Intel's dumping of the first-generation "Larrabee" GPU has touched off a debate on whether the giant chip maker will buy rival Nvidia to get itself back into the graphics processor game.
Intel officials Dec. 4 announced that they were shelving its first GPGPU due to development issues. The chip, first discussed in 2007, had been delayed several times, but was due out in early 2010.
The chip, which had been demonstrated at the Intel Developer Forum in September and shown off at the Supercomputer 2009 show in November, was supposed to compete with products from Advanced Micro Devices and Nvidia.
In an entry Dec. 8, blogger Robert Cringely says that by dropping Larrabee, Intel is setting itself up to buy Nvidia. Unable to compete with AMD's upcoming Fusion CPU-GPU by developing its own product, Intel will look to Nvidia as a way into the graphics chip space and greater traction in the HPC (high-performance computing) field, according to Cringely.
""There is a funky dance going on right now between chip giants Intel and nVIDIA and I just want to cut through the crap and tell you that no matter what the companies are saying it is likely to end with nVIDIA being purchased by Intel. Both parties know it and the only thing that hasn't been determined yet is the price, which is what all this posturing is about.""
Intel started working on Larrabee soon after AMD bought GPU maker ATI for $5.4 billion in 2006, he said. Dropping Larrabee not only gets Intel out from under a difficult development project, but also removes a barrier to buying Nvidia. Federal regulators would have much greater antitrust concerns if Intel-with its own GPU-bought Nvidia.
"With Larrabee, Intel could be seen as crushing a major rival. Without Larrabee, Intel is just trying to enter a new market," Cringely wrote.
Intel also is attracted to Nvidia's Tegra AXP mobile chips, which he said are superior to Intel's "upcoming Moorestown post-Atom low-power processors."
"Intel had to do something the minute AMD bought ATi," Cringely wrote. "Now with Larrabee gone Intel has no real choice but to buy another company to remain in contention. The only such company available is nVIDIA."
Not everyone is buying the theory. In a Dec. 9 blog post, Jon Peddie of Jon Peddie Research called the idea of Intel buying Nvidia "na???ve speculation," saying that Intel doesn't feel it needs Nvidia.
""The company has all the graphics IP it needs from Imagination Technologies, plus its own labs. It's not that Intel couldn't build a GPU, but rather that the company doesn't see today's GPU architecture as having long legs-they don't think it will scale and it certainly can't do MIMD [Multiple Instruction stream, Multiple Data stream]. To Intel, it is a dead end and why invest in that? I need to say this again because it really is a critical difference in the basic philosophies of the two companies. None of the events of the past week have had anything to do with hardware design. Yes GPUs are hard to design. So are CPUs. So is any billion transistor part. Intel simply doesn't see a future for the conventional SIMD [Single Instruction, Multiple Data] GPU architecture. Right or wrong, that's where their analysis leads them, and you can huff and puff about it all you want, Intel is not going to change its mind on that matter.""
Larrabee isn't dead, Peddie said. Intel will continue to make investments in developing GPU technology based on the x86 architecture, a sentiment echoed by other analysts following Intel's announcement. He expects there to eventually be an entire family of Larrabee processors.
The bitter history between Intel and Nvidia also would make it difficult to envision the two companies merging.
"The cultural differences, acrimony, and belligerences between Intel and Nvidia run so deep it would be impossible to blend the organizations without a few homicides," Peddie wrote. "It's unlikely, regardless of how big Intel's checkbook is, that the two companies could ever agree on the price. The Nvidia [board of directors] and shareholders of Nvidia would never approve a friendly acquisition by Intel, and Nvidia has a multi-voting technique that would delay any hostile attempt for over a year."
Peddie also was harsh about the speculation surrounding such a deal.
"It's naive to evaluate the computer industry as through it was a chess board and say if White takes bishop then Black has to take queen. It just doesn't work that way, never has," he said. "The PC industry isn't sport. If you want to forecast the industry you better understand its working parts, the history of its people, and the technologies within it."