 15 years, but I have only ever worked from Portland, Oregon. So I was remote before it was fancy and I optimized my house for high bandwidth. I've always optimized my camera for high quality. And I'm always surprised, even after all these years, when I drive to Seattle, how tall these little boxes are when I meet them in person. They're not shaped at all like they are when I work with them all day on Zoom and Teams. Well, an interesting but little known fact is that I taught over video as a consulting professor at Stanford in the 70s. So we had the Stanford television network and local companies in Silicon Valley would subscribe and they would show up. So half my students were in the room and the other half were in the valley sitting in rooms watching TV.

Stanford tends to do things ahead of everybody else, and there's a 70s video network. I think that as someone who really enjoys computing history, and I find myself when I am teaching, going back to history and reminding people that the thing that they think is new and fancy is not as new and fancy as they thought about. I assume you're familiar with Douglas Engelbart's

The Mother of All Demos from 1968? Of course. new and fancy as they thought about. I assume you're familiar with Douglas Engelbart's The Mother of All Demos from 1968? Of course. And this is just an important thing for people to know about where when someone says, oh my goodness, Google Docs and being able to type in a Microsoft Word document and see the cursor of the other person was being thought about by folks at the ACM and by folks at research centers back then. What was limiting us from doing all of those things? Was it memory, like you say? Was it connectivity? Why did it take so long for it to then hockey stick? Well, Engelbart's NLS, as I recall, and I was a youth row that ran on a PDP-10.

And a PDP-10 was a 36-bit word, and it was running about a mega-psych, a megahertz. So there's a limitation on Doug's dream. The reason PARC exploded is we took a lot of Engelbart's ideas and moved them off the PDB-10 onto personal computers where they blossomed. By the way, is it that important about what happened first or who did it back then? Why is that so important? That's a good question. I don't think the who did it first or the light bulb was invented by two different people on different sides of the worlds. What I find to be interesting is that someone had the idea, it was a great idea, but there was a constraint, whether it be an economic constraint, a social

constraint, a constraint of the laws of physics that prevented it from happening like that. And then looking back with the benefit of hindsight, I find it overwhelming to look back and say, there was a time I remember very clearly when there were no iPads. And then I turn around and then there's an iPad. And now I have a pocket supercomputer that I can talk to and ask, how tall is Brad Pitt? Because that's basically all I use my phone for now is when I'm watching a show. Who's that guy? You have trouble remembering how tall Brad Brad Pitt? Because that's basically all I use my phone for now is when I'm watching a show. Who's that guy? That's a little remembering how tall Brad Pitt is. Well, so I was reading an article in Ars Technica a couple of days ago where an individual bought the very last Encyclopedia Britannica from World Book because World Book is the only place that's

making encyclopedias. And they wanted to feel the 127 volumes of this giant book. But there's also a sense of maybe who are we if we're not offline? And I think about that. What does it mean when my kids maybe don't know who they are disconnected? So I try to take them out into the woods and out into a disconnected environment so that they might remember who they are outside the context of computers. That is overwhelming. And I only have 30 years of historical context. I can only assume it's either overwhelming for you or you just choose to not think about it. Forgive the slight digression, but you just mentioned that Worldbook owned Britannica. I think I mixed my two things there. Yeah, but they used to be arch competitors. Did one of them buy the other one? I think I was using it like the way we use Kleenex to refer to tissue. So it is, in fact, World Book that makes the boxes. I don't think Britannica makes physical books anymore, but World Book does. Okay.

So I have three or four of those old encyclopedias, including Britannica. Prior to 1957, so there's no mention of Sputnik in it and the vietnam war had not occurred yet and you know so on but going back to the other question we do it comes up a lot is who did what so when you try to tell the story of ethernet there's a guy from austin Texas, no, the other city, San Antonio, Texas, who was at Datapoint, and they introduced ArcNet, 19-something, 70-something, and that was beforeamax thing happened and Ethernet won and ArcNet didn't. So that's the people giving importance to what happened first and who serves credit for it. And so apparently that is important to people. It may be a source of learning. That is, maybe if you grovel over the events that happened and what came first and what came second, you can learn something about innovation.

