It seems that these days, many tech sector giants have a Moby Dick to deal with -- a large, gleaming market that they just can't succeed in taking over, no matter how much effort they put into the task. For Microsoft
Intel tried to establish itself in the discrete market in the late 1990s, only to beat an ignominious retreat. And with the canceling of the first chips based on its Larrabee platform, history might be repeating itself.
Why Larrabee matters to Intel
On paper at least, Larrabee, with its combining of features traditionally found in microprocessors and graphics processors, sounds like a pretty compelling idea. There's been plenty of talk in recent years about using the growing capabilities of discrete graphics chips to handle computing tasks typically managed by PC and workstation microprocessors. NVIDIA
In the world of PCs, things may look a little more subdued, but that's quickly changing. NVIDIA's CUDA and AMD's Stream technologies allow applications to be written that can offload tasks such as video encoding onto a graphics processor. And the latest version of Microsoft's DirectCompute technology, which is supported by Windows 7, also enables a higher level of graphics processor offloading.
Hence Intel's support for Larrabee, a discrete graphics platform that could theoretically offer a lot more raw horsepower than NVIDIA's and AMD's graphics processors, by using the x86 architecture found in Intel's microprocessors. In Intel's eyes, success with Larrabee would not only make it a player in a lucrative market that's been off-limits, it would also cover the company's flank, should its users start relying heavily on their graphics cards for major computing tasks -- the kind of tasks that drive users to buy high-end Intel processors, rather than settling for the cheaper stuff.
For now, however, with the project proving harder than expected, Intel has decided to put its Larrabee plans on hold. The company is still talking about eventually offering a discrete graphics solution, but details are pretty vague. If Intel does ultimately cancel Larrabee altogether, it wouldn't be much for Intel investors to sweat over the short term: For all the hype, only a limited number of applications support technologies such as CUDA, Stream, and DirectCompute, and it's going to take some time to get the ball rolling.
But the long term could be a different story. And if it is, Intel will regret not breaking into the market that's been its white whale.