News Analysis: AMD executives say they believe they can correct and learn from their chip production mistakes in 2008.
In the months leading up to the launch of Advanced Micro Devices' quad-core Opteron processor,
code-named Barcelona, company executives continually harped on its design as distinguishing it from Intel's quad-core offerings.
When 2007 began, AMD looked poised to follow up on the success of its dual-core Opteron chip by manufacturing a quad-core processor that would place four processing cores on a single piece of silicon. The design of these was a direct assault on Intel's approach to its quad-core offering, which tied two dual-core Xeon processors together.
It now seems that AMD's approach to quad-core design led to some of the numerous problems associated with Barcelona, namely underwhelming clock speed performance, a poor manufacturing ramp-up before its delivery and a bug that could disrupt the hardware. As the calendar turns to 2008, the company insists that it has learned its lesson and will be able to get its quad-core chips back on track.
AMD plans to have production-ready quad-core Opterons in the hands of partners by the first quarter of 2008 and OEMs are expected to deliver updated systems by the second quarter.
"They [AMD's OEM partners] want to be a little more cautious, and we fully agree with them there, so they are looking for what we are calling the 'fixed' version, and we already have that going toward production right now that will not have this particular erratum," Steve Demski, AMD's Opteron product manager, said in an interview.
AMD is now admitting mistakes with Barcelona and Phenom. Click here to read more.
"We begin sampling in Q1 and we'll go into production with that in the later part of Q1, and then after about a month of partners working with that, we then expect to see products from them in Q2," Demski said.
Demski explained that Barcelona's design—four cores on the silicon die—and problems associated with that complexity did not allow the company to deliver proper yields.
"We weren't yielding the volumes that we wanted and that's why we had lower volumes than what we expected to have," Demski said. "By then not ramping up the learning curve as quickly as we wanted to, we also ran into this erratum that we discovered several weeks after we launched the part."
The main lesson to learn from this, Demski said, is to get future chips into the hands of OEMs faster to allow for more testing and debugging.
Page 2: AMD's Next Moves
While AMD executives are reassuring customers that the company is now on track, some analysts suspect the chip maker not only needs to continue its innovation in the New Year, but also to combine that with near-flawless execution in 2008 in order to win back the admiration of the industry.
"It's a very complex chip, more so than Intel's current quad-cores, and I do believe what they are saying and that they would have liked to get Barcelona out there a lot sooner," said John Spooner, an analyst with Technology Business Research. "In this case, the company seems to have been overly ambitious. That's part of the problem, since AMD has to take big risks to compete with Intel."
Since AMD now has more market share than in past years—IDC showed in the third quarter of 2007 that the company's worldwide market share stood at about 23 percent—its failures are more noticeable, and the company has to offer its partners and the IT industry more realistic road maps and timeframes for its products.
To read more about AMD's road map in 2008 and beyond, click here.
"AMD has to find a balance between being daring and moving ahead with some caution," Spooner added.
The problems with Barcelona were compounded by the presence of a design bug that AMD disclosed early in December. The bug itself was part of the chip's translation-lookaside buffer, which caused problems for data being transferred from the Level 2 to the Level 3 cache. (The L3 cache was new in Barcelona.) This caused the operating system to malfunction and shut the hardware down. AMD sent a BIOS fix and promised to correct the flaw when new chips are manufactured.
In 2008, Demski said, AMD will also make good on its promises to ramp the clock speed up to at least 2.5GHz. In addition, the company plans on moving ahead with its 45-nanometer version of the Opteron—dubbed "Shanghai"—that will go into production in the first half of 2008, with products hitting the street by late 2008.
Nathan Brookwood, an analyst with Insight64, said while it is not fair to judge AMD from its performance in the last eight months, especially compared with the success it had between 2001 and 2006 with Opteron, the company does need to offer more realistic road maps and to back its commitments to its technology.
The most difficult part will be balancing AMD's innovation with a more conservative approach to the industry.
"AMD built up a lot of credibility in the last five years, and it's sad to say that a lot of it drained away in the last eight months," Brookwood said. "The one way to refill it is to make realistic promises, and then AMD needs to meet those promises."
Editor's Note: John Spooner is a former senior staff writer for eWEEK.
Channel News and Analysis 
