Microprocessor
Watch
Issue #20
MicroDesign Resources --- October 13, 1999
Editor:
Michael Slater
Contributors: Linley Gwennap, Keith Diefendorff, Peter Glaskowsky
Special
Seminar Invitation: Michael Slater on PC Processors
Michael Slater, founder of Microprocessor Report and Microprocessor
Forum and editor of this newsletter, will present his seminar
"Processors for PCs: A Business and Strategy Perspective"
on November 4 at the Westin Hotel in Santa Clara. This highly
regarded seminar will be fully updated with the latest Coppermine
and Athlon disclosures, as well as issues regarding the rollout
of RDRAM. The seminar runs from 9 am to 5:30 pm; registration
is $795. For more details or to register, see http://www.MDRonline.com/dm
or call 800.700.4004 or 707.824.4004.
In
This Issue:
- Rise
Revisits Roadmap
- Cradle's
"Universal Microsystem" Is Most Flexible Chip Ever
- Editorial:
Intel Commoditizes 3D Graphics
- Resources:
Processor Options for Digital TV
- Resources:
Deep Impact. Armageddon. Comdex.
- Resources:
Micro-32 Symposium, Workshops, and Tutorials
- Resources:
Wear Your Hard Drive on Your Sleeve
Rise
Revisits Roadmap
Rise
Technology is revisiting its roadmap (see MPR 5/31/99, p.
15) in light of the new competitive landscape that has resulted
from Via's acquisition of Cyrix and Centaur. The company withdrew
its presentation from Microprocessor Forum, where it had planned
to describe a Socket 370 chip code-named Tiger. It is now
reconsidering whether to focus its resources on Tiger or on
the mP6 II, a Socket 7 device with on-chip L2 cache.
Rise
had samples of its original mP6 rated at PR333 and PR366 (250
MHz) in a 0.25-micron process but decided not to ship these
chips. Instead, it is now sampling these speeds using 0.18-micron
silicon and is beginning to ramp up production. The company
expects to have PR400 and PR433 chips in production by the
end of the year. Pricing for the PR333 and PR366 chips is
$43 and $48, respectively.
With Intel's
bottom-of-the-line Celeron now at 400 MHz, Rise's current
offerings place it below the bottom of the desktop PC market.
Company officials say most of the customers are in Asia, with
interest coming from makers of consumer notebooks and mininotebooks
who are attracted to the chip's low power consumption; the
mP6-366 consumes a maximum of 5 W from its 2-V supply. Company
officials also indicated that they are receiving strong interest
from makers of Linux-based PCs, set-top boxes, and thin clients.
--M.S.
Cradle's
"Universal Microsystem" Is Most Flexible Chip Ever, But Is
the Market Ready?
Imagine
that you want to design a full-custom processor to meet the
exact needs of your application. You don't have time to design
an ASIC, and you don't want to learn a hardware description
language like Verilog or VHDL. A conventional microprocessor
would be too slow, and a field-programmable gate array (FPGA)
can't run your C code. You can't even find a media processor
with the right I/O interfaces. What you want is an off- the-shelf
chip that can be programmed with the software and hardware
elements of your design. You are the target market for Cradle
Technologies' new Universal Microsystem (UMS).
The UMS,
announced at last week's Microprocessor Forum, combines a
parallel array of microprocessor and signal-processor cores
with programmable logic and protocol engines to handle all
I/O. A simple DRAM controller is the only hardwired peripheral.
