Navin Shenoy, Intel's head of Client Computing.
Intel is in the midst of its biggest business transition to date. Just a few months ago, the chip giant announced it would be laying off 11,000 workers and taking a step away from the PC market to focus on wearables and IoT devices. Coinciding with those announcements was an executive shuffle that put Navin Shenoy, its mobile-client VP, in charge of the broader client computing group (which covers all consumer devices). At Computex this week, we had a chance to pick Shenoy's brain about Intel's path forward.
What do you envision being the next major breakthrough for PC form factor?
We're working on lots of things that are mind-blowing. To me, we have to figure out how to get to J.A.R.V.I.S. [Iron Man's trusty AI, not Intel's vaporware earpiece]. The ability to manipulate things wherever you are, look at things wherever you are, talk to things in a more natural way -- that's the next big breakthrough in computing. And it will be in so many domains; it won't just be PCs. It'll be phones, tablets and also new types of things we haven't conceived of yet.
I'm a fan and believer in the notion of ambient computing. ... Today you have "destination computing," where you go and sit down with something like an all-in-one. You have "carry with you" computing, like a notebook. And you have "pocket computing" -- your phone. But really, what computing needs to evolve to is ambient computing, where it's just around you and it just sort of disappears. I don't know if it's a form factor or not. I think the form factor becomes less relevant when it's just sort of around you.
How is Intel getting there?
We've spent a lot of time thinking about decompressing the technology problems associated with ambient computing. There are several: there's low-power, high-performance computing that you need almost always, in whatever device you have. Not everything will happen in the cloud. There will be some form of distributed computing, where some happens in the cloud, some will happen locally, it all just depends on the particular workload being done.
That's one aspect. [There's also] far-field microphone arrays, how do you optimize the mic the solution? What are the right algorithms to use? How do you apply those algorithms into either an audio DSP on the platform, or some other technology on the platform itself. How do you combine that with other devices that might sit in the home, like a home gateway? How do you optimize the WiFi connectivity between the gateway and your device? How do you do end-to-end workloads from the cloud and to the end gateway and client device?
There are so many technology problems to be solved. We're going to do what we usually do, and find the ones that nobody else is solving, or that we're uniquely positioned to solve. And we'll partner with the software ecosystem that maybe does other things, like the AI engines and the actual user interface.
I'm wondering how Intel will do that. These are things running on mobile processors... Where along the chain does Intel come in?
Well, they won't necessarily always run on mobile processors. And today even, they don't just run on mobile processors. I think you're going to see far-field in the PC domain, in all in ones, in all sorts of things. Some of them you'll see today, some you'll see in the future.
While today there are some examples that use mobile processors, like any other domain, computing is going to be this sort of spectrum. Some things that are extraordinarily formfactor constrained may have mobile CPUs. And things less constrained will have higher-performance CPUs. And so we'll participate wherever we can.
With mobile processors, Intel has admitted it missed out. What's happening with them now?
We're just really taking a much broader view, and thinking about where the world is going, not necessarily where it's been or where it is now. And if you kind of think about what we think about today, there's just a much broader way to think about connected devices. We're not thinking about a PC only. Or a phone only or a tablet only. We really are thinking about this broad array of connected things ... and figuring out where are the optimal places for us to add value.
So that being said, you will see us participating in phones in some way, shape or form. Our modem technology is going to be used on many networks around the world. Even on 4G, we think we have an opportunity with our modem.
We think as we go to 5G, there's going to be just a completely different way of thinking about the network. From narrowband IoT for things like a parking meter with very low bandwidth and low power to sub-6GHz kinds of things that pump very high data rates over a mobile device to 28GHz with millimeter wave, to 39GHz to 60GHz. We already have Wi-Gi 60GHz in the market today. We just have a much broader spectru, and a much broader way of thinking about how we can participate. Some of that will be medium to long-term. And some will be sooner.
An array of devices powered by Intel's SoFIA chip.
It's sad to hear about SoFIA (Intel's now-defunct smartphone processor). Now you're talking about data centers and AI. But the connection is the key there. How are you going to resolve that missing piece of the puzzle?
The place where we chose to exit is in the low-end of the smartphone market, with SoFIA. Quite honestly we didn't see an opportunity to offer our customers something they thought was differentiated or different. Or on our side, we didn't think we could make sufficient money. So we're optimizing for profitability. But we're continuing to look for opportunities to do something different.
When you think about the future, is the smartphone really done as a device like that, or will it continue to evolve? Of course, it's going to evolve. There will be converged mobility. If you think about what we did with the 2-in-1. ... We said we have the PC, we have the tablet, can we converge usage in some rational fashion? The first 2-in-1s weren't the best, but over two or three years we've seen them get better and better. Could we do the same thing with something even smaller? Perhaps we could.
All I'll say is we're not done experimenting and looking for opportunities to do something different. But we're going to be smart about how we do that.
The modem, we think, is going to be crucial; 4G and 5G modems are going to get increasingly complex, and there's going to be fewer and fewer companies in the world that can do it. You've seen companies like Broadcom, Marvell and NVIDIA get out of the modem. We're to the point now where it's a very scarce asset, and an increasingly valuable one.
Are you also actively working in the AI space? We're hearing more about it from your competitors, like Qualcomm.
