Intel announced last week that it hired John Sell, the chief architect of next-generation console chips at Microsoft. Sell worked on the chips for the Xbox One, Xbox One X, and the upcoming Project Scarlett.
I’ve interviewed Sell over the years on the different projects he has worked on throughout his career, ranging from Microsoft’s game consoles to its Kinect motion-sensing camera.
Sell is joining Intel as a security architect, and he will report to recently hired Raja Koduri, Intel’s chief architect and former head of graphics at Advanced Micro Devices.
As part of a new “architecture era,” Intel has been hiring high-profile semiconductor architects. Besides Sell and Koduri, the chip giant also hired Jim Keller, who worked on a variety of chips at companies such as Apple and AMD. Sell will work on security technology.
Sell was at Microsoft for 14 years, working on system-on-chip designs for the company’s gaming hardware. He also worked at 3DO, Advanced Micro Devices, and Apple. In 1980, he founded Ridge Computers, one of the original RISC (reduced instruction set computer) startups.
I talked to Sell about his history in chip design, as well as the trends in semiconductor design, the changes in game consoles over the years, and his work volunteering at the Hot Chips chip design symposium over the past 17 years.
Here’s an edited transcript of our interview.
VentureBeat: You scooped your own news last week. I guess you never know who’s watching your LinkedIn.
John Sell: [laughs] Well, other than updating that after being gone for three weeks, I didn’t have anything to do with it.
VentureBeat: I wanted to tap your brain on semiconductor design in general. Why did you decide to make the move to Intel at this time?
Sell: Well, Intel presented me with a really exciting opportunity. I was not, honestly, out looking. I’m very excited, very bullish on Microsoft and the things I was doing there. It is true that my work on the coming Xbox was mostly done. To be leading something as important as security at a company as important as Intel is a pretty exciting opportunity.
VentureBeat: What part of the organization do you report in to?
Sell: I report to Raja Koduri. As I’m sure you know, among his titles he’s the chief architect for Intel. He also leads an organization known as Intel Architecture, Graphics, and Software. That’s his org of several thousand people. He’s organized that into the six pillars that Intel’s been talking about since Intel Architecture Day last fall. I’m leading the security pillar, so I’m the chief security architect underneath Raja as the chief overall architect.
VentureBeat: Was that something you already had lots of experience within chip design over the years?
Sell: I was at Microsoft for about 14 and a half years. I started getting very — well, the Xbox has always strived for good security. I started getting very involved in that aspect about 10 years [ago]. Security, needless to say, has had a rather rapid increase in focus over those years.
In my own career, I’ve had a lot of experience in different areas. I started out as primarily a CPU architect. I became a bit of a graphics architect, a bit of a power management architect, a memory systems architect, a fabric architect, a lot of things. They were important issues for the projects I was working on. With Xbox, security became a really important issue. It’s certainly a very important issue now, as we keep getting freshly reminded.
VentureBeat: Have the machines gotten demonstrably much better at security over time? Or is that still the hardest thing to perfect, the notion that someone is always going to be able to crack security if they spend enough time on it?
Sell: Both hardware and software are getting much more security-conscious, exponentially so. The most important part of my role at Intel is set up a longer-range security strategy, but also, of course, to do what I can along with the team here with short-term issues as well. I would also say that no good security architect is ever going to be complacent and say, “I’ve solved it.” You can implement a new floating point format, test it completely, and be pretty sure there are no bugs, but it would be foolish to ever say that there’s no possible security problem.
One of the things that I’ll be doing here is striving to have a security design culture here, if you will, so that we anticipate the kinds of problems that can occur and minimize the chance of having them. But the issue with security is it’s kind of an anti-testing problem. You can test something to see that it does function how it’s supposed to, that two plus two equals four. It’s a more difficult problem to be sure that all the rest of the universe doesn’t have something wrong that can affect the security of your hardware or software. But a lot can be done in just structuring and designing security-critical components of our CPUs and other devices so that they’re inherently resistant. We could spend a whole afternoon listing all the kinds of attacks that can occur, but there’s a lot that can be done there.
