A dazed Intel shifts into comeback mode

On Monday, Intel will release a server chip, code-named Woodcrest, that's part of a trio of processors debuting this summer that have emerged as one of the most important product launches in Intel's 38-year history.

Like past Intel product releases, the new chips will lay the groundwork for a new generation of computers and generate sales for software developers and hardware makers. Hewlett-Packard, for instance, will come out with a wide range of servers based around Woodcrest, including one box that the company says will be the fastest two-processor server in its class. But more importantly, the chips will show whether Intel can regain its technological edge.

Confidence runs high.

"(On) average, we are 40 percent-plus better than the competition in performance and a little bit better in power, and the combination is 1.5X in terms of power/performance," said Pat Gelsinger, general manager of the Intel enterprise group. "Customers are very enthusiastic about this platform. We expect to see a rapid take up."

The notebook cousin of Woodcrest coming out later, code-named Merom, will provide 20 percent better performance without a power penalty, said Mooly Eden, general manager of Intel's mobile-platform group.

The buoyancy is a marked change. Since 2004, the Silicon Valley giant has stumbled through delays, cancellations and missed opportunities. Intel's hallmark products--its processors--have often lost in benchmark tests against chips from rival Advanced Micro Devices.

Inventory imbalances, once the sort of problem associated with AMD, have dented Intel's profits for three quarters.

During the same time, a newly confident AMD has won customers and kudos for its Opteron and Athlon64 chips.

As a result, Intel's market share has clicked down from the mid-80s to 77.2 percent. More ominous, AMD's gains didn't come at the low end. Instead, it's won large segments of the server market. In the first quarter, AMD accounted for 22.1 percent of all x86 servers, compared with almost zero percent during the same period in 2003.

Along with the new chips, Intel is in the midst of a thorough re-examination that some say may prompt Intel to jettison unprofitable divisions, lay off employees and cut back middle management. For CEO Paul Otellini and his lieutenants, the pressure is on.

"This is a fairly unique situation in Intel's lifetime," said Dean McCarron, an analyst with Mercury Research. "You're seeing the entire road map of Intel being rewritten with one stroke of the pen."

The company has also committed itself to coming out with a new chip architecture every two years, a fairly aggressive cadence, even by Intel standards.

For its part, AMD has scoffed at Intel's trumpeting.

"It's driven by the fact that they can't talk about their current products, because everybody knows their current products aren't very good," Henri Richard, AMD's chief sales and marketing officer, said in an interview earlier this year. Richard has also said that Intel slipped because it got complacent.

Getting to the root of things
The seeds of Intel's problems, arguably, can be traced back to 2000. The Pentium 4 generation of processors focused on getting tasks done as fast as possible and running at high speeds. The heat generated by these chips was so excessive, however, that Intel never got to its goal of cranking them up to 4GHz.

Then in 2003, AMD's first Opteron-Athlon 64s came out. The chips ran slower but more importantly, these chips came with features that added performance without adding heat. They featured an input-output technology called HyperTransport, as well as an integrated memory controller. Together, these two features probably added about a 15 percent performance boost alone, AMD executives have estimated.

These chips also featured 64-bit processing, which adds performance. Almost no one who has bought a desktop or notebook with a 64-bit chip actually runs 64-bit software. As for servers, executives at Sun have said only about 30 percent of the Opteron server customers run the 64-bit software. But it was a great marketing tool. Intel eventually put in 64-bit capabilities.

Gelsinger is blunt: Intel had the wrong chip for the wrong time.

"We didn't deliver the best products (in the server market) for a number of years," he said. "Competitors took advantage of that, and now we have to come back and deliver the best products to win back the confidence of our customers and end users."

One of the low points came when Dell announced this year that it would put AMD chips in one of its four-processor servers.

"I was pretty personally dismayed by it. Not dismayed at Dell, but dismayed that we didn't deliver the products necessary for our most loyal, long-term customer to remain fully on Intel," Gelsinger said. "It's ironic that it came on the eve of when we will deliver better products."

The picture, though, has been changing since the end of 2005. Since then, Gelsinger noted, Intel has put out chips on time, or ahead of time.

Unlike the Pentium 4 chips, the new Core chips will do more work at lower speeds. Coincidentally, Intel engineers started warning about the looming power crisis at the time the Pentium 4 was coming out.

A new processor architecture can have a huge effect on a chip company. In 1999, AMD was faced with mounting financial losses and the departure of its No. 2 executive, Atiq Raza. In October, an AMD engineer named Fred Weber unveiled plans to build a 32/64-bit chip (which can run both kinds of code). When it came out as Opteron almost four years years later, AMD's seemingly never-ending melodramas became a smooth story of success.

This time around, the engineering heros could be Intel's Israeli design teams. Intel Israel came up with the low-powered Pentium M for notebooks and are credited with designing the Core architecture behind Woodcrest, the desktop chip code-named Conroe coming in July, and Merom, which comes shortly afterward.

Even with the performance gains, Intel likely won't be able to turn the clock back completely, analysts have said. AMD has forged tight working relationships with HP, Sun and IBM.

Government agencies and some large corporations have also become accustomed to buying AMD-based servers. AMD is now trying to get these customers to buy desktops and notebooks based around its chips and break the decades-long dominance of Intel. HP has tweaked its desktop line to make AMD desktops more palatable for this crowd.

"There is a segment of customers that are still Intel loyalists, but it is becoming more of a server-brand articulation," said Paul Miller, HP's marketing vice president for industry-standard servers.

The chips from both companies differ enough to keep alive the debate over which one is best. Both companies have four-core chips coming next year. AMD will sport a shared cache of memory on its quad-core chip, a first for AMD and one of the hallmarks of what makes a dual- or quad-core chip into an integrated whole. (Intel already does this.)

Intel's quad-core chips will be made up of two dual-core chips with shared caches inserted into a single package. The architecture behind the Intel chips, however, will be relatively newer. Each chip will also be tuned for a different type of memory.

"As all industries become more dependent on computing to improve productivity and visibility in inventory, it makes sense that there's going to be some diversification in the chip world," said IDC analyst Richard Shim.

So far, The Woodcrest-Merom-Conroe generation looks pretty good. Dell has measured up to a 152 percent performance increase delivered by the Woodcrest processors--combined with a 25 percent decrease in power consumption--compared with Intel's first dual-core Xeon processors, said Jay Parker, director of Dell PowerEdge servers.

"There's times where new products align with a new architecture that gets a 5 to 15 percent improvement. When you're talking about something this dramatic, that's meaningful and drives customer behavior," he said. HP's Miller said that power consumption and virtualization will be two strong selling points for Woodcrest servers.

The navel-gazing also sometimes obscures perspective. Intel, pulling in more than $8 billion in revenue and $1 billion in net income a quarter, is still one of the biggest companies in IT. Apple Computer became a customer last year when IBM couldn't provide notebook chips that ran cool enough.

Even if AMD executes flawlessly and fully builds out its factories, the company still couldn't produce enough chips to pass up Intel any time soon. AMD's company's goal is to achieve 30 percent market share by 2008. The company has also yet to commercially produce chips on the 65-nanometer process, something Intel started doing last year.

Still, it's up to Intel to show that it's as good as it used to be.