That would be a reason to do it. ACM ByteCast is available on Apple Podcasts, Google Podcasts, Podbean, Spotify, Stitcher, and TuneIn. If you're enjoying this episode, please do subscribe and leave us a review on your favorite platform. So given that's a pretty, that's a good attitude. That's a very positive, healthy attitude that you have. What does it feel like when you got a Turing Award, but I noticed in the interviews, you acknowledged immediately, like you're receiving this award, but there's so many people that have been involved in the evolution and the growth of this, but then they choose to say, here's Bob's award. And then you said, well, there's hundreds of people involved in the ever growing standard of ethernet. I thought

that was very magnanimous of you to do that. Well, it's a feature of ethernet. Yeah. And its design is that it's it's part of a uh you mentioned the open source community is there's a different community of which ethernet is a part where and the a feature of the ethernet brand is its susceptibility to new technologies it's like built-in. Backward compatibility and interoperability are two big Ethernet values, and they allow Ethernet to persist.

So I frequently ask, I'm asked, what's going to obsolete Ethernet? And the answer is tricky. Every time a new technology comes along that threatens to obsolete Ethernet, they have a meeting and decide to call it Ethernet. It will persist forever't replace books because they're still books. And the IEEE has over a thousand standards and they're building on top of the land man group of standards. And this is ethernet. And that is the legacy is the building on top of the standard rather than trying to make yet another standard. And still things work. I'm sure that you and I

are talking over a million different switches on a million different brands between my network and your network and all the things in between. And it all just works because we all agreed that it should just work. And not just one Ethernet, but a dozen of them. Your traffic is reaching me and mine is reaching you. Yeah. Even from the 20 foot coax that goes from my switch into the garage, then quickly switches into fiber and then into the ether, presumably, and off into where you are on the other side of the country.

I am impressed that I still have a sense of awe over this, that like every once in a while on a random Tuesday, I'll be on a video call and you're very clear, you're in high definition. I go, oh my God, I'm talking the speed of light to the gentleman over here on the other side of the country. How amazing is that? And my kids are utterly unimpressed. I wonder if I'm, am I supposed to also be unimpressed or is it okay to have that sense of awe? Do you have any sense of awe? My kids are a little older than yours they're in their 30s in fact my son had time my daughter had time to work for eight years for facebook oh wow son coincidentally got to work for eight

years for apple computers so both of them are so immersed in what you and i find to be awesome they don't find it so awesome. It's just like, you know, air. It's just everywhere for them. That's so wonderful. Yeah. There was a young person at work who was having an issue with their Git repository, and it was a carriage return issue. And I think they were in their 20s. Yeah, exactly. So you just facepalmed. And I said, oh, well, that's a carriage return line feed. You know about that? And I said, there's a very funny history about that.

And they said, what's a carriage, and where is it returning to? And I was like, oh. And how do you line? Why do you keep a line? What is that? So I didn't want to turn into professor mode, so I always ask permission before turning into professor mode. And I said, would you like to know?

And they said, all right, all right, I've got 20 minutes. Let's hear it. And we talked about asking, we talked about the carriage on a, on a typewriter and how it returned. And we went on, on, on. It's like, it's 2023 and you're on a Mac or a PC and you're upset about a carriage return. And it's because of decisions that were made in the thirties and forties about how to encode text. And what a wonderful thing that was. And I just loved that kind of history. And they just really didn't care. They just tolerated the whole lecture. Since you raised character, it turned line feed ASCII.

So in 1970, we were developing an internet protocol called Telnet, which allowed you to log into a remote host and some of the hosts had the convention that they only use carriage return they didn't use line feed and others had the convention they insisted on carriage return and line feed and so the the displays were all screwed up when you as you went from the 10x pP-10s or the Multics, which was a typewriter that only had line feed. I forget how we resolved all that. I think maybe we just switched to ASCII and accepted line feed.

Well, I remember doing, when I was in networking class in college 30 years ago, having conversations about different networking cards using Big Endian and Little Endian and whether or not it would be the CPU that would do the flipping or whether that was now built into the hardware. Should the network card be responsible for worrying about Big Endian and Little Endian? Because is the Ethernet, does Ethernet care versus the computer cared? And thinking about those bits flipping on the way off of the network and into the cpu and back and it's in my brain and i know that stuff but i don't think anyone's thought about that in many many many years many years big indian and i still don't understand

the difference well i bought i built a piece of hardware for the ARPANET, and it was the seventh one built, a very low number, so no one really knew what to do. So I made mine hugely functional. So I could have a mode, big N-D-N and little N-D-N and 8-bit, 16-bit. 36 was a very popular word size then, so I had a register where you could specify the word size, and it was full duplex. It could send and receive at the same time.