VentureBeat: I was trying to remember how far back in history you went with the 3DO CPU team that went on to become the WebTV folks and then the Xbox and Xbox 360 team.
Sell: If you think way back, I cofounded Ridge Computers, which was the first commercial RISC machine. After eight years of working on that, the founders and the venture capitalists had a parting of ways in terms of the best strategy for Ridge going forward. Arthur Rock, who had a hand in getting Intel started, steered me toward Apple, where I led the team there creating the Power architecture with IBM, the PowerPC Macintosh architecture.
At Apple, I was recruited away by Trip Hawkins to 3DO. I went there as the CTO. You’re correct that when 3DO got out of the hardware business, most of the hardware team ended up at WebTV. Microsoft bought WebTV. I went to Advanced Micro Devices, AMD. The team, by then, was at Microsoft creating Xbox, and they kept trying to hire me away. After eight years they succeeded. That began my Microsoft career. I came to Microsoft just before the first silicon arrived for the first Xbox 360. My primary first job, the task there — that was a 90nm design, so I was making the 65nm version of the Xbox 360.
VentureBeat: I remember Ken Kutaragi gave a talk a while back at one of the big chip conferences. He talked about how much they shrank the size of the Emotion Engine CPU for the PlayStation 2. It was something like 13% of the original die size by the end of the product’s life, six or seven years. That was the model for game consoles, where you put in as powerful a chip as you could for that $400 launch price, and then shrank it as much as you could every year or so to drive the cost down. That was how you made use of Moore’s Law. And the PC was almost the opposite, where you’re always trying to double performance instead and stay at a constant cost. Is that the way the world still works, do you think? As far as how game consoles and the PC have progressed.
Sell: I can comment on the past in the game consoles, at least. I’m not that directly involved here at Intel. But what you described as far as the consoles is definitely the key over the last 15 years or so. It seems to be changing a bit going forward. The rate at which new consoles are coming is picking up. Developers may not love that, but — because the performance of consoles has gotten so high now at the high end, with things like Scorpio and the coming Scarlett, they’re starting to be a family of consoles. Both Sony and Microsoft have a $250 version and a $500 version. Those factors are changing things.
The other thing that will be an exciting growth opportunity for Xbox, certainly, and I think also for others, is cloud games. The technological pieces of that are coming together to where it’s practical. There hasn’t been a combination of performance and cost-effectiveness in the past.
VentureBeat: As far as you’re concerned, do you believe the time for cloud gaming, game streaming, has arrived, or do we have to wait another 10 years for it to work in a satisfactory way for consumers?
Sell: I think it’s on the cusp of arriving. It’s about to arrive.
VentureBeat: That’s where you’re then, in some ways — I’m not sure what the implication for the developer. The player no longer has to worry so much about buying a new machine every few years. The developer is developing for the datacenter and what’s there.
Sell: There are some unknowns. You’d be best off, I think, asking some of my former colleagues at this point. [laughs]
VentureBeat: But you do think this is a new vector that could be interesting for the design of these machines?
Sell: Right, I do. However, as I said before — I’m bullish on what Microsoft is planning in that area. I can’t talk about details there, and the details have changed beyond what I knew. But the opportunity that I have here at Intel is really exciting too. It’s a bit different, but it also includes all of those areas. One of Intel’s most important businesses is the datacenter. We’re working on new initiatives in several specialized areas, GPUs and lots of other things as well. For what I’m working on personally, it’s a broader opportunity.
VentureBeat: It’s a wide range, right? You’re going from IOT on up to server chips.
Sell: That’s right. A lot of security issues are common. The scale, the bandwidth, the details vary, but there are a lot of common — Intel’s casting itself, quite rightly, as being more of a data-centric company rather than just a CPU company. Everyone wants their data to be secure, whether it’s at the IOT or edge or in a server room and getting it securely back and forth.
VentureBeat: What do you think about this bouncing back and forth you’ve done between the horizontal chip companies and the vertically integrated companies that do everything, like Microsoft?