I'm not sure about all these buzzwords and what people even mean by AI anymore. ... Right now, to put it in perspective, machine learning is really only 1 percent of the workload today. But it's growing really fast and we're super excited about it. On the training side, and on the actual recognition side, we're doing quite a bit of work to optimize the data-center solutions we have. On the device side, that's where you get into things like computer vision, cognitive computing. And there's a number of things that we're looking at; we haven't talked a lot about this. I'm sort of in the "talk less, do more" space. But suffice to say there are a number of really interesting things we can think about doing with various solutions.
You can imagine taking FPGA [Field Programmable Array chips, which can be reprogrammed on the fly], graphics or CPU core technology and combining them in interesting ways. You can imagine taking things like RealSense and using it to capture the world in a different way, and do some interesting things. Stay tuned, you'll hear more from us.
You mentioned FPGA that's something that could be hugely transformative. Are there any updates in terms of what Intel plans to do with that maybe on the desktop side?
I'm not the expert on FPGA. ... Most of the effort on FPGA with Altera [which Intel acquired last year] will be on the data-center side, and to some extent, IoT. There are some IoT applications that would take a data-center Xeon and FPGA and will combine them to do interesting things. We're exploring in my business on the client side whether there are opportunities, but for now, most of our energy and emphasis in FPGA are on IoT and datacenter.
The idea of FPGA in a desktop processor is really intriguing, especially with something like Viv coming along, the AI platform which basically writes itself. Is that something you're thinking about?
There's new domains that are emerging, like autonomous driving. It will utilize a lot of interesting PC technologies, a lot of interesting data-center tech, it'll have all sorts of machine learning. The datacenter might do the learning part. But the training part, taking all the algorithms and teaching the system what all these things are, creating a big database, and then pushing that information out, [happens] on the car itself. And the car will have to recognize "Is that a cat, or a leaf?" and do that in real-time. So there are places where we are looking at [FPGA]. Whether that makes it into a PC or not, maybe someday it would.
It's one of these things where we want to tap into the maker creativity. We will never be able to conceive, sitting inside of the walls of Intel, what people could do with something like Curie. We have some ideas. We think it's kind of cool to apply within sports. We think it's cool to apply to entertainment and music -- Lady Gaga integrated Curie in her Grammy performance. We're experimenting with it. But the whole reason we did the America's Maker show [a competition in which inventors were tasked with using Curie in their projects] to just tap into the imagination of hundreds of thousands of makers out there.
You mentioned a home gateway during the keynote, could you talk more about that?
What we announced was the home gateway system-on-a-chip (SoC), combined with a MU-MIMO WiFi chip. We're offering both of those in combination in a reference design, which companies like ASUS and Arris can take to market as retail gateways.
So what are those, exactly?
The way to think about a gateway is, on the one hand it brings in whatever access you have to the internet. It might be fiber, it might be DSL or it even might be a wireless 4G backhaul. We have an SoC that combines any access method, so the service providers love that. They can be agnostic, since their box can support anything. And once that signal comes into the home, the WiFi chip becomes the means by which the content can be reattributed into the home. And we can do interesting tricks where the WiFi we have on the client side on the PC can be optimized to work with the WiFi in the gateway.
The gateway itself, as the OEM configures it, can be more than just a router. It can be a place where things are actually stored. It could be like a NAS [network attached storage], it can be a modem and a gateway. There's all sorts of combinations being rethought now. This is going to evolve dramatically over the next three or four years.
The central data center sitting in Seattle or Virginia or something, will not be sufficient to redistribute content to the consumer as the world moves from full HD to 4K to 8K. Or as the world moves from normal flat content to VR. You will have to sort of take the network and redistribute it out closer to the user. And we believe that gateway can be an interesting access point into the home, where various combinations of compute will be reconfigured.
More compute will move into that device, as well as more storage and networking. Things will start to get queued, where you'll do progressive downloads of various things based on user profiles. You'll have various content sort of on demand there locally in your home. All those kinds of things are going to happen over the next several years. Our vision for this thing is pretty expansive and broad, if you think about it.
Is this something Intel has done before?
We were in the cable-modem area of the home gateway. About a year ago we acquired a company called Lantiq, which brought us a number of other technologies like DSL, 4G and fiber. So now we have all the various means by which you will connect to the internet inside of a single SoC. And we basically leverage Atom for the horsepower, along with those other IPs we've acquired from Lantiq.
The 10-core i7 processor is really interesting, will anything trickle down to the 7th generation Core chips, like Turbo Boost Max 3.0?
We'll see. (Laughs) Turbo Boost Max is really interesting technology. Let's say you've got 10 cores running, let's find the most favored core that's running. There's a distribution curve in our manufacturing profile, so one core may run slightly faster than the others. Historically we would just say, we'll run all the cores at the frequency of the weakest link. Now we're able to, on a 10-core processor, say one core is faster than the other nine, and we can dedicate single-threaded workload to that core so you can get faster speeds. It's a really innovative idea. It tends to be more applicable to higher-core-count CPUs.
Some of our commenters don't think it's that expensive, for what you get.
(Laughs) I love those guys and gals. They're just amazing.
At our Intel Extreme Masters events, you go to a sports stadium and there are professional gamers there. And there's 20,000 to 30,000 people there, watching other people play games, they're fanatics. And then online on Twitch, there's a million people watching those people watch other people play games. It's the most mind-blowing thing.
All they want is the best. If we could give them a 20-core processor, they'd buy it. If we could give them something slightly higher, they'd buy it. I love people with insatiable demand for compute.
Richard Lai contributed to this report.
Stay on top of all the latest news from Computex 2016 right here.