Dialogue on Open Source,Makers,and Industry Innovation: Eric Pan and Eben Upton | Maker Forum

Since changing? It's fascinating in a lot of ways. In a lot of ways it's still the same. You still got that kind of energy. The thing I've always appreciated about being here is how fast people do things. The time scale that people do things on is days and weeks, not weeks and months or months and years. And you can see that energy, that energy is still here. I think, obviously, you mentioned the geopolitical situation. What's fascinating is that doesn't seem to really be hurting it. Maybe there's a sense of distance between America and China or between Europe and China now, but it doesn't seem to have hurt the energy here, which is super exciting. So I guess, like a lot of our communities, it's the same but more. Another three years of Moore's law, your Wi-Fi modules now cost half, everything costs less, is more powerful. That's fascinating. I think when I started 10, 15 years ago, my assumption was we were coming to the end of that era of explosive growth, and it's kind of wonderful to see that still happening. Wonderful to see the interaction of that kind of explosive growth at very short time scales.

Yeah, I mean, that's obviously the biggest change since I was last here. I guess we were probably shipping Pi 3 when I was here last. Pi 3 was out, and you've gone through Pi 4, Pi 5, so you've got roughly a 10-fold performance increase. When people think about the big chip, the BCM2712, that's now roughly 10 times the performance. But for me, as somebody who's been making chips for most of my adult life one way or another, it's super exciting now making our own silicon. Obviously we had RP2040 appear in 2021. Perfect time. If you could have picked a time to launch a microcontroller, the year when nobody else had any microcontrollers was probably a good time to do that. That's been a wonderful thing for us. It's just the sense that we'll talk a little bit about maker businesses. One of the pain points for maker businesses over the last few years has been the supply chain environment. It hurt us, it certainly hurt many of our customers. We saw maker businesses who effectively could not do business for a period of two years because they had no access to semiconductors. So to arrive with a semiconductor product that sustained a lot of these maker businesses through the hard time was super exciting for us. It's been really touching the way that the RP2040 platform has been taken to heart by the maker community, maybe the way that a generation before they were using Arduino or Microchip parts. It's become quite central to the making movement. And then JLCPCB, which is this interesting organization, is a hub for the maker movement. That partnership that we have with them around helping people build those products at scale has been wonderful to plug the Shenzhen time scales into the growth of the ecosystem.

Obviously Raspberry Pi is an interesting platform because for the first time we have RP in a big chip. The RP1 South Bridge we announced – I hope most of you are aware – we announced Raspberry Pi 5 at the end of September. We are now manufacturing at scale and shipping them. It's the shipping at scale that deserves the woo. The amount of effort the team has been putting in. This product is roughly 150 times faster than the Raspberry Pi 1 from 2012. And for the first time, we have three custom chips on there: a custom Renesas, a custom power management chip, another generation of Broadcom BCM2712 of course, but then for the first time we have the Raspberry Pi-ness of the Raspberry Pi. If you think what is the Raspberry Pi, we've always seen ourselves as a PC company, but we know that most of our customers see us as an edge computing platform, an interfacing edge computing platform, for which the interfacing capabilities of the Pi are extremely important. What we've been able to do is take those interfacing capabilities and crystallize them into this RP1, 20 square millimeters of 40-nanometer silicon that we designed at Raspberry Pi. So you can see that certainly we're becoming more of a semiconductor company to the extent that it allows us to build more interesting electronic products. The question is how much of our business is going to be about actually selling chips. I don't think we're really planning to sell RP1 to other people, but the microcontrollers – certainly in terms of volume sales to third parties – we're probably consuming a million RP2040 ourselves for Pi and accessories, but we're selling four, five, six million units a year to third parties. So you can certainly see in terms of volume, that bit of our business is looking a lot more like a semiconductor business than an electronics business.

