Interview with Eben Upton

The past 12 months have been pretty eventful for the Raspberry Pi Foundation – with a number of new hardware releases (Pi NoIR, Compute Module, Model B+), a new website, a huge push on educational materials and outreach, a number of new employees and even a visit to Buckingham Palace! We talked with Eben Upton about the recent changes and what they mean for the Raspberry Pi.

Raspberry Pi Geek : Has the runaway success of the Raspberry Pi allowed you to expand your charitable and business goals further than you had originally thought possible?

Eben Upton: Yes, I think so. Obviously we are doing much more engineering than we were expecting to do, particularly more software engineering. We have been able to devote a lot more effort to things like the display board and to getting things like the Compute Module to market. The Compute Module was never really even on the agenda early in the Pi's life cycle – that one really is James Adams' brainchild, and it was something that he was very passionate about from day 1, so it is nice to see that go into the market and be really successful. Then, on the charitable side, we have been able to hire more people [and] produce more material, and we are obviously running things like Picademy, the free teacher training program, which has been very popular. We certainly didn't see ourselves getting into teacher training when we started doing this, but it was something that Carrie Anne [Philbin], when she joined us, really wanted to do. Quite a lot of these nice things that we have been able to do are things that people have come into the organization wanting to do. Ben Nuttall, for example, has also developed the GitHub-based documentation and resources section of the new website.

RPG: The recently released Model B+ brings a number of improvements over the Rev 2 Model B. What do you see as the most important upgrades here for the average Raspberry Pi user?

EU: The interesting thing is, I think that they are all kind of equal weight. Every change in the Model B+ is actually a fairly small change – but there are six or seven of them and when you add up six or seven small changes, you end up with something that is really big. Obviously the additional USBs are very useful for a lot of people. There have been a lot of people who have had to plug USB hubs into the Pi and so the combination of more USBs and the better behaved USB power chain are two things which together have made quite a lot of difference. Obviously the extra GPIO pins are great for people doing hardware hacking and the slightly prettier form factor should be beneficial for people working on case designs and for embedded projects.

RPG: Were most of the improvements and changes based on requests and feedback from the community and educational establishments, or were they mainly internal ideas?

EU: I think they were all feedback really. Obviously the form factor tweaks I was quite keen to do, and the power consumption tweaks were something we also really wanted to do, but in general it was mostly about user feedback.

RPG: With a board like this, it is always a balancing act between the functionality of the board and the cost of the components to keep the price down for the end user. Did you consider any additional upgrades such as on-board WiFi and Power over Ethernet (PoE)?

EU: I think both of those are really good examples of useful things which, when you price them up, are barely worth considering as they are very expensive. They are both things that we would love to have in the future, but right now it doesn't make sense. There wasn't anything not already on the board that was in the 10 to 20 cents range that felt useful. We mined out all of the things that cost 10 to 20 cents that felt useful, and the remaining things that feel useful are all kind of several dollars, and that is where we hit a brick wall on the ability to put stuff in. WiFi and Power over Ethernet are great examples of things we would love to put in a Raspberry Pi but just can't do at the moment given the price.

RPG: Was the release of the Model B+ specifically timed to be before the new UK Computer Science curriculum comes into force in September 2014, or was that just a coincidence?

EU: No, it just happened that way. This is just how it fell out of the inventory planning. Obviously it is challenging to introduce a new product like this – it requires a lot of notice. RS and Farnell had pipelines with the manufacturers of Model B orders, so we had to plan really carefully to make sure that they both finished their Model B orders at the same time and were able to transition to the Model B+. Effectively it is just a number that came out of a spreadsheet, but the timing is very convenient for us.

RPG: The Raspberry Pi obviously has a very complicated and intricate PCB (printed circuit board) design in order to fit into such a small, neat package. Was the Model B+ update a challenge in that regard or relatively straightforward? Did you start from the ground up or just tweak the old design?

