Yesterday, NVIDIA unveiled its new graphic card architecture, entitled Fermi. More than a simple transition, it marks a true (r)evolution with previous systems and design, and those new models are definitely NOT ONLY dedicated to manage display. When looking at specification, it is impressive:
3 billion transistors engraved at 40 nm
384-bits memory bandwidth supporting up to 6 GB of GDDR5
512 CUDA cores organized in 16 clusters of 32 units, each of them featuring 64 KB of L1 cache
NVIDIA Parallel DataCache technology: 768 KB of L2 cache common to all 16 clusters to speed data sharing
32-bits floating point computing
NVIDIA GigaThread™ engine
ECC support
So, yes it looks more like a CPU/GPU than a GPU as we are used to see them from NVIDIA or AMD/ATI. If you add that this GPU will be able to execute C++ codes you will then understand that the true target is not AMD/ATI, but rather Intel and its