So there is that idea. I think we are a very end-to-end company. Even the silicon – the silicon we get from Broadcom, some of the IP, the video decoder or the camera processing block in BCM2712, that was developed at Raspberry Pi and provided for integration into the circuit. So you can see we're doing everything from semiconductor IP development, chip development, electronic product development, software, community, ecosystem management. And of course, if you stretch out into the foundation, where teacher training, teaching kids – you're doing that whole thing. And now two physical shops in the UK. So you can see we have this end-to-end orientation in the business. I think we do it for the same reason that Apple does it, which is control. If you want to make really great products, there's no substitute for touching every stage of the design process. You can optimize and optimize. If you look at a Raspberry Pi 5, we have a chiplet architecture. What today would be called a chiplet architecture, where you split the functionality of what would effectively be one chip into two chiplets. Those two chiplets, BCM2712 and RP1, are connected together by a four-lane PCI Express bus. If you look at the PCB, you see freeways, motorways of lanes, parallel lanes between the two chips. Why are they parallel lanes? Because we control the ball-out of the two chips at the two ends. We control RP1 ball-out because we designed it. We control BCM2712 ball-out because we have a very close relationship with our supplier. That means because the positions of the balls match to allow you to have parallel lanes, you don't spend PCB trace swapping lanes around. You have a cheaper PCB. So that level of control across the entire stack, and then down at one end designing the video codec IP that decodes video frames, and we're writing the operating system at the other end of the stack. We can arrange the pixels to come out of the video codec in the format which makes life easier for the software. If you control all this stuff, you can build better products, you can build cheaper products, you can build more.

The strap line on our website when we present the company to people now is 'Computing for everyone'. If someone adds two numbers together, I want them to do it on a Raspberry Pi platform. That's the vision. We really want to provide the underpinning for all the work you do, for all the work that all of you in this room do. We want to provide the computational substrate, regardless of whether you want to build a little $1 device on top of RP2040 or you want to build a $500 device and put a Compute Module 5 into it. We want to be the computational substrate for people's work, for all your work. We want to take away – there's an enormous amount where we succeeded, I think, where we've helped people stop doing things they don't want to do. A lot of you, we've met a lot of companies over the years, a lot of our OEM partners who are making hardware, making Linux computers basically in small quantities, thousands, ten thousand a year, in order to run some piece of software. We see this a lot in thin client, all digital signage thin client, where a company has a software platform and their differentiation is in software, yet they're building hardware because they need somewhere to run their software. What we've allowed these companies to do is stop building hardware and just use our hardware and focus on what they're good at. That's the aspiration, to just be an enabler. Obviously we have experience now being an enabler for people making electronic products. The question really now is can we do the same for semiconductors?

The next thing, same five. Can we talk about Pi 5? There are a lot of improvements, a lot of changes, advancements. What will be the most important change from the previous version that enables new applications?

Oh, that's a very good question. I don't think we're deprioritizing education. I think what we're seeing is we've got to acknowledge that our business is now an industrial business. 70-80% of Raspberry Pis now sell into industry. Industry doesn't necessarily mean 100,000 units to Siemens or someone. Often it means selling 100 units to a maker who is going on that journey from making one of something for yourself to making a hundred or a thousand to sell to other people. So we've got to recognize that change in our business. But education is super important. I think we will try and do a lower-cost SKU of Raspberry Pi 5. I doubt we'll do a $35 Raspberry Pi 5. If you look at the cost structure of the platform, some of the work we've had to do to get about a three-fold performance increase over Pi 4 – some of the things we've had to do to enable that are expensive. But I really don't want people to see it as a deprioritization of education. It's the addition of a new focus on industrial. And of course, on Raspberry Pi as a PC. We've always told ourselves that Raspberry Pi is a PC. What we're making is a PC. It's always been true, but only Raspberry Pi 1 was a PC for 1% of users who just want a little bit of performance. Each generation has broadened out the pool of people for whom Raspberry Pi is a PC. Raspberry Pi 4, I think, was a PC for the average user. During the pandemic, we shipped a lot of Raspberry Pis to disadvantaged kids in the UK who didn't have computers to study at home. It was a good enough PC for them. The fun thing about Raspberry Pi 5 is that it's a PC to me. I'm a reasonably power user. If I sit in front of a Raspberry Pi 5, I forget that I'm not sitting in front of a traditional x86 PC.

It's an interesting flight. Certainly, it's one of the reasons why we built semiconductors. The observation that Arduino makes money for their chip vendor but not for themselves – out of the classic, I mean, Atmel and stuff of course is a different business, but the Uno business made a lot of money for Microchip. So we sort of saw, well, why would we never do a microcontroller product before? We never did a microcontroller product before because we just would have made money for our microcontroller partner. You always chose the hard road. We definitely chose the hard road. If the semiconductor company is going to make the money, the solution is to be the semiconductor company. As you said, we chose the hard road, and it was a long hard road, but it was worth it because it means I don't mind when people say, 'How do you feel if people select a JLCPCB device instead of a Pi?' That's why I don't care, because they're still my customer. Either they're my customer directly or they're my customer through JLCPCB.