EU: It's a ground up redesign, with the old design in hand. Pete [Lomas] had done a lot of extremely cool stuff with the Model B, so it was a very helpful thing for James to have that stuff to hand when he was creating the Model B+. A lot of this was surrounding the BCM2835 chip and how you get signals out of it. The BCM2835 is a 0.65mm BGA (ball grid array) so most of the sites in the grid are occupied by balls, so it is challenging to route out. So I think it was useful to see what Pete had done and look at ways of improving it. But certainly out of the board level, it is a ground up redesign starting with a blank PCB and working from there. The Model B+ for example was a mounting holes first design where the first thing to go down on the board was the mounting holes and everything else was built around that structure. Whereas on the Model B it was kind of a mounting holes last design as you can see by their placement in the space that was left over.

RPG: With the Raspberry Pi being so successful and the buying power of components therefore being higher than expected, did the economies of scale help to allow features in the Model B+ that were not originally possible in 2012?

EU: Yes, it is certainly a lot easier negotiating component prices when you are saying we need 200,000 of these a month, than it was when we were saying we would like a few thousand of these to get started with and we might give you some volume later. It is a very different experience now. A great example would be the SD card connector. We had no real option to improve the pricing of the old SD card connector because it was a fixed element of the design with a single source. However, when we were looking at new SD card connectors, we were able to go and have a conversation about significant volume which allowed us to move to this much better microSD card connector, with no significant cost penalty.

RPG: Some of the more eagle-eyed Raspberry Pi fans may have noticed that earlier in 2014 you updated the PCB design on the Model B and moved around a number of the minor components, as well as adding a LAN9514 USB/Ethernet chip and updating the CE/FCC device classification to Class B. Were these changes purely for certification reasons, or was it also a sort of proof of concept for some new Model B+ circuitry?

EU: Mostly component supply issues. Certainly the use of the LAN9514 was – you can use the LAN9514 in the place of the LAN9512 as long as you tie the hats off right. You tie the unused bits off and it just deconfigures those, then when it starts up it probes and effectively behaves just like a LAN9512. It was certainly useful though to move to the LAN9514 a little bit ahead of time. The change, which I think came in, in February, was a component supply issue. We had hoped to get all the way through to the Model B+ launch without changing the Rev 2 Model B but in the end there were some DFM (design for manufacture) changes that we needed to roll up, for the sake of about four months that the updated product was in the market. It was worth it, as we were still going to do around 800,000 in that time before changing over to the Model B+, and of course we still make the Model B as there are still industrial customers who want them.

RPG: Is the Raspberry Pi Model B going to be discontinued in the foreseeable future now the Model B+ has been released?

EU: Yeah, once there is no demand, it will be discontinued. While there is demand, and there is still very significant demand, it will stay in production – and it is nice to have a DFM tweaked version of the Model B so we know we can be comfortable manufacturing it for another year or so if we have to. Most people should really start to switch over to the Model B+ though.

RPG: On the official Raspberry Pi blog, you recently introduced HATs (Hardware Attached on Top) – the new standard for Raspberry Pi add-on boards. What was the motivation behind this move, and what does it mean both for add on board manufacturers and for the end user?

EU: The motivation was to make it a little bit easier for the end users to use add-on boards. The HAT is two things – a mechanical specification and an electrical specification. We thought very hard about what makes a physically robust top board and then we have written it all down in one place and given it this stamp of approval that you can call it a HAT as long as you build it in the best possible way. On the electrical side, there is this self-identifying EEPROM system where you have DT (device tree) blobs which allow you to automatically configure the system to load the device drivers you need for your top board. So it is a simplification and a standardization thing, and we hope that most people, unless they have a compelling reason not to, will choose to build HATs rather than building ad hoc top boards for the B+.

RPG: Was this HAT specification something you planned from the outset when designing the Model B+ or something that was introduced more recently?

EU: Yes, it was another thing that James Adams and Jonathan Bell were keen to put on the Model B+.

RPG: Are there any plans for a similar update of the Model A in the future, a Model A+ perhaps?