If it
is truly faster, easier, and cheaper to implement sophisticated
chip designs in a UMS device than with existing alternatives,
Cradle's success is virtually guaranteed. If, however, the
complex UMS mix of processors and programmable logic is not
as easy to configure as Cradle promises, the full potential
of the new architecture may never be realized. Cradle hopes
to gain a share of the competitive networking and consumer-
electronics businesses, and it may also be able to earn a
portion of the low-volume but high-margin field-programmable
gate array business. It won't be easy for Cradle to break
into any of these markets, but its Universal Microsystem could
be the right tool for the job. --P.N.G
Editorial:
Intel Commoditizes 3D Graphics
With the
introduction of the 810E chip set, Intel has pushed integrated
graphics from the low end to the mainstream of its product
line, leaving discrete 3D accelerators for only the most performance-conscious
users. By next spring, the majority of new PCs are likely
to be using a chip set with integrated 3D graphics--most with
Intel's name on them. This represents a huge change from early
this year, when nearly all PCs contained a discrete 3D-graphics
chip.
While
integration has been a powerful force in PC design for years,
this transition is unusual: in many cases, PC users will see
a decrease in performance when comparing the new integrated
parts with older designs. The 3D core built into Intel's 810
and 810E is competitive with the least expensive discrete
3D chips available today but behind the chips typically used
in midrange PCs. The success of Intel's integrated devices
indicates how few OEMs and PC buyers care about 3D performance.
Intel
has helped foster this attitude in the consumer market with
the creation of the Celeron brand. Intel believes that anyone
buying a low-end PC cares only about megahertz, not actual
performance. The success of the Celeron processor, with its
slow front-side bus and lack of SSE, substantiates this belief.
Buyers in this segment don't seem to care that the 810 chip
set has limited 3D performance. They just like the big megahertz
number on the CPU.
More
sophisticated consumers are looking for good deals on Pentium
III systems, which can be found for less than $1,000. Many
of these less expensive Pentium III products will be switching
to the 810E in the near future. This change will drive some
consumers--particularly gamers--to move up to an 820 system
that will probably come with a more powerful, and more expensive,
Intel processor.
Most
PCs are still bought by businesses, large and small, and few
business buyers are interested in 3D performance. There are
still no mainstream business applications that have 3D components,
and nothing seems likely to appear in the near future. Thus,
most business buyers are happy with "checkbox" 3D--as long
as the PC has a 3D accelerator, they're happy. For these businesses,
the 810 and 810E fit the bill, and carry the Intel brand to
boot. Some business PC buyers have a bit more foresight. Seeking
a four-year lifetime for their PCs, they don't want to ignore
an area such as 3D performance that could become critical
during that period. In addition, the incremental cost of a
more powerful system is a small fraction of the total cost
of ownership (TCO) during that system's life. These buyers
will be well served by seeking out 820-based systems with
moderately priced but powerful 3D accelerators available from
a variety of chip makers.
Intel's
integration strategy leaves these 3D chip makers to compete
in a shrinking market for discrete graphics accelerators.
Naturally, the biggest impact has been on market leaders ATI
and S3, which have been the vendors of choice for low-cost
and midrange 3D chips. These are the very sockets being eliminated
by the 810 and 810E. As a result, Intel is likely to become
the leading vendor of 3D-graphics accelerators within the
next few months.
Intel's
flank attack exposes a key fallacy of the 3D market: that
performance is the most important factor. Sure, there are
a bunch of gamers and a few professional users who care, but
without mainstream 3D applications, most PC makers aren't
putting a priority on 3D performance. When several vendors
offer chips with similar features, OEMs will naturally choose
ones with the best performance in their price class. But by
integrating 3D into the chip set, Intel has sidestepped this
comparison and reduced system cost.
Over
time, emerging applications will increase mainstream interest
in 3D performance. (Indeed, Intel itself is promoting some
of these applications, hoping to boost sales of its high-end
CPUs.) Intel must be careful that its brand name does not
become associated with inadequate 3D performance. It's quite
possible, however, that emerging applications will be satisfied
by whatever level of commodity 3D Intel can deliver at the
time, making moot any competitive performance issues.
Other
3D vendors must figure out how to compete in this new world.
Creating their own integrated chip sets, as Nvidia has done
and S3 is doing, is the obvious approach, but these vendors
will find it difficult to displace the 810 family, given Intel's
entrenched position in the chip-set market. The high-end 3D
market remains viable and active, but Intel's integration
strategy has quickly halved the available market for third-party
3D chips.