It's slightly neither of those. We talk about the other fruit company, the one in Cupertino. There's a way of thinking about Apple which is that in their classic era when they were run by Steve Jobs, they were really a company that built the things Steve wanted. And it turned out there were just a lot of people who wanted the things that Steve wanted. That's a powerful way, and that's how we design products. We make the products we want. So it is slightly the community, but it's the community because we are community members. We're building the products we want, and because we're just like you, often the things that we make for ourselves you want to buy. I think it's a really powerful way of designing products. If you compare it to a strategic view, a marketing-led view, a focus group-led view, it's really powerful. I have some experience from elsewhere in my career of what happens when you try to design products by marketing strategy. It's very easy to design products that nobody wants because you say, 'Well, this guy told me he wants this thing, this guy told me he wants this thing, and this guy told me he wants this thing, so I'll make a product in the middle that hopefully will cover all those people.' By trying to appeal to all those people, you appeal to nobody. When you design a product, you want at least one person wants it. And if one person wants it, then maybe two people want it. That's how we design our products. We just look at the things we want, just as a bunch of geeks in a room building things we want to build.

It's enormously important to us for a few reasons. One, as I say, there's a huge sense of speed here. This is the most demanding market, the least giving market in some ways. People have the highest expectations in terms of speed and availability and agility. If you want to succeed in the Chinese market, you've got to have products that people really want. And you have to remember how insulated we are in the West from competition. Thinking about the semiconductor space, you can buy the most incredible semiconductor products here and nobody in the West knows about them. You have Western semiconductor companies who are selling absolute garbage compared to what you can buy very cheaply here if you're embedded in the local community. So I think it's important to us as a stress test. One of the things I'd be most pleased about with RP2040 is that it's been able to succeed not just in the West where it's competing against high-margin, high-cost, old-line semiconductor companies, but that it can succeed here in China where it's competing in a much more brutal environment. People still appreciate it. So it's important to us as a kind of extreme example of an environment for selling products. It's important to me because there's enormous domestic demand for all of Raspberry Pi in China. So it is a hugely important end market for us. And finally, it is our second largest market in terms of total sales after the United States. But those sales are a mixture of domestic demand – although given the population here it's clearly underpenetrated, it should be five or six times the size of the United States – so it's clearly underpenetrated. But we also have systems integration demand. As I say, the vast majority of Raspberry Pi use is industrial and embedded. If you're embedding Raspberry Pi power into a product to sell in America, where are you doing that embedding? You're probably doing it within 100 miles of this room. So there's a huge amount of systems integration business here. And it's a market that's increasingly ours. We have a lot of fantastic partners here who help us address that market.

I don't feel particularly lonely.

I think we are a very strange company. We're very British. We don't have a lot of people outside the UK, and we don't have any first-party representation in China because we have this longstanding relationship with element14 who provided a sort of virtual presence. I do think it's getting towards time that having somebody west of Cambridge – east of Cambridge – would be nice. We have a few people in North America now, but having somebody east of Cambridge might be quite nice. It's always a balance for us and for the foundation, because remember we're two different organizations. The foundation was my shareholder, the foundation is the thing I founded that now owns most of my company. It's a balance in terms of trying to be scalable, trying to be able to address all different markets around the world without having hundreds and hundreds of people, versus this challenge of how do you support the community, particularly a very vibrant community we have here in China that could be bigger. I love the thing about asking people if they've used Raspberry Pi. The question I used to ask before – when I was hiring people, when I was getting students into Cambridge or hiring people to come to Broadcom – the question I used to ask was, 'Have you ever written a computer program that you didn't have to write?' If you've only ever written a computer program because someone paid you to do it or because you had to do it for your university course, I probably don't want to employ you. That sign that you're in some way elite – the maker movement is this surprisingly elite movement of people who are doing deeply technical work that they don't have to do. It's incredible. All of us in this room are doing something really very unusual and incredible. We should be proud that we do that. Maybe go watch the football as well, but at least we take some portion of our time and spend it doing this self-consciously elite activity.

