Fermi, GT300 looks much more like a CPU than any GPU before. Support for C++ and ECC memory are definitely something that high compute industry can leverage and GPU with its hundreds of cores can do a very fast search.

The first step for Nvidia is that its current generation Tesla is inside of some super computers, including Bloomberg's, which definitely helps to open the doors of highly profitable HPC market to Nvidia’s GPU.

The downsides of GPU based high performance computing are that they are not as versatile as CPU and of course that they do not run X86 code. Still, if you want the fastest data base search, a GPU cluster will beat any multi-core CPU-based high performance computing server farm. ECC support is also something that these guys needed and since GT300, Fermi offers this, as well as L1 and L2 cache, the fight between GPU and CPU in high performance computing market can get to the next level.

With all this in mind, Larrabee makes a perfect high performance computing backbone, and is one of the main reasons why Intel went down this road. GPU simply puts CPU to shame on any parallel process as Intel’s 2010 processor has six to eight very deep and smart cores, while Nvidia has 512 shallow but fast cores that can do the job.