And the card was three feet square to do all that functionality. So part of the progress in making them cheaper was eliminating all that generality and finally settling on bytes instead of 36-bit words. That's the feature. Oh, well.

Yeah, that's funny. Even now, I see an article from IBM, which is Indian-ness guidance for open source projects, where they have a whole conversation about thinking about Indian-ness because you're coding usually for little Indian processors. That's basically the way things are working now, but people still have big Indian support guidance. So every once in a while, you'll hit a bug in something. Like a couple of weeks ago, I was dealing with some home networking stuff and I had to think about jumbo frames. And I hadn't thought about jumbo frames in many years. It's always an abstraction until it leaks, isn't it? Yes. We still have jumbos.

I guess most are. That's the wrong way to say it. The jumbos are the new thing. Right. Jumbo frames can do 9,000 bytes, apparently, in a frame. Yeah, I don't remember the numbers. Yeah. But I do remember setting the maximum packet size welcome to your new computer and max out the packet size because you need to do that because the defaults are wrong you know what the

consideration was what's that i don't know it may surprise you. The Alto personal computer had a disk. And we imagined that that disk, you'd want to put data on it, and you'd want it to go really fast, as fast as you could. And that would be the dominant method of transfer, would be disk-to-disk over the network. Well, there were sectors. The disk had sectors. And so the maximum packet size was set to be large enough to contain a sector plus the metadata, the overhead associated with that sector. And that's how long.

So we made the packet size as small as possible, but still capable of carrying a sector at a time. And that Alto had a two and a half megabyte one platter disc cartridge. Yeah. Well, you put in the pizza oven. That's so fun. I know I've mentioned this a little bit before, but I am finding as I am now in getting into my fifties that knowing this stuff is overwhelming. And I wonder if there's what to do with understanding the full stack or whether I just let it go and accept that these abstractions don't matter

anymore, at least for people who aren't hardware designers. I'm sure that for networking engineers, they always will matter. But as a user, I keep wanting to say, like when I learn about chat GPT or AI, I want to walk the stack all the way down and I want to understand what's happening. So how far down in the stack do you go in your automobile? I love that you did that because I use that analogy all the time because I'm making my kids drive stick shift, but they want to just take an Uber. An Uber, that's even higher level than an automatic.

They just want their body to be teleported. And Uber is the closest thing we have to teleportation. But I want them to build a kid car. Let's make a car from scratch in the garage, right? But then you start getting ridiculous. Like, well, you have to smelt your own iron? Are we going to become blacksmiths? How far down do you go?

And where are you going to get the oil from? I'm going to have to drill into the geothermal at this point. But then you'll need a hard metal to go into the granite, so you'll probably need to do some metallurgy. And now I'm at the Renaissance Fair trying to figure out why it's so hard to make an internal combustion engine. Yes. So anyway, you need to apply that same thinking to the computer stack but apparently you and i are stuck on that there's something special about that stack and we want to know about that stack all the way and then there are these programmers they're called full stack programmers is that what it's

called yeah full stack engineers my son's a full stack he does both he does javascript on the server but also in the client and that makes him full stack javascript on the client and the server i'm gonna ask him about big indian and little indians and you should do that because i'm over here over that i mean it is possible to live a rich full life without running into that concept and that right there i love that you said that it is possible to a rich, full life and never know that stuff. Because my dad made me learn how to change my own oil and change my own tire. And then he gave me a AAA card. And AAA has never failed me. And they've always changed my tires. But there was a joke that a friend of mine told me that he says, there's not a lot of software engineers on The Walking Dead. So when the zombie apocalypse comes and the zombies

eat us all, it's always the preppers that have the ability to chop wood and build stuff and blacksmiths. And there's no one there to reboot the router because that skill doesn't matter anymore. So I think my dad tried to prep me for the zombies teaching me to chop wood and carry water when I am just obsessed with the computer going all the way down to the bottom. I bet you Portland has that in common with Maine, which is there's a lot of people up here who do five different jobs. That is very true. That is very true. There's even digital preppers I learned out about this recently where you can get a Raspberry Pi and a couple of terabyte disk and go do what's called the internet in a box. So you download the Wikipedia, Stack Overflow. You download several terabytes.

