Inside Intel: Where next?
By Ed Sperling, Editor in Chief -- 10/23/2007
David “Dadi” Perlmutter, senior vice president and general manager of Intel’s Mobility Group—and the de facto keeper of the Intel Architecture—sat down with Electronic News to talk about WiMax, the company’s on-chip graphics engine, the future of Intel’s chip design and the ultimate number of cores that will be on a processor. What follows are excerpts of that conversation.
Q: The first WiMax product from Intel came out almost three years ago, but there seems to be momentum growing now. Is that just due to ramp-up time, or has something fundamental changed?
Perlmutter: The first products were fixed. It was a good value proposition, but the networks we are going to use now are mobile. It allows mobility, not just a DSL replacement. The scale of what we are doing is much harder, though. It’s not just components that go into a modem. The components have to go into building networks and they have to be put into notebooks and handheld solutions. It’s a big deal.
Q: So it’s no longer just Intel driving this? It’s now the rest of the world?
Perlmutter: That’s correct. It’s a huge ecosystem that has to work together to make it happen.
Q: At what point do you believe you’ll have critical mass?
Perlmutter: I think it’s going to take several years. We’re going to see critical mass when it goes beyond U.S. networks. You will be able to connect when you go from the U.S. to Asia with a wireless connection. Today you see the same thing with WiFi. You go anywhere in the world and you connect. It doesn’t matter which network is there.
Perlmutter: There is a lot of discussion about using different frequencies to allow coverage.
Q: But we’re talking global coverage, right?
Perlmutter: Yes, and it’s going to take several years. It’s still a long road ahead of us.
Q: Doesn’t this open a lot of doors for Intel, too? If you can offer connectivity technology, you can reach far beyond Intel’s traditional PC market.
Perlmutter: Definitely. It will open up a lot of opportunities. I think next year you’ll see that like WiFi, which was limited to notebooks, WiMax will extend beyond that. Mobile Internet devices and handsets are going to be powered by WiMax solutions.
Q: From your vantage point, is Intel now more open to working with partners?
Perlmutter: Intel has been working with the industry for many years. There are a lot of examples of that. With PCI Express and USB, Intel has focused on creating a solution that is not proprietary. It’s true that it’s very much around the Intel technology. But when you talk about building a network, it’s even more than that. You have to partner or it won’t happen.
Q: Intel is talking about integrating a graphics engine into its chip. Is this a discrete engine that Intel has developed?
Perlmutter: Yes.
Q: So this is brand new?
Perlumtter: Yes. This is all new. It was started several years ago. Intel was looking at how to expand its core business. We’ve been expanding into adjacencies. The question was whether we could take something that we were very good at—the Intel Architecture, CPU architecture, process technology—and just go beyond what we traditionally did. You go up and down, and you combine building blocks. The future will tell how successful we are, but this time I have a lot of confidence. We know the industry and the world knows what we have to go.
Q: From the outside looking in, Intel has always had the PC as its core market. It seems something has changed. Is that correct?
Perlmutter: This is probably easier to see from the outside than from inside Intel. It’s a gradual change. The PC is still an extremely good business for us. But we had to look at our business and expand it.
Q: And clearly Intel saw there are 1 billion cell phones being sold.
Perlmutter: Yes. That was the thinking behind Silverthorne. But you probably still won’t run the Internet on a cell phone. It’s a huge, tough software problem. We turned the software problem into a hardware problem. We said, ‘If you could run the same architecture as you run on a PC today, just smaller and way lower power, then the software problem is solved because it’s already done for the PC. That was the challenge we gave the team. We are running ahead of our original plans, and we may go even deeper. There is a possibility of a mid-range or high end for small form factor, so the user will use it more as a computing device than just voice. We have internal numbers of how successful this will be in several years after we introduce this into the marketplace.
Q: Intel produces an enormous number of iterations of every chip, and that number has been increasing since the introduction of the first Pentium. Will that ever go down?
Perlmutter: No, it will continue to go up because it’s a mature market. The trick is using the same technology across iterations or you need to start from scratch. What we are good at is turning a single permutation from quad core to dual core to single core to larger cache. In the future it’s going to be an even more complex permutation, and we will do it with a relatively small number of engineers. One size cannot fit all. How you scale a solution becomes harder and harder, and how you engineer your development to make solutions without increasing your workforce is a very important thing. We are getting better and better at it.
Q: How do you accomplish that?
Perlmutter: We think more and more about modularity—how you build your solutions so you can add and chop things out. It’s very much a planning process where you start out thinking about what you’ll need at the end of the day, not interrupting the course and saying, ‘I need that.’ It’s very hard to do as an afterthought.
Q: At the high end, how many cores will Intel put on a chip—in Larrabee and beyond?
Perlmutter: You have to look at this as something that is general purpose versus something that is not trying to solve a single application, but a more confined set of problems. A Larrabee solution in some mathematical problems that is very redundant is very useful. That could be a database application or Photoshop or graphics, some of the media stuff, things people do in mechanical design or oil and gas exploration. For these kinds of things, a smaller device that is very good at crunching data—it is very effective.
Q: How many cores to you use on that?
Perlmutter: It can use 16 or 64—whatever. Then the issue is that you can make the cores smaller. If you look at the graphics cores from Intel or even Nvidia, they have more than four cores and sometimes more than eight small, dedicated execution cores for dedicated specific functions for graphics.
Q: Given this strategy, Intel’s list of competitors will grow.
Perlmutter: Yes, but you can not expand yourself and just compete with the good old guys you’ve always competed with. And while that’s true, the good old guys also are looking at the same problems.
For the first part of this series on Intel, see "Special Report: Inside Intel," where Pat Gelsinger, senior VP and general manager of Intel’s Digital Enterprise Group, talks about the company’s new target markets, integrated graphics and the next big thing.
© 2009, Reed Business Information, a division of Reed Elsevier Inc. All Rights Reserved.