SUNNYVALE, Calif. – When AMD went looking for a graphics partner in 2006, so the rumor in the Silicon Valley goes, it initially courted nVidia. Such a pairing made sense; nVidia was only a few miles away in Santa Clara, it had outstanding graphics and made chipsets and was a market leader.
But the deal never came together, for any one of several reasons. Whichever you choose to believe, AMD (NYSE: AMD) had to settle for what many observers believed was its second-best option, ATI Technologies. Not only was ATI across the continent, it was in another country, Canada.
Now, instead of a local partner, AMD had to pick up a third major office, in ATI’s native Toronto, on top of its Sunnyvale headquarters and Austin, Texas, where many executives are based. Plus there were all of the cross-border headaches.
Two years later that deal is paying off handsomely for AMD. Graphics are about $1 billion of AMD’s $6 billion in annual sales, and growing. It now has its first mobile platform to compete with Intel’s Centrino, it’s working on a CPU-GPU combination technology called Fusion, and ATI has come roaring to life, making significant gains on nVidia in the market thanks to a strong new product offering.
That success comes from basically adopting a multi processor strategy. For its most recent video cards, ATI decided not to follow its typical course of using one large, bleeding tech GPU. Instead, ATI put two GPUs, with perhaps less than bleeding tech performance, together. When running in parallel on the video card, the duo more than made up for any lags in individual performance.
“It was a big risk, because it was a departure from what worked for 10 years,” said Rick Bergman, senior vice president and general manager of the graphics products group.
It paid off. ATI’s business grew 55 percent from Q2 to Q3 of this year, and 40 percent over Q3 of 2007.
Next Page: Looking ahead to Stream computing
Page 2 of 3
Looking ahead to Stream computing
In 2009, Bergman sees “HD everywhere” and support for high definition video will be in most products coming from AMD. The company is moving from a 55nm design to 40nm design, and come December 10, it will release a new version of its video drivers that allow Stream computing.
Bergman was one of several AMD execs to speak here at financial analyst’s event where the company discussed its future plans.
Stream combines the ATI graphics with AMD CPU power to do hybrid computing not possible, or very difficult to do on either just a CPU or just a GPU. One of its uses is to convert media from one format to another, like video format conversion.
The December 10 driver release will unlock the ability to do Stream processing in many ATI Radeon cards that have the ability to run Stream but currently cannot.
“We’re really starting to see the benefits of the ATI acquisition come to life,” JoAnne Feeney, managing director and senior research analyst with FTN Midwest Securities told InternetNews.com. “The old AMD was only a CPU company. AMD could never play in the notebook space because it only offered the CPU.”
Vendors want the complete package of CPU, graphics and chipset from one vendor, not just because it makes buying easy, but because by controlling all three chips, the supplier can optimize them for maximum performance and power efficiency. Intel has had that for years with Centrino, now AMD has it with Puma, said Feeney.
Getting a grip on operations
Doug Grose, senior vice president of manufacturing and supply chain management and the future CEO for The Foundry Company, the new company created as a spin-off of AMD’s fabrication plants, followed Bergman to the stage to discuss 2009 priorities. That would be a transition to a best-in-class foundry model, complete conversion of the Dresden plants to 45nm production and begin working on 32nm.
Currently, AMD’s German foundry makes all of its processors, but the graphics processors and chipsets are farmed out to Taiwanese firms TSMC and UMC. That will change over the next 12 to 36 months as that business comes to The Foundry Company.
Grose has his work ahead of him with the fabs. The company has to start Fab 4X, the one planned for upstate New York, as well as get the German facilities up to volume manufacture of 45nm product. Mid-2009 is the target date to get Fab 36 completely converted to 45nm.
As for the move to 32nm, AMD will start this quarter with SRAM test chips. To get to 32nm, it will move to the high-k metal gate technology Intel (NASDAQ: INTC) has been using. AMD can get this technology from manufacturing partner IBM (NYSE: IBM). IBM developed its own high-k metal gate manufacturing process around the same time as Intel.
The company expects to move to 32nm by 2010, using high-k metal gate.
Next page: No change in guidance
Page 3 of 3
The next speaker was Bob Rivet, newly promoted to chief operations and administrative officer in addition to his duties as chief financial officer. He said that despite the huge cut in Intel’s projected revenues the day before, AMD won’t change its guidance because it’s too early.
Despite AMD’s seemingly endless bad luck, Rivet said there were bright spots. “If you stand back and look at things, we actually did make a lot of progress in 2008,” he told the crowd.
Rivet said that AMD is poised to reach $1.5 billion in operating income and profitability by mid-2009. The graphics segment has returned to profitability, it has disposed of its DTV business and plans to sell a handheld device graphics business as well.
Dirk Meyer, in his first analyst briefing as CEO, then took to the stage to finish off the morning. “We have a clear-eyed, achievable plan to make this a profitable company on a sustained basis,” he told them.
AMD is not focused on enterprise client business, but SMB and the power desktop because “that’s where all the money is,” he said.
His mood got a little dark when someone in the audience asked about Intel’s “serious questions” regarding The Foundry Company. “Our competition is doing their best to spread fear, uncertainty and doubt about it,” he said. “I’ll just say we know what our deal is. We are confident we are well within our boundary box for IP licensing.”