Well, I guess I'll ask you this question, but I'm kind of asking the community question for you, I suppose. Which is, what can we do to be helpful? What are we not doing now? You mentioned obviously better local support for the community. What else could we be doing, either in respect to the community or in respect of technology, to help the broader Chinese maker community get to scale?

I think it's a very good start coming to Maker Faire. Beyond that, I think a lot of engineers after using Raspberry Pi now go into product design, design the silicon inside their products. I think if we have more Chinese documentation support, a community with official presence, it's going to collect more ideas or suggestions. I think also people want to expose themselves. If they have done some great work, but maybe it's a language barrier, maybe Chinese are more shy than others. So maybe Raspberry Pi could be a platform not only to buy from but also to show off.

Yes, absolutely. Explaining to parents why letting their children be part of this thing that we've created together – explaining to parents why the maker community is important, why there's more to life than exams. We have some young makers here. Explaining why it's important to play. While it may look like play, it's developing things inside your mind. It's important. I don't have a good solution for it, but I would say that the people I knew as a child who've been most successful as adults have been the people who played most.

It's interesting. 30 seconds on Raspberry Pi and AI. Raspberry Pi doesn't have an integrated AI accelerator, and it probably never will. Why doesn't it? Because I think it's only of interest to a small number of Raspberry Pi users. It's an expensive feature to provide in terms of silicon area, so you are taxing 90% of people to pay for a feature that helps 10%. So I think we do two things. We try to put fast CPUs into our products so you can run neural networks. I think the Llama 7 billion coefficient model will do about a token every 10 seconds on a Raspberry Pi 4, and about a token every three seconds on a Pi 5, which is not super fast, but you can do some work. So the idea is some workloads can run on the CPU. If you want to run a workload on the CPU, our goal is to be a good host for external accelerators. In the good old days when people said AI, if you said AI to me, you were talking about CNN, image recognition. The Coral accelerator – we try very hard to be a good host for the Coral accelerator. So I think you'll see faster CPUs, PCI Express, and USB 3 to talk to accelerators. It's an exciting thing. I think really what we've got to do as a community – because we know what we're talking about – I think we need to be careful. There's a bit of danger with LLMs. People think that it's going to make computer programming not be a thing anymore. Anyone will be able to write a computer program because they'll just be able to say to ChatGPT, 'Write me a computer program.' This is insane. It's insane because telling ChatGPT about your computer program is just computer programming. So it's crazy, but it sounds plausible. There's a real risk for those of us who look at education policy that governments will deprioritize teaching people about engineering, making, computer programming, because they will imagine that LLMs are going to make that go away as a discipline, rather than making computer programmers more productive. The historical lesson is when you make computer programmers more productive, you increase the demand for computer programmers. But we're going to do our best as an engineering company, and increasingly the foundation is thinking very hard about how to position AI within the computer science curriculum.

Okay, the shortage. It's been a horrible couple of years. We were in shortage for basically 25 or 26 months, from about April 2021 to about June or July of this year. We are now back in stock everywhere, which is great. Obviously Raspberry Pi 5 – people keep buying them, we keep making them. We had great support from our silicon vendors all the way through the shortage. It was never that we couldn't make Raspberry Pis; it was that there was a lot of demand. In fact, I think what had happened early on was people had moved to Raspberry Pi in expectation that we were safe, and of course we weren't safe. It turned out we made last year the smallest number of Raspberry Pis since about 2017. So it was very painful for us. What can silicon companies do? Our suppliers were very good to us. It would be great if they'd given us twice as much silicon, but it's very hard to go to your supplier and say, 'Please double my silicon supply.' You're lucky if you get what you got last year. I think some other companies did treat their maker customers badly. I know maker businesses – I talked about 2040 saving maker businesses. Microchip, NXP, ST – they just threw their maker customers under a bus. I think that's a shame. What could they do better next time? Hopefully there isn't a next time. This last few years was so bad, probably as bad as it's been since the invention of the point contact diode in the 1960s. It's never been that bad. But next time there'll be cycles, and I hope that next time people remember that the maker community is extremely important as a route for market access. I think some of those vendors have permanently damaged their position in the maker community, and that's a shame because they have good products. The availability issue in non-traditional markets – my chief commercial officer is here this weekend. He had a lovely trip to Ghana and India with me last year. He gave a talk in Nairobi. He said that for 30 years he'd been in senior roles in distribution, all with 'global' in their title, and the first time he went to Africa was January last year. It is incredible the extent to which the distribution community did not see these countries. Africa for distribution is South Africa, or maybe a little bit of Francophone North Africa. There's a whole block in the middle that people just don't see. We're actually building Pico – we make Pico in Nairobi now. We're probably going to move all Pico production to Nairobi. Why are we doing that? It's not quite as economic as Asia, but we have a contract manufacturing partner there and we're trying to bootstrap their relationship with distribution. If you have a few million dollars of money flowing through there and they're buying resistors and capacitors, then they're going to be more visible to distribution. We're building out a distribution capability in Sub-Saharan Africa. Once we've done that, and we have a little more work to do in South America, once we've done Sub-Saharan Africa, really all that's left is a relatively small number of countries. I think it's another area where the Shenzhen community, particularly the Shenzhen community, can be helpful. I went to see building industrial agriculture robots in Ghana. They built a robot for planting seeds and had a complicated PCB with technology. I said, 'When did you get the PCB?' They said, 'Well, JLCPCB, when else?' So there is a sense that the more aggressive, more enthusiastic engagement of China with Africa, compared to the UK or North America's engagement in Africa, can be a vector, a route by which African countries can access state-of-the-art technology that otherwise they'd be stranded from. People sometimes talk negatively about China's involvement in Africa, but I think it can be an enormous force. Look at the amount of infrastructure in Kenya that's been paid for with Chinese money. It can be a very positive force.