EU: The Model A obviously hasn't sold as many units as the Model B and just as the Model A is derived from the Model B, there will be a Model A+ which is derived from the Model B+. I don't think it will be a particularly significant upgrade except in so far as you get the extra GPIO and the better power consumption but you will still only get half the memory and the single USB port. Additionally you will get the form factor compatibility with the B+ which we think will be useful for people. I think we need to be a bit clearer about articulating why the Model A is the Raspberry Pi you want. There are a lot of people who just see the Model A as the poor cousin of the B when in fact it is the thinner, lighter, lower power consumption version of the B, which also happens to be cheaper. We perhaps need to do a better job of articulating that to people so we will try, I think, when we do the Model A+ to push that message more.

RPG: For applications like the Raspberry Pi-based ballooning that Dave Akerman is doing, as well as any other applications that do not want to use too much energy, the Model A seems to be a very useful board?

EU: Before we had the Model A, Dave was actually basically making Model As out of Model Bs in order to make them lighter. In fact before we had the Model B+, Dave was making Model B+ power chains for his flights by de-soldering the LDOs (low dropout regulators) and replacing them with switches so he could get more life out of his batteries. I am hoping that we are steadily converging on the "Dave Akerman Pi" – the Pi which is perfectly optimized for sticking under a balloon! [Laughs]

RPG: I am guessing the price will remain the same for an updated Model A, as it did with the Model B and B+. Are you able to give any more information about what the Model A+ will look like and what improvements it is likely to have over the Model A?

EU: Yes, that is the goal. We really don't want to go up from US$ 25 because it has always kind of been our signature price. One of the important things about the Model A is that we promised everyone the $25 computer  – so we have to make one. We are definitely not going to go up from there.

RPG: In previous interviews, you have mentioned that there may be a possibility for a processor upgrade on the Raspberry Pi in a couple of years' time. From what I understand, there are no more powerful SoC (system on a chip) packages that would be a straight swap for the BCM2835 and the LPDDR2 RAM chip on the Raspberry Pi is already at its maximum available value of 512MB. This would surely, then, call for a fundamental redesign of the Raspberry Pi from the ground up, along with a large amount of new drivers and software to go with it. Is this why it is not something you have considered doing sooner?

EU: That is the real difficulty with moving away from the BCM2835 platform in that we are pretty much back to square one if we do that. We are wedded to the BCM2835 really. We know we will have to [change the processor] at some point, and when we are able to do it for $35 we will, but we are a long way away from being able to do that at the moment even with the economies of scale that we have, so we will need to wait for a new chip to come along that we can use.

RPG: In a world where technology now moves so fast, it seems to be industry practice to increase the capability of hardware before optimizing the software that runs on it. It is also probably quite rare to spend so much development time on one piece of hardware, as you are able to with the BCM2835 and the Raspberry Pi. Have you been surprised by the scale of performance enhancements that have been achievable purely through software changes?

EU: Absolutely. We always encourage people to go back and try running some software from the first year, such as the very first release that we made – the performance difference is pretty staggering. That is community effort and that is money that we have spent. I think this is going to happen more in the future, as we are not going to have Moore's law forever, and I think we are going to see more focus on software optimization over time. But right now we are very unusual in having picked a platform and nailed our colors to the mast and said we are just going to make this platform better by attention to detail.

RPG: You have provided us with so many great surprises already in 2014 that I am finding it hard to imagine how you will top that next year. What have you got in store for the rest of the year and moving into 2015? Any Christmas presents we should be saving up for or new software releases to look forward to?

EU: [Laughs] They wouldn't be surprises if I told you. There will be surprises. There is a lot of stuff that we have talked about in outline that we are intending to do. The display stuff in particular; I think people are going to be really pleased with how that product comes out. We are going to keep trying to surprise people. We have got engineering staff and they are working on cool stuff, but I don't think between now and the end of the year, say, there is going to be anything where you are going to say "wow, they never even gave us a clue they were working on that," but I think it is more going to be about good conclusions to announced engineering activities. Things like the display board and the web browser. People, I think, are going to be very pleased with how those things turn out.

It is also about attention to detail on the software side and doubling down on the education mission. It is actually trying to solve the problem. Ultimately it doesn't matter if we sell 10 million Raspberry Pis if we don't get more kids programming computers, because then we have failed. So it is about keeping going, improving the software, and deploying the resources and revenue that we make from selling Raspberry Pi to really make a difference in education.