SAN FRANCISCO -- Even the biggest bunglers get a second chance.
Advanced Micro Devices
, the do-over comes Thursday with the launch of its new server microprocessor, dubbed Shanghai.
After the disastrous launch of its previous chip left AMD bleeding red ink and market share for much of the past 18 months, AMD can't afford to drop the ball again. Every aspect of the Shanghai rollout must be flawless if AMD hopes to have any kind of future.
The good news is that AMD is off to a fine start: The new processor is shipping on schedule with specifications that appear to be on target (even ahead of target, if you ask AMD).
The question is whether the conservative mentality that AMD developed as a result of its near-death experience will prevent it from getting a leg up against
, which is prepping a major upgrade to its own chip line.
And with the state of demand and the global economy in question, the upside of new products may be limited for both companies. On Wednesday, Intel shaved nearly a billion dollars off its
Nathan Brookwood, a semiconductor analyst at Insight64, says AMD is back on track with the release of Shanghai.
But, he notes, AMD doesn't have any products in its pipeline that will deliver the type of performance advantage to leave Intel in the dust.
"They have made a conscious decision, at least as I read the tea leaves, on programs that have relatively low technical drift, because they don't want to miss anything," he says. "The offshoot of that is they're not taking those big risks where you might get a giant leap."
For AMD shareholders, who have seen the stock crumble 79% in the past 12 months, simply fielding a working product in large volumes is progress.
And for AMD, which is undertaking a radical organizational makeover that will
into two entities -- one focused on chip design and the other on chip manufacturing -- limiting the risk associated with new products is probably not such a bad idea.
The company will provide Wall Street with a much-anticipated briefing Thursday, providing key details of its new, post-split financial model.
Pat Patla, the head of AMD's server and workstation group, says the company learned its lessons from the fiasco of the company's Barcelona chip, which shipped with disappointing specifications and suffered from a technical glitch that forced the company to delay availability. AMD tried to do too many things with Barcelona, he says, noting that 80% of the chip's original design was new.
In developing Shanghai, AMD has made a slew of organizational changes to ensure that things run smoothly, such as centralizing the management of the teams responsible for the various aspects of creating the chip.
AMD also brought its customers in the loop sooner, giving them a say on which features and bug fixes to prioritize throughout the process. That's a significant change from the last go-round, where AMD customers didn't even receive production units of the new chips until a few weeks before the product was officially launched.
Shanghai doesn't represent a brand new microarchitecture, so much as a refinement of AMD's existing Barcelona architecture.
In Shanghai, AMD has shrunk the width of the transistors to 45 nanometers, from the 65-nanometer transistor chips in its Barcelona chip, allowing AMD to lower its manufacturing costs, improve the chip's performance and add new features. The result, says AMD, is a chip that is 30% faster than the Barcelona processor, and is well-suited to handle tasks like virtualization.
"I do get the sense that they
AMD hit their targets and they did it pretty well," says analyst David Kanter, of Real World Technologies. Shanghai is "definitely a good sign that the folks at AMD are executing," he says.
Computer makers including
are all expected to announce servers that feature the Shanghai chip at, or shortly after, its official launch Thursday.
AMD won't have very long to enjoy its moment in the sun, though. The competitive landscape will undergo a tectonic shift in the first quarter of 2009 when Intel releases its Nehalem microprocessor (Intel will actually ship a desktop PC version of Nehalem in the fourth quarter 2008, but its mainstream, two-socket server version of Nehalem isn't expected until early 2009).
In Nehalem, Intel is actually mimicking AMD's microarchitecture, by integrating the so-called memory controller directly onto the chip, and revamping the way data flows between the processor's multiple cores. That's a nice validation of AMD's engineering, but it also means that Intel's new chips will receive a significant performance boost.
AMD argues that it will have its own follow-on to Shanghai, code-named Istanbul, shipping in the second half of 2009, and that that product is the better comparison with Intel's Nehalem.
And AMD points out that Intel's Nehalem chip won't be compatible with Intel's previous generation of hardware -- customers that want to use the Nehalem chip will need to upgrade the motherboard and related chipsets in the servers too, making for a more expensive transition.
But regardless of whether AMD is fielding Shanghai or Istanbul, the fundamental chip design advantages that have long differentiated AMD's products from Intel's will disappear when Intel releases Nehalem.
"The era of AMD being a leader on the CPU
central processing unit side is kind of coming to a close," says Kanter, of Real World Technologies.
"I think AMD, to some extent, is going to be forced back into their role as a value supplier," he says.
AMD will have a bit more breathing room in the market for processors intended for four-socket servers, currently AMD's stronghold, and a class of product in which Intel is not expected to adopt the Nehalem architecture until the second half of 2009.
That's not to say there won't be market segments where AMD's products continue to provide better performance or more desirable characteristics than Intel's offerings. Analysts note that raw performance is not the only criteria that customers consider when evaluating which chips to power their servers with.
Some industry-watchers also suspect that AMD may try to catch up with Intel's manufacturing lead, by producing chips with 32-nanometer circuits sooner than AMD initially planned. And nobody really knows what AMD may have up its sleeve with its next-generation chip architecture, dubbed Bulldozer, expected in the 2010 time frame.
But it's clear that for the next couple of years at least, the rules of the game governing the microprocessor market have changed.
The burden is now on AMD to demonstrate how it can adapt and thrive in this new world order.