Sell: From a technical point of view, we still cover — in some ways we cover just as broad a range. It starts lower, with the devices, in a way that a few of us — including me — paid attention to at Microsoft. Obviously, it’s not the center of what Microsoft does. It goes up through — there’s an awful lot of software people at Intel. You can’t do anything without lots of firmware and software. As Microsoft takes care of its underpinnings of Windows, Intel is very active working with Microsoft, as well as with Linux in all its different flavors.
It feels like the shade is different, but it’s not a whole different world. Of course, I was part of the hardware part of Microsoft.
VentureBeat: Did you feel like there was a core group of folks you had always worked with over there? I wonder if that was the reason you stayed for so long at Microsoft.
Sell: I did a lot of great colleagues there, who are still there. I’ve got a few that I’ve worked with over the years, and Intel as well, including my boss, Raja, but even more so people like Jim Keller. We were fellow senior fellows at AMD at the same time, back in the late ’90s. Intel is a big place. There are a lot of others. I’ve been here months now, spending most of my time with the security architects of course, and there are a lot of good ones.
VentureBeat: You were heavily involved in Hot Chips over the years. What was your motivation there?
Sell: Hot Chips is a unique conference, as I know you know. It’s run by volunteers. It’s not a for-profit conference like some of them. It doesn’t require putting together a paper that usually gets written a year after the chip came out and all that. It’s a presentation format. I’m actually on all three of the committees — the steering, the organizing, and the program committees. We’re fairly maniacal about not allowing too much marketing in the presentations.
A couple of years ago, John Hennessy joined the steering committee when he retired from Stanford. He’s also been on the program committee two of the three years when he had time. When we had our first steering committee meeting after he joined, we kibitzed back and forth about MIPS versus Ridge, his past and mine. But that’s lent a little bit of an uplift of prestige and importance to the conference. Intel’s going to continue to be one of the corporate sponsors.
VentureBeat: What did you learn about how the job of the chip designer has changed or evolved, just by being involved in that conference? Are there just as many jobs for these kinds of people in Silicon Valley now? Have you seen that dissipate all over the world?
Sell: I think there are just as many jobs. Maybe the rate of growth of those that work at the — if you look at the percentage growth, it’s probably smaller than in some higher-level software. But the absolute number continues to grow.
The job has changed. Even with my poor eyesight, I could see the transistors when I started. [laughs] Another publication pointed out that the original Xbox One had more transistors than the entire production run of the first microprocessor. It just shows you how technology has moved forward.
The job of the architect has gotten — mostly architects don’t get to touch the hardware much anymore. There are much more sophisticated simulation and design tools, so we can do more complex, more advanced architectures and get them to come out right, without having some horrible performance bottleneck. We couldn’t do that without those kinds of tools. Doing everything by guesswork doesn’t lead to the optimal result. It never did, but it’s even less so as complexity has increased.
It’s harder to distinguish between a software and a hardware engineer these days. A hardware engineer still needs to understand the physics and the limitations and the special conditions of building the devices.
VentureBeat: I wonder if there’s a good metaphor for describing this type of job now. Some people used to say it’s trying to keep track of the traffic in New York City. The amount of detail you have to make sure is correct is as complicated as something like that.
Sell: It does amaze me that we can fully utilize this much silicon when we have double-digit billions of transistors. When you stop and think about it, it does seem amazing. That’s partly because of the tools. I don’t remember so much from 20 or 30 years ago, but the verification of the designs probably always has been, and certainly remains — it’s a bigger effort. There are more people working on that than there are architects.
I was commenting about things being designed so they don’t create unnecessary security exposures. They’re inherently more secure. The same thing applies to designing to deal with complexity. There has to be a clear benefit, rather than complexity for complexity’s sake.
VentureBeat: I remember the CISC versus RISC debates were always about what level of complexity you wanted.
Sell: You know I’m a RISC guy. I suppose I have to keep that to myself at this point.
