More news about the delay in Merced...
Intel's Merced chip may slip further
By Stephen Shankland
Staff Writer, CNET News.com
July 8, 1999, 6:15 p.m. PT
Prototypes of Merced still aren't out, and it looks like Intel's next-generation processor may slip until the
third or fourth quarter of next year.
This latest lag, which could put off Merced's release by a few weeks or months, won't be as significant as the
delay Intel imposed last year when it shifted Merced's debut from the end of 1999 to mid-2000.
Still, the revised calendar may sap Intel of some psychological momentum and will likely solidify arguments
that Merced, Intel's first server and workstation processor designed to compete directly against the most powerful
Alpha and UltraSparc processors, will primarily serve as a test vehicle. Customers and computer companies will
buy it, but will hold off on migrating to Intel's chips until McKinley, Merced's successor and the second "IA-64"
chip, comes out in late 2001.
"My sources tell me first silicon was originally scheduled for the first of June, and now it's late August, so I think
there's been some slippage over there," said Linley Gwennap, senior analyst at Cahners Microdesign Resources.
While customers don't see first silicon, a delay here pushes other events into the future. A yearlong wait is
typical between first chips and first shipping systems even with understood architectures, Gwennap and others
say.
"One year from first silicon to first system on a brand new architecture
...seems a bit optimistic," he said, adding that he expects the first systems to
arrive in the fourth quarter of 2000.
Intel denies that the schedule has slipped, saying that the company is still
on track to produce the chips in volume midway through 2000. The first
Merced prototypes aren't due for several weeks, but the company remains
on schedule to get samples out in mid-1999 and chips into volume
production in mid-2000, according to Ron Curry, director of marketing for
the IA-64 processor division. Some people "are taking statements of 'mid'
far too literally," he said.
"We're on track for what our [original equipment manufacturers] need to be
able to move this thing into production," Curry said. "There have not been
any problems with the program since we got past the hiccup last year."
The company has also ramped up an aggressive venture capital fund to
support application developers building products for the new architecture.
Other sources at Intel, however, added that volume production typically occurs a year after samples arrive, and
that systems only start appearing a quarter or more after production.
In May 1998, Intel said it would have to delay Merced from late 1999 to mid-2000. The delays happened
because Intel found it harder than expected to develop several aspects of the new architecture concurrently,
Curry said.
Intel has eight different operating systems running on simulators of the Merced system, including Windows NT,
Linux, and several varieties of Unix.
Merced only a test bed?
Analysts agree that Merced will be used chiefly for debugging IA-64 and putting the new architecture through its
paces. The buyers of the high-end servers where IA-64 will initially appear are the most conservative in the
industry, they say. "Merced is going to be like the pace car lap at Indianapolis," said Insight 64 analyst Nathan
Brookwood.
As computer makers and their customers have eyed the issues with Merced, staunch IA-64 supporters
Hewlett-Packard and SGI have extended their own chip road maps.
Curry acknowledges that Merced "is the initial member of the [IA-64] family," but says that many computer are
investing in the architecture for actual use, as well as mere testing.
Though Brookwood hasn't heard about new delays to Merced, he said he had expected Merced to have "taped
out" by now--in other words, to have finished the design and sent it off to be manufactured. But a bigger
problem is that Intel faces the issue of developing not just the Merced chip, but also supporting chips and the
compiler, the software that translates the programs people write into language understood by the chip.
Compilers are always important to chip performance, but they're even more critical to IA-64.
The Merced schedule further diminishes the significance of Merced, the first chip in Intel's plan to bring its
dominance to high-end computer systems, putting more pressure on its sequel, McKinley, Gwennap said.
"Had Merced come out sooner, or if it had come out with really killer performance, that would be bad for the
[Ultra[Sparcs and the other competitors," Gwennap said. "At this point, everyone is expecting it's going to be
late and slow, and the real advance is going to come from McKinley. What this does is puts a lot more pressure
on McKinley and for that team to deliver."
Each delay means it's harder for Intel to keep up with Moore's Law. "Every time you slip a month, you'd better
be adding 4 percent to your performance," Gwennap said. "At some point, they've got to get something out."
Ashok Kumar, an analyst with Piper Jaffray, said the hardware issues with Merced are less significant than the
software. "Operating systems are going to be barely ready," he said.
Minor delays to Merced aren't that important in the scheme of things, Brookwood said. "On a multiyear effort
like Merced, worrying about two weeks one way or the other is not terribly productive," he said.
"Frankly, even in 2000, the financial impact of Merced is not that great," added Joe Osha, a financial
analyst at Merrill Lynch.
However, the effects on Intel's image can be much larger, Osha said. "Psychologically, it is enormously
important. A lot of people investing in Intel are very much focused on whether they can execute this
transition," Osha said.
IA-64 is key to Intel's long-term plans to keep profitable by selling more expensive, high-end chips. The
popularity of sub-$1,000 computers and the aggressiveness of competitors such as AMD have reversed
Intel's historical annual increases in the average chip sale, Brookwood said. "In order to keep the
average above $200, they've got to sell more high-end chips," he said.
Further delays to Merced are possible, but may not be significant, said Mark Edelstone of Morgan
Stanley Dean Witter. "At this point, most people understand that Merced itself is not going to be a very
compelling product for Intel," he said. "If you look at the performance of Merced, it's really not going to
be meaningfully different from what you get at the high end of the 32-bit architecture."
Edelstone expects samples of the first Merced chips to arrive in late September at the earliest.
Curry said that manufacturers aren't accusing Intel of being late. "We expect to have samples within a
few months to OEMs. That's within their expectations. None of them are walking around saying it's late.
As of today, the product has not slipped the schedule," Curry said.
fred |