So what was the first one? How do I see it playing a role? Just like the cross learning, the opportunity for art and design to come to engineering. Yes, so this is interesting. I am an enormous geek, and I've been an enormous geek since I was five years old. If I hadn't had a chance to learn about computers in my home and school organically, you could have got me interested in computers with a thought about is and functions and variables. I'm interested in computers for their own sake. That makes me part of about 1 to 5% of the population. The rest of the population, if they're interested in computers, if we're going to get them interested in computers, we have to position computers as a way of solving a problem that they have. They're not interested in computers for their own sake. They're interested in computers as a tool, as a screwdriver, as a hammer, something that solves problems. I am always more excited when I see a Raspberry Pi in an art classroom than I am when I see one in a computer science classroom. Because what it means is that somebody is seeing our product as a way of building an art installation, as a way of measuring something for a geography class. These non-traditional uses of computing are super important. By providing low cost and good support material, I think there's a good chance we can encourage that. Your second question is about economics in developing countries. I see Raspberry Pi as a PC. In our developed countries, we have a lot of PCs. Although my country's government, which is made of rich people like most governments, sent all of our children home in March 2020 to study from home. 700,000 families in Britain had their children sent home with no computer at home to study. We had children trying to study on their mother's mobile phone, trying to do a day of school work on their parent's mobile phone. A family of five with three kids huddled around. So even in Britain, there aren't enough PCs. In the pandemic, we were privileged to be able to use Raspberry Pi to address that shortage. The same effect is magnified in the developing world. There are enormous opportunities. Most of us in this room were very lucky to grow up with access to computing hardware. There are enormous opportunities available if you have access to a computer in the developing world. Sometimes 1%, 2%, 5% of people have PCs. So Raspberry Pi could play a role there by reducing the barriers to accessing this incredibly beneficial technology that has benefited all of us.

Yes, well I'm really glad to hear that. That's the business we're in. I mentioned industry is enormously important to us, but it is still the case that trying to change people's lives at a young age – I have people working for me now who got their first experience with Raspberry Pi. It's a closed loop system now. It's the most exciting.

Cool, I do recognize this. Wow, there you go. Look at this. Got the Pi. I see Mr. Computer. Oh yeah, there you are. This is the computer. BBC Micro.

My favorite bit of launching new Raspberry Pi products – favorite but also terrifying moment – is we do a credits list. If you go to the Pi 5 launch blog post, there is a credits list. But this is the first launch where the credits list got so long that Liz made me put the credits list on a separate page. Over 500 people. It's wonderful. When you do a big something like this, or something like the Raspberry Pi 5, and you write down all the people who helped – I really would encourage you to do if you launch products: do credits. It's terrifying because the one guy you forget will hate you a lot more than the 100 people you remember. But if you can get it right – what I did this time was I outsourced the list to other people at work, but I also have a backup for any forgetfulness. This is wonderful. Thank you very much.

Wow, well I'm going to take this over. Please pass all my thanks. And we'll show them the radio. Okay, thanks.