You put it on a disk. Then you fire up the Raspberry Pi and you have a little intranet that presumably you would run when the zombies eat us all. And you'd have a little local network and you could look up. You could Google for stuff except on your little Raspberry Pi, and that'd be a way to digitally prep. Okay. I better get working. You should get on that. So as we get towards the end of our chat here, and thank you for rambling with me and playing, I do appreciate it. Do you feel that you're going to see the breakthrough that you want in geothermal in your lifetime,

you know, in the next 30 years? Oh, yes. Jamie and I, Jamie Beard and I worked together. We founded GEO at the University of Texas and were funded by the Department of Energy to do startups that do geothermal. And our goal was this decade, commercial scale this decade. And that implied something that was complicated. In order to get scale, Jamie figured, and I came to agree, we're going to need the oil and gas industry to help. They have scale. The trouble is the hippies running geothermal hate the oil and gas industry, and that's counterproductive.

So one of the goals of geo, supported by the department of energy was to encourage startups in geothermal, not 2050 or one of those silly years that they keep naming. No, in the 20s. In the 20s. In the 20s. And you think this is going to be an overnight success in 40 years where it's not in the zeitgeist right now. Like, you know, my parents aren't thinking about geothermal, but then the stories will start happening and it's going to pop like that like we've seen ai pop well we saw the internet pop in 1995 when the worldwide web came public through netscape i guess it was 93 or 4 it popped uh you know suddenly the there was a reason to have to have it all and it

it popped. So I've been around for a few. Personal computers popped at one point. The internet popped at one point. And AI has popped four or five times. Marvin Minsky was my undergraduate thesis advisor. I knew about AI back in the 60s. And it popped, and then it popped again.

And it's popping now. And it might break through this time my theory of the past pops was that the uh ai ran out of data but now we have the internet so the ai really maybe doesn't have to run out of data maybe there's a lot enough data to feed its growth incidentally here's a fact. Microsoft, your employer, has placed an order at NVIDIA to buy every processor they make. Yeah, it's insatiable. They want all the GPUs. And they want it for chat GPT. And so AI is cycle starved. I also claim it's connectivity starved. We have 10 to the 11th neurons with, on average, 10 to the 4th synapses per neuron.

So that's 10 to the 15th. Whereas GPT-4 has a trillion, it sounds big, a trillion parameters, a trillion synapses. But a trillion is only 10 to the 12th. So there's a 10 to the 3 difference between the connectivity of the human brain and of the CPTs. So AI is connectivity starved, it's compute starved, it's probably memory starved too i don't know or that's the infiniband networking standard that they use to talk between all of the things when they are training these models these can't be can't be too fast it just need they need more and they need more gpus and they need more speed but infiniband is 400 gigabits or something like that well ethernet's 800 now yeah i met with jensen huang who's an awesome guy and i got in to agree that the power of his products is is not in the g's it's in the pus no not in the pus it's in the PUs. Nope, not in the PUs, it's in the Gs.

Processing units are beside the point. It's the connectivity among them that has made the difference. Indeed. And we got to the subject of InfiniBand, and he was a little embarrassed to be talking to me because Ethernet has been a constant enemy of InfiniBand. And he seems to think that his own company, NVIDIA, has chosen InfiniBand over Ethernet. And I think he's about to get surprised. Well, thank you so much for spending

as much time with me as you did, sir. I very much enjoyed this conversation and your time. Oh, a great honor. Thanks for inviting me. We've been chatting with Dr. Bob Metcalf, the recipient of the ACM Turing Award, recognized for the invention, standardization, and commercialization of Ethernet. He's now working on making geothermal happen, and I'm looking forward to hearing about that soon. This has been another episode of Hansel Minutes in partnership with the ACM ByteCast, and we'll see you again next week. ACM ByteCast is a production of the Association for Computing Machinery's and we'll see you again next week. That's B-Y-T-E-C-A-S-T.