At the Intel Developer Forum in September, Intel officials demonstrated its planned “Larrabee” graphics chip, running a scene from a 3D video game called Quake Wars on the processor.
Now Intel officials are shelving the project, saying the development of the processor hadn’t advanced far enough to offer it as a product. Larrabee was scheduled for release in 2010.
The decision to delay the development will hinder Intel’s plans to expand its architecture deeper into the videogame arena and to compete more directly in the area of parallel computing, where Nvidia and Advanced Micro Devices are pushing their graphics technologies.
Both AMD and Nvidia are aggressively pushing their products into more mainstream computing arenas, particularly in the HPC (high-performance computing) space. AMD officials see their ATI GPU business as a key differentiator in their competition with Intel.
AMD offers a development framework based on the OpenCL standard, which lets businesses applications run in both GPU and CPU environments. Nvidia is using its CUDA technology to drive its graphics technology into mainstream computing environments.
At the Supercomputing 2009 show in November, Nvidia officials unveiled new Tesla GPUs based on its new “Fermi” architecture, which officials said will offer the performance of traditional CPUs at a fraction of the cost and power.
Also at the Supercomputing show, Intel demonstrated an over-clocked Larrabee chip topping the 1-teraflop (trillion floating point calculations per second) mark.
However, even after that performance, Intel officials said they didn’t see a strong demand for graphics processors in mainstream computing. In an interview, Boyd Davis, general manager of Intel’s server platforms group marketing, said the company’s upcoming eight-core “Nehalem EX” processor-due for release in the first quarter of 2010-will be able to handle the bulk of the highly parallel workloads.
Demand for GPU-CPU co-processing is limited, Davis said.
Intel officials are now saying that Larrabee will be available as a platform for developers to design applications for parallel computing workloads.