VentureBeat: Not sure who won that? [laughs]
Sell: When you get down to the microarchitecture and issues like how we approach design — certainly there are areas where simplicity is what you want and what you can do because it’s under the covers. But keeping things simple in the most critical areas in security design is really valuable. It’s a lot easier to prove and correct.
VentureBeat: Given the kind of work you’re doing, do you think this is the last kind of work that might be done by AI? Do you see some part of the job being done by AI to your benefit?
Sell: We’re looking at areas where AI can help. I think there’s going to be some areas, you might say, at the base, at the root of trust — it’s going to be more complicated than that. But there will be some core areas where AI is not really the answer. There are absolutely some higher-level areas where we’re looking at it.
VentureBeat: What about the notion of AI replacing a chip architect?
Sell: [laughs] You can call me biased, but no, I don’t see that happening. It’ll be like we were discussing a few minutes ago, where you don’t draw the individual transistors. Humans are going to work at a level above whatever we manage to create, in the sense of tools. AI will be a tool. I don’t subscribe to the notion that machines will replace us. There’s some that do, but I’m not one of them.
VentureBeat: The level of abstraction and how tools can make this understandable or effective for someone like an architect to work — what do you notice about where these tools are and where they’re going, how much they’ve changed over the years?
Sell: I’m not an expert in this area, talking about specific toolchains, but I think the tools are obviously doing a lot of the drudgery of the work. I may be overstating this, but it’s also — we don’t have to count transistors one by one and meter them out. We, the architects, can spend our time on larger concepts rather than trying to micro-optimize the use of every transistor when you have only a few thousand or a few million. Tools, among other things, do optimization.
But when you have a timing closure problem or something like that today, it still takes a human. To do the things the tools haven’t been able to solve, it takes a human to go in and fix it, rather than having the chip not being able to have nearly as high a frequency, or take more power to run at a given frequency. The very hardest problems, the tools don’t solve them. They solve more of the day-to-day ones, so that lets us do more complex things, and us humans can worry about the hardest problems.
VentureBeat: How do you feel about big teams and small teams, the chip designers at startups versus the ones at the big companies?
Sell: I’ve worked at both. Big companies that are doing a good job do provide a more complete product. All the Is and Ts are dotted. They’re everything from more thoroughly tested to just having all the infrastructure around the product, to have broader use. Small startups end up aiming at a more specific niche.
Of course, there are startups that have literally billions of dollars raised or at least hundreds of millions. Maybe in some ways, there’s not so much difference there. Ridge was started in the proverbial shed, or in its case more specifically a garage. The end result — if they grow into big successful companies, that’s what they are. But the initial market that they can address is just smaller.
VentureBeat: Were you ever a big gamer yourself?
Sell: To be honest, I really only got into playing games when they were creating a problem for the hardware. I’m just not very good at it. [laughs] I’ll tell you this, too, because I’ve said this publicly before. The biggest reason is that I’m not very good at it. I’m good at designing the systems for them, but I’m not very good at playing them.
VentureBeat: I wondered how much of the final product that a chip designer needs to understand in order to start at the right place for designing a chip, designing a system, and the hardware that’s ultimately going to be used by a lot of computers or in the datacenter. How much of that side of things do you wind up having to see and understand?
Sell: Well, the more the better. This includes things like security architecture. You can’t have the security be at odds with the way the end user wants to use the product. You want it to be secure, of course, but you don’t want to give up other things. It’s the more, the better. As an architect, you want to know how the software and the applications are going to use the hardware, rather than designing its features in a vacuum.
There’s a whole continuum of “customers.” If you’re designing a CPU, your customers are the people designing the kernel operating system, but also all the apps. They’re all using the features. Then there’s the end user, particularly if it’s a device, or if it’s in the server room, everyone cares about cost, heat, form factor, reliability. Again, the more the better. You can’t know everything, but you need to know as much as you can.
I spend a fair amount of my time listening to the people here at Intel who are working directly with our customers, and I try to go talk to some of the big customers personally, to hear what their issues are directly.