The speed of the microprocessor and Intel’s Pat Gelsinger’s career growth
have been running on parallel tracks for the last 20 years. At 32, Pat became
the youngest vice-president Intel ever had. A bit later, he took over as the
chip giant’s first CTO. Overseeing the entire technology research and
development activity at Intel, Pat today works as the top visionary of the
digital world. Excerpts from an interview with Dataquest:
You seem to be excited about the possibility of extending Moore’s Law
into other areas. How soon do you think Moore’s Law can be adopted?
In the communications area, particularly the wireless communications
segment, there is a lot of enthusiasm in applying Moore’s Law to build current
products in a better way. However, the concept has not picked up in the
application of those capabilities to new markets. We are working with some
companies to start applying things like sensor networks to new application areas
assuming that radios become free, ubiquitous and fully integrated. There are a
few challenges though. For instance, having built the antenna, how do you put it
into the laptop and other devices so that it does not interfere with the rest of
the circuitry? We also need to identify new market applications for these
capabilities. Industry participation is critical in expanding Moore’s Law.
Intel does not build products but building blocks. We are looking for companies
with systems, software applications and platforms that will complement our
innovations.
Your confidence in Silicon seems to be near total as you talk about it
driving the chip speed into tens of GHz in the next 20 years. Is there any other
compound that Intel is trying out for high-speed processors that uses terahertz
transistors?
My confidence in Silicon stems from the fact that it can be manufactured
easily. You can combine other materials into Silicon very effectively. Compounds
such as Gallium-Arsenide (GaAs), Silicon-Germanium (SiGe) and a few others are
very expensive. Besides, they are not scalable and cannot be manufactured. As a
result, GaAs has achieved the reputation of being ‘tomorrow’s technology for
ever!’ We have been talking about this compound for the last 15 years. It will
not go away. It is needed for some extremely high performance applications. The
other aspect is that because of the materials used inside Silicon, I now believe
that biological computers will never make any difference. However, I do think
that we might use biological materials and their characteristics in Silicon. The
concept of bio-enhanced Silicon may catch up in the future but Silicon will
remain the fundamental substance.
What are the limitations of Silicon?
Lithography and the power issue are the main limitations of Moore’s Law.
Power does not go away. We also need to engineer new technology. In fact, we
need to attack that aggressively.
How interested is Intel in the Carbon nanotube technology?
We are quite interested in this one. The approach could be like that of
bio-enhanced Silicon, which would be very useful in the future. However, it will
be decades before mass production of products based on these technologies takes
place.
The next spurt in innovation in software applications will come through
Web services. With the DotNet environment, the Web services solutions, UDDI?
Universal Description, Discovery, and Integration- an XML-based registry for
businesses worldwide to list themselves on the Internet.) XML (Extensible Markup
Language) and other concepts reaching the first generation of maturity, we
expect high activity in these models.
On the PC front, Intel is working with its OEMs to consistently reduce the
form factor. On the other hand, notebooks are getting more and more powerful and
loaded with as many features as the PC. Moreover, the Tablet PC is coming this
year. Where’s the PC headed in light of all this?
The distinction between PCs and the mobile desktops is increasingly getting
blurred. In Japan for instance, big notebooks are used as desktops. They are
transported from one place to another probably once or twice during their
lifetime. They are mobile and yet not mobile. The desktops are shrinking in
their form factors. We are comfortable with that. We expect the line of
distinction to blur further in the future. However, there are some innovations
that will differentiate the desktops from the mobiles. Take for instance,
wireless applications. My vision of the mobile version of the wireless device is
what I call 111A.
This means the device should weigh a pound, have a day’s battery life, be
an inch thick and always connected, anywhere.
And I should be able to seamlessly roam in the LAN or WAN environment.
Continuous innovation in the area of wireless applications will ensure that the
PC is connected to other devices in the home or office. At home, one could have
the PC island, the TV island and the music system island. The personal computer
can now become the server for the network that wirelessly connects all these
islands. Connectivity within the home will be the focus of the PC and component
industry over the next two years.
Talking of competition, what about AMD’s view that speed is not
everything to a processor?
We have never said that megahertz was the key to performance. We do two
things. We base on our marketing and industry perception of the hardcore
technical things that you can measure. The other thing we do is benchmarks.
Those things can be measured. This is how we think we can communicate to the
users and make further progress.
You have been very secretive about the Banias chip for notebooks, expected
later this year...
No. Not as yet. Part of the reason why we are secretive is that we ourselves
have not seen the baby. All I can say is that things are looking good. We are
very optimistic about maintaining our schedules.
The P4 will touch 3 GHz and 533 MHz bus speed by end-2002. Are other PC
sub-systems keeping pace with processor development?
We have put in a lot of effort to deliver a balanced effort. As we move to
well above 3Ghz in the future, I/O system scalability will be a key issue and
that is the driving force of the PCI Express (3GIO) effort.
What is Intel’s take today on integrating more functionality into the
processor or the chipset, taking those functions away from additional systems or
plug-in cards? We already have ‘software modems’; Intel has graphics
chipsets, which do away with graphics cards. Now, with the Banias (new mobile
chip), the accompanying chipset will include 802.11 functionality. How far does
this go? Is this changing the processor/chipset role from ‘the heart of the PC’
to ‘almost the entire PC’, with other functionality in software? What is the
role of sub-systems vendors, such as graphics companies like Nvidia or others?
This is the continuing thrust of silicon innovation. The cycles of
innovation, standardization and integration continue. We see no end to this. In
the future, things like wireless communication, optical networking, video and
speech inputs, new models of storage and new types of outputs will continue to
provide the next elements of these cycles.
McKinley is one of the biggest chips ever. While you keep pushing silicon
and keep up with Moore’s Law, is this the path? You have declining yields with
such large chips, and difficult manufacturing... Are such expensive, low-volume
chips worth it as against cheaper, more efficient chips deployed in
multiprocessing or massively parallel architectures?
Both the ‘big’ and the ‘parallel’ approaches add value and in many
cases are complimentary. On very large database machines, the very large caches
have very high performance value. Further, to enable higher degrees of MP, it is
essential to minimize bus traffic, which is exactly what the large caches
accomplish. Caches minimize latency and bus bandwidth. In this way, the on-chip
vs off-chip cache is a very good tradeoff and is a big factor in the improved
performance of McKinley. Finally, the yield is not an issue given Intel’s
tremendously strong process/manufacturing capabilities.
The McKinley has an on-chip 3MB level-three cache, a 256KB level-two cache
and a 32KB level-one cache. Is this complexity really necessary? Do you get a
big performance gain versus the Itanium’s external 4MB level-three cache? Are
you satisfied with the Itanium’s success in convincing CIOs that Intel chips
can power high-availability, high-performance enterprise systems? The
traditional RISC and UNIX systems, including Sun and IBM, seem to be doing just
fine...
Yes and no. It is still early and the overall program is later than we would
have hoped. However, we are gaining momentum, the design wins and the amount of
excitement for McKinley is high.
Manoj Chandran in Bangalore