If this
trend continues, the discrete 3D chip will go the way of the
math coprocessor. A high-end niche will remain for quite some
time, but to be successful, 3D chip makers must counter Intel's
inevitable integration efforts. --L.G.
Resources:
Processor Options for Digital TV
Microprocessor
Report's November 4 dinner meeting features a presentation
by Cees Hartring, VP and GM of Philips' TriMedia business
line, and Gert Slavenberg, fellow and cheif architect of the
TriMedia processor, titled "Processor Options for Digital
TV Applications." The dinner presentation will describe the
requirements for digital TV processing and then review and
compare a number of processor options. Preliminary information
will be given on Philips' next-generation media processor,
currently in development, which will boost performance by
more than an order of magnitude.
Microprocessor
Report dinner meetings are a great opportunity to network
with your colleagues, enjoy a good meal, and hear about the
latest developments. Registration is only $99 (less if you
also attend one of the seminars also offered that day). See
http://www.MDRonline.com/dm
for more details and to register, or call 800.700.4004 or
707.824.4004.
Resources:
Deep Impact. Armageddon. Comdex.
An enormous
inert mass is hurtling toward us, and nothing short of an
atomic blast can turn it away. Yes, it's once again time for
Fall Comdex, "celebrating" its 20th year in Las Vegas on November
15-19. Not just trapped inside Windows, the show this year
is paired with the Linux Business Expo. The enormous gathering
features the usual keynotes (Gates, McNealy, Chambers, and
Mr. Linux himself), the usual slew of product introductions,
and the slogan "the new TechnoMillennium." Registration fees
range from $1,095 for the conference to $1,999 (get it?) for
the full "Millennium Passport." Register before 11/5 to weasel
into the exhibits for free. For more information, access http://www.comdex.com.
If you
can take some time out from the show floor and the private
meetings, there's something else worth attending at Comdex:
Michael Slater will chair a conference session on Wednesday
morning, November 17, at 10:15, titled Future Watch: PC2020
-- Beyond Faster and Cheaper. With panelists Jeffrey Harrow
(Compaq), Phil Hester (IBM), Gerald Holzhammer, (Intel), Scott
Smyers (Sony US Research Labs), and Carl Stork (Microsoft),
a lively and insightful discussion is all but assured.
Resources:
Micro-32 Symposium, Workshops, and Tutorials
If something
more thoughtful than Comdex--and a destination more interesting
than Las Vegas--appeals to you, attend the thirty- second
annual IEEE/ACM International Symposium on Microarchitecture,
to be held November 15-18 in Haifa, Israel. The symposium
has long provided one of the premier forums for discussing
and debating issues relating to instruction-level parallelism,
compilation techniques, and microarchitectures. On Monday,
November 15, you can choose from three workshops and tutorials,
covering feedback-directed optimization, media processors
and DSPs, and low-power processor design. The three- day symposium
on November 16-18, keynoted by Intel's Fred Pollack, covers
a wide range of current microarchitectural topics presented
by many leading academic researchers and a few architects
from industry. To get more information or to register, access
http://huron.cs.ucdavis.edu/Micro32.
Resources:
Wear Your Hard Drive on Your Sleeve
If clipping
a cell phone, a pager, and a Palm computer to your belt doesn't
make you feel enough like Batman, pretty soon you'll be able
to wear a full-blown PC, too. The future of wearable computing
is explored in a fascinating white paper, the result of a
study conducted at Carnegie Mellon University (CMU). The researchers
drew upon their backgrounds in anthropology and industrial
design to study spaces on the human body where solid and flexible
forms can rest without obstructing free motion. The paper
is available for free on the Web. Download it in Adobe Acrobat
(PDF) format at http://panopticon.edrc.cmu.edu/design/wearability/front.html.
|