 We're here at the Lenaro Connect, and who are you? John Masters, Chief Arm Architect at Red Hat. And you've been working on ARM Linux for the last 22 years, is what you said? What were you saying before? Yeah, I sort of been working on Linux for about 22 years. So I started using Linux when I was 13 and I'm in my mid-30s now. So it's a very long time to be involved. But it's actually, I'm a relative newcomer compared to some of the people in this space. So your business card says ARM enthusiast? No, what does it say? I'm architect. ARM architect. But you're the ARM, what do you call it? Not cheerleader? Cheerleader? Fanboy? Yeah. What should we say? But I mean, it's been a while now. Is it happening fast enough? Now it's happening. Now there's some huge chips. You're going to be very, very powerful. You're going to take over the whole server market, right? Well, that's one of the aspirations that the ARM ecosystem has, right? So I'm often seen as a fanboy of ARM servers. And the reason is that I think there's a disruption happening in the server market now, where you're seeing the fact that Moore's Law has basically come to an end. Some people don't believe that. But I think it pretty much has come to an end. And single-threaded performance is not what is defining the industry anymore, because it's not increasing at the same rate it used to. And so actually what's defining the future of the industry is machine learning, accelerators, lots of additional workload optimization that's happening outside of the core. But the core is still important. You've got to have a good enough compute story. And the opportunity for ARM is that they now have a chance as the ARM server cores are rapidly catching up and meeting the same kind of performance of the other architectures out there, right? It means that ARM has an opportunity to get into the mainstream server space. And I think that's going to start to happen over the next year to 18 months. You're going to start to see some big name brand OEM and cloud vendors starting to get very interested in the space. But is ARM also solving the single-threaded need? Well, I think. Because maybe the Thunder X2, maybe the new Qualcomm chip set are quite powerful, you probably have the secret benchmarks, right? You know how good they are. Well, I think the fun thing is, right? So if you want just the fastest single-thread performance, actually the fastest single-thread performance in the industry isn't your regular X86 servers or your ARM servers, the fastest single-thread, the highest clock speeds possible are actually some of the more big iron traditional IBM machines and so on, right? I don't just mean power. I mean like mainframe machines, right? Those clock very, very high and so on. Mainstream compute is a trade-off. You want to have a certain amount of single-thread performance, but you actually don't want to take all the downside of running your clocks that high and having your thermals that high and having to have the kind of cost of manufacturing that comes with having that kind of quality of silicon that you need. So actually what you want is you want that sweet spot. You want good enough performance. And so hitting the kind of performance that you get from say an X86 class part today, that's where the ARM vendors are getting to now with their second to third generation designs. And so you mentioned a couple of vendors there. Very exciting ones. Yes, yes. Well, I've been obviously involved with all of these guys for many years and I'm very, very bullish on those two vendors in particular, also the others, but what Qualcomm and Cavium are doing is very exciting. These are mainstream server-class parts. They're not wimpy cell phone cores. These are ground up server designs intended to be put into a server space. Big chips. Big chips. And the thing with big chips is, why do you have a big package? You have a big package because you've got memories on there. So everyone who's building a chip now has thousands of pins. And that's because you're running all these IOs off there. You've got lots of cores. You've got lots of memory. Some of them are 10 nanometer and they're still huge chip. That's right. It's going to be really interesting how big it's going to be. That's right. The actual dies can't speak to how big the dies are, but certainly the packages are huge because you've got so many pins coming off it because you're running so many IOs off there. You've got so much memory. You've got so much PCI. You've got multi-socket. All the standard server features. And some of these companies, I'll mention, Qualcomm, is actually quite happy to be very aggressive against Intel. Right. And Holly, the other guys too. Yeah, I think everybody wants to really be a significant huge player and the arm is going to reach to 25% by 2020, right? So I'm a big believer that arm can take a significant chunk of the server market. I don't know if it'll get 25% by 2020. Maybe we'll see how it plays out. It's going to depend on how you measure what a server is because a lot of the traditional enterprise data center guys are not going to replace their existing x86 machines with arm servers overnight. And in fact, you probably don't even want them to. There's not really as much value in just replacing something you already have. The interesting opportunity is adding something that's net new. So we're doing a lot of work in the NFV and telco space with people that are blazing trails using arm servers for telco workloads and running basically LTE phone calls and the like. That's a very interesting opportunity. There's a lot of opportunity for cloud. There's a lot of opportunity for enterprise, but it's not just sort of subtractive or not just replacing. It's about adding something new. And yeah, the