updated 08:25 am EDT, Mon August 4, 2008
Intel Larrabee Details
Intel today provided some initial details of its Larrabee graphics architecture and thus its first attempt to directly challenge stand-alone graphics chipsets from AMD and NVIDIA. The semiconductor firm believes it has solved the problems of extending graphics features by using multiple, complete x86 processor cores that are more consciously designed to recognize multiple program threads and support Hyperthreading to run more than one thread at the same time; it also adds a dedicated vector processor per core to aid in graphics-related tasks.
This approach of including many generalized cores not only mimics the parallel computing inherent to modern graphics chips but theoretically gives software programmers a flexibility absent from normal graphics hardware. The foundation lets many x86 developers add new graphics effects or rendering techniques simply by creating new software, according to Intel.
The design's multi-core awareness is also said to make it uniquely scalable in a way that bottlenecks most multi-core CPUs and graphics chips. Larrabee typically avoids the diminishing returns of adding more cores and should become proportionately faster as Intel adds cores. The efficiency often scales up to within 10 percent, Intel says, though it doesn't yet have benchmarks to demonstrate the results.
Intel declines to elaborate on the likely number of cores or the clock speeds of given processors and notes that Larrabee won't be available until 2009 or 2010, when many of the company's main processors will themselves contain eight or more cores each.