Intel's decision to spike its "Larrabee" discrete graphics chip will be a boon to Advanced Micro Devices and Nvidia, according to an analyst.
In a research note Dec. 7 on the Intel move, Hans Mosesmann, an analyst with Raymond James Equity Research, said the "Larrabee debacle" will cost the giant chip maker money and leverage as the industry moves forward into more general-purpose GPU computing.
"The strategic implications are significant as we view the world of -computing' as moving unambiguously to the use of more and more graphics transistors for general purpose," Mosesmann said in his report. "Basically, the world doesn't need a quad-core or eight-core CPU for mainstream applications, but it does appear eager to use of late a solid dual-core CPU and a good GPU as a -co-processor' (ask Apple)."
Intel's inability to bring out Larrabee will cost the company about $400 million to $800 million a year in 2010 and 2011, or about 10 to 20 percent of the GPU market that was available to Intel, he said. Much of that money will now go to AMD and Nvidia, he said.
It also significantly weakens Intel's position against AMD and Nvidia in this space, Mosesmann wrote. Going forward, Intel platforms will be more dependent on discrete graphics technology from AMD and Nvidia, and Intel's negotiating position against Nvidia in their ongoing legal dispute over cross-licensing of the bus interface is hindered given that Intel's future GPU offering will more closely resemble a traditional architecture, rather than the Larrabee's x86-based design.
"Intel now has an extremely weak -integrated' graphics roadmap," Mosesmann wrote. "With no relevant investment in its existing and rather ancient graphics core (GMA4500), over the next couple of years Intel platforms will be relatively weaker vs. AMD platforms or older Intel platforms that use Nvidia -ION' chipsets."
Intel officials on Dec. 4 said they were shelving Larrabee due to development issues, and instead will use it as a platform for programmers. The chip was first discussed publicly in 2007, and after several delays, was scheduled for release in early 2010.
Several analysts said that creating a new architecture is difficult, and that Intel's struggles shouldn't be unexpected. They also said they expect Intel will continue to develop its own graphics chip that will come out in the future.
The lessons learned through the Larrabee work will not only help Intel in this effort, but also as it looks to ramp up the number of cores in its traditional CPUs.
In the meantime, AMD-which entered the GPU business through its $5.4 billion acquisition of ATI in 2006-and Nvidia will get some running room as they continue to push their graphics technology for mainstream computing workloads.
While some Intel officials have downplayed the need for CPU-GPU co-processing capabilities, Nathan Brookwood, an analyst with Insight 64, said that demand in the HPC (high-performance computing) space is growing and will continue to do so.