Qualcomm Centric really is a very interesting example. I think Centric and then also ThunderX too, they are good examples of the kind of server class arm parts that I've been waiting for for the past five or six years that I've been working on this. And I'm guessing if things we ramp up, it's only some kind of company that's a big G, a big F, or big something. If they make a big order and suddenly things happen quickly, then you have a phone number. You can just call up a whole bunch of Red Hat guys to come and ramp up a whole bunch of solutions very quickly, right? That's right. And operators are standing by, ready to take your call and your contribution towards subscriptions right now. No, I mean, I think exactly, though. I think there's a big team, right? There's a big team. And they are all ready to make the swap over to ARM. Well, the philosophy has been that actually Red Hat has a strategy called multi-architecture, which basically means things that are not traditional like said, E6 servers. So it's not just ARM. There's other architectures as well that you obviously see in the industry right now happening as well. Well, not RISC-5, because, well, I speak to the RISC-5 thing. So RISC-5 is a very interesting architecture. But RISC-5 is not something people are putting into servers. And the reason, they might put it into servers on an adapter card. They might do it in IoT. It's very interesting for Red Hat to follow these areas. But the thing with RISC-5 is if you want to put it in a high-class server part, high-performance server part, it's going to cost $1 billion. And it'd be very interesting if someone did it at RISC-5 startup with a server processor that'd be interesting. But the cost of building that would be at least $1 billion. And I'm not seeing people with that kind of investment dollars. So from a Red Hat perspective, we're looking at this multi-architecture future, which is really around having more than just X86 as an offering. And that's because that's what customers want. But what customers want is they want the same, we call it Red Hat Enterprise Linux or REL. They want the same REL experience, regardless of which architecture that they choose to run on. So actually, what Red Hat's done for the past few years is build its portfolio of enterprise Linux offerings for all these different architectures. And so they're already there. They're available to customers today. There's actually a beta release right now of REL 7.4 for ARM that people are actively using and testing on machines. And so if there is a big customer demand, we're obviously ready to go and respond. Ready, but can you even ramp up, get things done? Even like a bit public with new ideas, maybe. Right, well, people ask me. It's kind of like apps people need, also. Yeah, that's true. People asked me at this conference. They said, at Lenara Connect, we're here in Berlin Game, they asked me how many people at Red Hat work on ARM? When we started six years ago, it was myself and a couple of colleagues that started the team. And now we're at a point where there are about 10,000 people working on ARM. And the reason for that is everyone in the company has some involvement with ARM. Everyone in the company has some involvement with X86. Everyone has an involvement with every architecture because we make it just another architecture. And so a lot of people are using ARM where they're doing software development and their binaries are getting built for ARM. And they're not even aware that that's happening because it just happens. Our build systems, when you build some software that most of our developers are working on, it actually just gets compiled for everything at the same time. And we treat it as just another architecture. And once ARM has taken over the whole market by 2022, then you're in the right spot to be the next CEO, right? Well, I'd love it. I've said to my CEO many times, he's actually one of the main reasons I love working at Red Hat. Jim's an absolutely amazing guy. But if he wants me to be CEO one day, then great. I'm ready and willing to take on that challenge. Hey, and if he does, he'll come back and there'll be ARMs everywhere. It'll be amazing. So I think there's a big opportunity to be proven right, if you like, that I think ARM has a. But by being proven right, you're not proving someone else wrong. There'll still be a lot of other architectures. There'll still be a lot of different things. Intel are going to make some amazing ARM server chips very soon, right? You know, the interesting thing with some of these other players, right, is I'm a big fan of competition. So the moment that you have really high-end, fancy, high-performance ARM cores that are challenging some of the existing vendors, what it's going to do is it's going to make them innovate, right? So it's good for customers, good for the industry, good for everybody, because the Intel guys will respond. That's what they're good at doing. They've responded a lot to everything that ARM has been doing the last six years. And you know what? There's a whole bunch of new Intel server chips that are inspired a lot, but what you did. And Intel in a row did. And yeah, I think that's true. And I think also, if you look at, say, watching just AMD and Intel together right now with Epic, right, the response to the surprising success for some, they were not necessarily expecting Xen and Ryzen and Epic to be as successful as it has been. But you've seen the reaction. You've seen Intel in the client space coming out with the Chlorine 9 series. You've seen the response there on the server side. And they also came with micro server, small server systems. Exactly. As a response to the whole. To the micro server. Atom was in the response to ARM, right? Yeah, and Xeon D and so