Intel Outside

By Ed Sperling and Jessica Davis -- Electronic News, 9/17/2004

Pat Gelsinger, chief technology officer at Intel, sat down with Electronic News to talk about the future of computing, new markets where silicon will become prevalent and the processor giant’s overall direction. What follows are excerpts of that interview.

Electronic News: Intel seems to have shifted gears from the race to more performance in a single processor to multiple cores that can do more things. What’s prompted that change?
Gelsinger: It’s not that we’ve given up the race for more performance. We’re doing it in a different way. That’s what the whole dual core effort is all about. Actually it will be many cores.

Electronic News: How many processing cores are we talking about?
Gelsinger: Many.

Electronic News: Dozens?
Gelsinger: Sure.

Electronic News: So this is almost becoming a system in a package, isn’t it?
Gelsinger: Sure. The number per die per chip is going to be a matter of research and great debate over what’s the right tradeoff versus system and versus chip. But the number gets to be a very big number over time. There will be different configurations, too. Maybe in a mobile I’ll have two, in a desktop I’ll have four and in a server I’ll have eight. As Moore’s Law allows us to go to bigger numbers, that will get bigger over time. But the tradeoff between how many cores are on a die versus how big each core is will have to be researched.

Electronic News: This sounds like a shift in Moore’s Law. Are we talking transistors per processor or transistors per chip?
Gelsinger: This is per chip. Let me elaborate. Moore’s Law speaks explicitly about doubling the number of transistors per chip every 18 to 24 months. That’s what Moore’s Law said. Over the last 30 years of Moore’s Law, we’ve had two other wonderful effects. We were able to increase frequency and decrease power at the same time that we’ve doubled the number of transistors. We’ve been able to pay that off by increasing the clock speed and decreasing the power, so that when you go from 180 [nanometers] to 130 the power goes down and the clock rate goes up and the chip shrinks in size. That era of Moore’s Law where I get all three for free -- transistors, power and frequency -- is over.

Electronic News: Can you continue doubling the number of transistors?
Gelsinger: I can continue to get transistors as far as the eye can see, at least for the next decade plus. However, with technical issues like limitations of single-thread performance and the challenges of voltage scaling, we don’t get the other two for free. We get some improvements, but not the doubling or halving effects that we got for free. That has changed the nature of Moore’s Law. Also, the whole environment for multicore and multiprogramming has matured. We have been working on this in servers and supercomputers. We introduced this into the client with hyperthreading a couple years ago. We can now bring that into the client environment, as well. Now is the time to make this move to multicore.

Electronic News: Is part of what’s driving this a shift from a PC-centric computing model to a distributed computing model, in which a cell phone may be doing some processing that previously was done on a computer?
Gelsinger: The simple answer is, no. What’s driving us into multicore are the factors I talked about. Distributed computing has benefited from this. More distributed computing environments -- peer-to-peer, grid, Web services -- are beneficiaries of this multicore direction.

Electronic News: But isn’t this a little of the chicken and egg scenario?
Gelsinger: Yes. As I have more powerful end points and more powerful computers throughout the network I’m able to do more distributed computing models, and as I do more I’m able to drive more into those multicore processors. They reinforce each other. This transition to multicore isn’t being driven by distributed computing, but it is benefiting by it.

Electronic News: Does that mean Intel will shift into new markets it hasn’t been in before?
Gelsinger: Look at this more as how we deliver value to today’s computing markets. If you think about computing today, there’s the big enterprise, the volume server, the desktops, the mobile, the PDAs and cell phones. You’ll see multicore in laptops and above. All of those will benefit from multicore. There aren’t any significant markets for computing that we’re not already in. This may accelerate them, this may expand them and create new opportunities within them. We already have a strategy to be there, so it’s not like this takes us into new markets.

Electronic News: Still, it’s not necessarily going to be a PC-centric world. Where will computing take place?
Gelsinger: Everywhere. What you think of as the PC becomes harder and harder to distinguish going forward. This applies to all computing markets and we’re already applying it to some markets. We’re using it with Itanium in the server market. I do believe, however, this will establish some exciting new classes of applications that will open up new markets for computing.

Electronic News: Such as?
Gelsinger: The medical market is one example. It’s the least penetrated for computing, which is pretty amazing. It’s 18 percent of the world’s GDP and it’s the least penetrated for computing. Why is that? There’s a lot of reasons, some regulatory, some the structure of that industry. But there’s a lot of problems that can’t be solved with today’s computing. When you look at what multicore can be used for, recognition mining and synthesis, and new classes of applications open up. Imagine if you take your medical records on a USB stick and plug them into your PC, you start Googling against your medical records and testing against databases. My blood sugar has gone up, my blood pressure has done this. You can do data mining against that. The Internet has done that already. People are walking into doctors offices with printouts from the Web. They’re taking more control of their health. But your doctor obviously would never spend the kind of time that you can to do this kind of data mining and synthesis of different technologies, looking at different records. We just haven’t had the level of technology available in the past to do that kind of search.

