Thoughts on Intel's Larrabee GPU

Intel (INTC) recently announced its plans to get into the discrete graphics business with its announcement of the Larrabee architecture to compete with NVDA and AMD. I don't think that Intel could compete with NVDA and AMD and performance because they lack the experience. NVDA's CEO lampooned Intel recently on this point, which seemed to backfire on the share price.

If Intel wanted to get into the graphics business, they would have done it by now, so I think their are other motivations for their entry of the discrete graphics business.

Using my limited technical knowledge, it looks like that Larrabee is based on the x86 instruction set and is able to do GPU computations.

This sounds familiar with NVDA's CUDA, which allows programmers to easily use the GPU to perform specialized calculations. Using the GPU, these calculations are orders of magnitude faster than a regular CPU. Instead of having a big array of CPU's, you can do the calculations on a GPU.

While the applications are CUDA are limited, it is a threat to Intel's share of the supercomputing business. Intel needed to find a solution to compete, so they proposed the Larrabee chip which used the x86 architecture making it easier to program for compared to CUDA.

Graphic performance is probably a secondary consideration for Intel, since their solution doesn't seem to be the most efficient design. Intel's main goal is to provide a solution that will be competitive with CUDA.

Whether it will be successful or not is another question.