on. So I think what happens in the industry is when you have these competitive offerings, it just drives everyone to innovate faster. And I think, you know, I don't know what Intel's working on. So I'm just speculating here as much as you are. But you know what, they're the leading manufacturer of silicon technology still, right? Others are catching up. They say this. I mean, others are catching up. You look at the roadmap for there's something called the International Semiconductor sort of roadmap, right, that talks about the different node sizes over time. But the ARM was on 10 nanometer before Intel. Well, it depends on how you measure the nanometer size, right? So Intel measured their minimum feature size slightly differently from others. They're very conservative in that number. So actually Intel 10 nanometers, pretty true 10 nanometer. Other people might have maybe a 12 LP or 14 nanometer that's kind of the same. But over time, they're all kind of catching up with each other. But the interesting thing is Intel's still the lead manufacturer. So to your point, if they ever did want to fab, you know, ARM server cores, right? I'd love to see that. Maybe that's what they're doing for LG right now. Not just smartphone chips. Who knows? LG servers. Who knows? I mean, it's something that accompanies that size, right? It has the opportunity. If they see what's going on here and if they decide, you know what? This is a fabulous thing. We love our existing data center stuff. We love X86. It's great. We'd love to have a 10 or 20 Intel ICINEs join the NARO. Sure, by next connect. Sure, why not? Why not make it 100? 100, yeah. You know? We should consider this place. It's a nice place to be, the NARO. Well, I think the interesting thing with ARM is if you look at what we've gone over the last six years, right? So you see Moore's lore is kind of grounding to a halt. You see every year, your single threaded performance is not getting 50% better. It's not getting 30% better. It's actually getting 3% to 5% better if you're lucky. What's happening is people say that their servers are getting faster, and that's because you're giving them more cores. Well, if you give them more cores, they run a bit slower. They've got to clock it down a bit. So actually over time, what's really happening is we're getting more cores and we're getting more parallel workloads. But we're not getting more single-thread performance. And what's happening is rapidly everyone's catching up to the point where everybody's got similar numbers of cores, similar performance, same kind of stuff. So at that point, where's the value? And I don't think the value in the longer term is in pure compute. I think the value is in offload. That's why I'm also involved in C6, the cache coherent interconnect for accelerators work group. And what we're doing there is we are working on building a cache coherent interconnect, similar to, say, IBM's Cappy or OpenCappy, or NVLink, or many different technologies that are out there for plugging accelerators very close to the chip. And I think those kind of technologies are going to be very interesting in future server designs. And I think that if you're an Intel or an IBM or one of these long-standing incumbent players, you know, look at the space and you're going to say, well, compute's kind of boring. So do I really see it in Intel or in AMD kind of giving up X86 and moving off it over time? Maybe? Who knows? But maybe the compute piece is kind of boring, you know? What we're doing with ARM is not trying to just replace the compute. What we're trying to do with ARM is we're trying to democratize the architecture to the point where you have 10 different vendors, 20 different vendors, more who will have high performance compute, but then they can all give you something innovative on top. So driving innovation, driving innovation is better for the world. And it's the only way we're going to have billions of people with smartphones that have all kinds of smart features that go part through the cloud. Otherwise, you know, things are going to burn. I think open innovation, open standards, these are the important things. There's a great book that I encourage people to read. Rod Canyon, who was the CEO of Compact back in the day, and he went to HPE and then HPE and so on. He's doing other things in the industry now, but he wrote a book called Open. And his book's all about the disruptive power of getting industry players together and building open standards. And he famously did this on the PC. He helped to create what is now known as the industry standard architecture for x86 PC servers and laptops and workstations and all these other devices. But that wasn't the standard when he began, when they began their work back at Compact in the early days, everyone had a different PC, they were incompatible, they ran different versions of DOS and Windows and so on. And he got together with different vendors and he wrote a book about it. It's called Open, it's a great read and it goes through the disruptive power of coming together. And you look at Lunaro, you look at the ARM ecosystem, it's the same kind of strategy. It's getting all these vendors together and saying, we are stronger when we work together than we are if we all try to just, shoot at each other, right? So we used to joke, it's kind of like the pirates coming to battle with the navies of the world, right? What you don't want to do if you're a bunch of sort of disruptive pirates coming in, you don't want to shoot each other. What you want to do is you want to work together. And so that's what we're trying to do here is we're trying to all coordinate or collaborate together, build open standards, build interesting alternatives, but let people innovate on top of that technology.