Merced mired (comments???)
By Michael Kanellos
Staff Writer, CNET NEWS.COM
August 6, 1998, 4:00 a.m. PT
Touted as a major milestone for Intel and the computer
industry in general, Merced, the company's first 64-bit
chip, appears to be losing its luster because of delays,
performance issues, and upstaging by other chip vendors.
Industry experts and even
Intel executives have made
comments that call into
question the commercial
viability of Merced, which
has been pushed back to
mid-2000. It now appears
that the chip which will
propel Intel deep into 64-bit computing territory will be
McKinley, the Merced successor that will likely come out
in 2001.
"If Merced slips another couple of quarters...this could
lead Intel to not market Merced as a product but use it
only as a development vehicle," said Linley Gwennap,
publisher of the Microprocessor Report, writing in the
most recent issue of the respected industry newsletter.
To be sure, Intel has been sending out signals. In May of
this year the chipmaker said that Merced would be
delayed due to its complexity.
Just as telling was a comment made last year at a chip
conference by Fred Pollack, Intel Fellow and Director of
Measurement, Architecture, and Planning. In saying that
the chip following after Merced (McKinley) will "blow
your socks off," he left many attending the conference
with the implication that Merced would not.
And McKinley will not only bear the 64-bit flag but also
become the first Intel chip to use copper interconnect
technology, according to sources familiar with Intel's
plans. Copper, which conducts better than the aluminum
circuits found in current chips, is expected to lead to
faster processors.
But to get to McKinley one has to go through Merced.
"Sources indicate that [Merced]...is facing problems that
could jeapordize [its] existence as a viable product,"
Gwennap wrote.
Merced's dimming commercial appeal arises in part from
its delayed launch. "There's no question that more and
more lately we're hearing the mantra 'Wait for McKinley,"
said Keith Diefendorff, editor-in-chief of The
Microprocessor Report.
Intel, however, believes strongly in the chip and has
stated clearly that the chip is simply delayed, an issue
which often arises with the design of any highly complex
chip.
Merced is expected to debut at a clock speed of
800-MHz--compared to today's 400-MHz chips--and be
capable of processing six to eight instructions per every
"tick" of the chip's clock cycle. That would constitute an
architectural improvement over current and competing
chips, according to the report authored by Gwennap.
Additionally, Merced will use an advanced manufacturing
process referred to as 0.18-micron. The measurement
describes the circuit line width of a processor. Current
Intel chips are made according to the 0.25-micron process.
The smaller the number, the more powerful the chip.
Yet despite these impressive gains, Gwennap points out
that they would be considered more of an accomplishment
if the chip came out in late 1999, as originally planned. By
mid-2000 some of these goals will be neutralized by gains
made by other chips.
Though quite different from Merced in some respects, the
21264 Alpha chip from Compaq, will likely be available at
speeds greater than Merced. Compaq is sinking
considerable resources into Alpha and touting it as a
highly viable 64-bit technology--especially since it is
already on the market.
IBM and Sun have made similar predictions about speed
gains for their own chips.
Adding salt to the wound, IBM, the other large,
longstanding U.S. chipmaker, has also been stealing the
technology march on Intel over the past few months. This
week the company announced a new chip breakthrough
called SOI which should yield faster chips. And in
October of 1997 Big Blue was the first to announce its
intention to use copper. Several other chipmakers have
followed.
Hewlett-Packard (HP) is also showing increased signs of
development activity, according to sources. Although the
current stated roadmap for its proprietary PA-RISC
architecture processors stops at the 8700 chip due next
year, others are in the works.
"I'm not convinced [Merced] will run faster than an
equally fast RISC [Reduced Instruction Set Computing]
chip," said Dieffendorf. "I've not seen enough
architectural cleverness to say superscalar RISC is dead."
RISC is a chip architecture employed in Sun's Sparc
design, IBM's PowerPC chip, and HP's PA-RISC
processors.
But as it turns to McKinley, Intel expects to boost
performance significantly. While the chip will share the
basic "IA-64" (Intel 64-bit) architecture with Merced, two
fundamental changes will be apparent in its design. First,
the chip will in all likelihood be made on the even more
efficient 0.13-micron manufacturing process. Second, if
McKinley comes out on the more advanced process, it
will use copper interconnect technology instead of
aluminum.
Diffendorfer, among others, said that McKinley's 2001
release will roughly coincide with Intel's adoption of the
0.13-micron process. Howard High, an Intel spokesman,
said that Intel will only use copper technology with this
process, that aluminum will not be used on chips for the
0.13-micron generation. A Merced successor, he admitted,
can roughly be expected around the time this
manufacturing technology comes out.
Until then, Merced, and what to do with it, will be an
issue. Speaking on a purely speculative basis, Diefendorff
said that one of the problems in developing a "new"
architecture for Merced and McKinley may have come as
a result of Intel's business methods. Intel typically takes a
top-down approach to chip design, hiring fleets of
designers once a basic strategy is outlined. Merced seems
to have been no different, with employees inside of Intel
stating that a number of engineers came to the project.
"They do deploy an unusually large number of people
compared to anyone else in the industry," he said.
Unfortunately, smaller is often better. Nonetheless, he
added, "They have done a lot of chips. You can't argue
with their methodology."
One factor that may help is HP's role in developing
McKinley. HP has greater experience in designing
high-end computer systems and processors, and sources
close to the company said earlier that it is participating
more than in the Merced project.
Gwennap's report predicts that McKinley will probably be
a commercial success. A number of workstation and
server vendors have already shifted business plans to
accomodate the chip, he noted.
Yet ironically, Intel's recently released Xeon line of
Pentium II chips for workstations and servers will compete
with the Merced technology. Xeon chips are expected to
hit 700 MHz by late 1999, according to MicroDesign
Resources, and will be made under the 0.18-micron
process. The chips are based around the familiar 32-bit
Pentium II architecture, however, making it a safer bet for
computer vendors.
"If the performance is similar to Merced, there isn't the
incentive to push to the new architecture," noted Kelly
Spang, processor analyst with Technology Business
Research. Instead, computer vendors may say, "`Let's
squeeze more profit out of Xeon.'"
A third factor will be the Year 2000 problem and the timing
of the release of Windows NT 5.0, observed Ashok
Kumar, technology analyst with Piper Jaffray. Merced will
be coming out at a time when corporations will likely be
grappling with their Year 2000 problems and before the
real migration to high-end corporate use of NT. The
incentive to consider an IA-64 chip really won't come until
2001, when McKinley comes out.
Despite these glitches, Dean McCarron, principal at
Mercury Research does not see long standing roadblocks
to the proliferation of Intel's 64-bit architecture.
"In reality, no one is moving away," he said. |
