Let's discuss how these "Larrabee" chips are basically a bunch of Atom processors on a single die, then let's discuss how Larrabee measures up per-core/per-clock to Atom, and then to Nehalem... The Gflops per core/clock are a bit unrealistic for Larrabee, so basically, we will come to the conclusion that Intel is lying through their teeth on these performance claims, and that we can expect less... MUCH less in the real-world... not to mention that everything that makes GPGPU tech hard to implement in real-world code will pretty much still apply to Larrabee, there is still PCIe latency to deal with, and only those apps that truly lend themselves to parallelism are going to benefit.