The architecture preview for Fermi states that it supports C++, but to what does that mean? Could you take a program written in C++ for an i7 and run it on Fermi? How programmable does this make Fermi in comparison to a CPU by AMD or Intel?
My understanding is - there will be some library files for c++ for fermi. If your code included these libraries and factions. The coding will redirect the process to fermi. Another idea is - you can direct compile your existing source code without any modify to a fermi binary executable program.
Keep in mind though, you probably won't be able to just take any code in C++ and just put it there It would have to be C++ and x86
Fermi can't run x86 instructions natively....so, the "middle man" will still have a job to do. Trying to compare "Fermi" and Larrabee at this point, is completely pointless. Neither are on the market....nor are either expected for at least another month or more. Wait to compare them, until both are actually ready for market.
it should just be a matter of taking any C++ program that isint highly dependent on the underlying processor architecture, compile it using whatever C++ compiler Nvidia have/will have for Fermi and running it on the gpu. The problem is regular C++ code will most likely run like crap, the most efficient code will be highly parallel code. As for Larrabee, well have Intel ever done a good gpu? time will tell but its pointless speculating on it for the fact that its release is so far away, and most likely it will cost and absolute fortune.