Intel is in the process of developing multi-core processors that are capable of increasing floating-point performance into the spheres that are reached by computing clusters today. The projected performance of Intel's 80-core prototypes at the time it could become available is expected to hit one teraflop. At this point it's also interesting to note the floating point performance of a Pentium 4 processor: A 3 GHz model reaches 0.006 teraflops, which underlines the impressive capabilities of the 80-core prototype. We have to add that floating point performance is more important for scientific applications and number crunching than it is for desktop applications, but it might enable geologists to analyze drill test samples on their way home instead of processing it in expensive server farms.
This is simply crazy. Intel is again losing focus. Earlier they were just trying achieve higher and higher clock speed instead of reviewing and rebuilding their CPU architecture. AMD showed how you can make faster CPU with lower clock speeds. Intel realised this only after about 4 years. Now with the coming of multi cores, Intel is again trying to squeeze more and more cores into the die, and do something foolish like all cores communicating through the external FSB.
They should try something innovative.