Sony, IBM, and Toshiba jointly created a chip they dubbed the Cell processor. Its initial target is the PlayStation3?
. In its initial incarnation, it combines 8 processor cores on a single die. Each core has a dedicated block of memory called local-store, and each core has DMA access to a common memory pool as well. Because common memory is accessed only through DMA, its not a true shared memory model. Multiple simultaneous inter-core data transfers are possible, which means that communication from process to process can occur without hogging the common memory bus bandwidth. This architecture is designed around streaming data between processes without going out to the slower memory bus (slower is still 20+ GB/s). With about 300GB/s total inter-core bandwidth available, that's clearly the place you want to keep your data. The Cell is capable of 200+ Gflops compared to the 4-5GFlops of Intels current architecture, but in order to make use of it all, you need a strong stream based archecture. So far, it seems like a good fit for FBP.
See also http://www.blachford.info/computer/Cell/Cell0_v2.html - thanks, Tyler!