Electronic News: As a result of this, does your battle shift against the embedded companies?
Gelsinger: Potentially.

Electronic News: We would think it’s more than potentially.
Gelsinger: I see a lot of those as computers. I’m not sure what the difference is.

Electronic News: The difference is that’s the world not owned by Intel.
Gelsinger: A lot of embedded applications take PCs and stick them inside of other pieces of equipment. You don’t know there’s a PC in there, but there’s actually a whole motherboard. Where does it become a PC? It’s hard to tell.

Electronic News: Let’s go back into the health arena. What’s coming down the pike?
Gelsinger: The whole issue of personal health. Your cell phone may record how much exercise you’ve gotten today, if I’ve traveled what kind of nutrition I got, and then when I get home I’ll connect it via Bluetooth so it can automatically get updated. Some of those queries happen without me knowing it, as long as my privacy is protected.

Electronic News: So everyone gets networked in?
Gelsinger: On a more grandiose level, we’ve said we want to have everyone on the planet connected. Right now we have 800 million. We have almost 5 billion yet to go, and we want them connected 7 x 24 with both interactive and proactive devices.

Electronic News: Describe proactive devices.
Gelsinger: I walk into a room, these devices sense me and turn on the lights and turn on the heat. I become energy conscious efficient, I become healthcare conscious efficient, I become security conscious efficient. I come home at night and I get a notification on my PC that says there was somebody different at the house today because I have video recognition at my front door. This person didn’t fit any of the people that normally come by the house. Maybe it’s fine, maybe it’s not, but the computational requirements to look through a screen and identify a person -- that’s a really hard computational task.

Electronic News: How far behind is the software? It sounds like the hardware is way ahead.
Gelsinger: Hardware will always lead. You can’t create software for which there will be no hardware available for five years.

Electronic News: But we’re really shifting computing out to the network, which is a whole new model. As you do that, is there a longer lag time for the software to catch up?
Gelsinger: There might be. But we’ve seen this in the history of computing. Eventually a new level of abstraction emerges that simplifies the programmer’s job, and so they go from maybe being five years behind to one year behind. That’s the nature of the symbiotic tension.

Electronic News: It sounds like one of the next big hurdles will be integrating wireless and wired across a number of different platforms.
Gelsinger: That’s one challenge, and it’s what Web services is trying to address. This is a case where the software guys are almost inching ahead of hardware. They have an abstraction that works well, XML, but the hardware is barely capable of processing that data. When you move from compact data types to complex string data types for everything, those string data types are recursive. This is a big jump in computational requirements. You need a lot more performance.

Electronic News: Intel has dabbled in many different markets over the years and pulled out. Why?
Gelsinger: Most of the markets we got into and chose to withdraw from weren’t reliant on the fundamental core competency of the company, which is silicon. If you look at our balance sheet, we have $5 billion for R&D. The largest chunk goes to silicon. We have capital assets of $30 billion, most of which goes to silicon. Anything that doesn’t leverage those two assets isn’t the best investment of shareholders. We’ve done services, video cameras and consumer things, and they didn’t leverage those assets. Things that do leverage those assets we’ve stuck with, like flash. In communications, we’re focused on the silicon level.

Electronic News: Will the future of chips remain in silicon?
Gelsinger: Other materials will migrate to silicon, and I expect that to continue for at least a decade and probably two or three. There isn’t another material domain that shows anywhere near the manufacturability, scalability and cost structure of silicon. We’re exploring how we take nanotechniques and add them to silicon and extend silicon. People suggest biological computing, molecular computing, optical computing made of different materials. Most of those are nonsense. They are so far from anything that is manufacturable that it’s ridiculous for now. In two decades, they might be interesting. In the meantime, it’s how we can harness some of that research and apply it to silicon. I call it the silicon scaffolding.

Electronic News: Would Intel ever get into MEMS production?
Gelsinger: Absolutely. We’re just adding stuff to silicon. I have a fab in Israel that’s building those right now.

Electronic News: Will Intel stay with it’s own fabs, or will it start parceling out some work to third-party fabs.
Gelsinger: The large preponderance of Intel is as a vertically integrated supplier. But as we look at new markets, our big fabs are mostly built around chipsets and processors. They’re not well tuned to communications chips yet, so we’re building a lot of our communications chips in third-party fabs. Some of that is, we did and acquisition and they were using TSMC so we’ll continue using TSMC. Some of that is, our process recipe isn’t as good someone else’s. We’ll continue to do that. In some cases, it allows us to hedge our capacity.

Electronic News: Is there any percentage you can ascribe to that?
Gelsinger: We’ve never given out that number, but well over 50 percent is from our own fabs and it will continue to be that way forever.