Disruptive Trends by Data-Stream-Based Computing

Abstract

Computing is an extremely conservative science. After about a dozen of technology generations the dominant basic principles have mainly remained the same. Although von Neumann has not invented this mainstream form of computing, he and his co- authors have been in the mid’ 40ies the first to clearly describe the strikingly simple machine paradigm which I would like to call the von Neumann paradigm. Other paradigms, like dataflow machines, reduction engines, or others, have never been commercially significant. Commercial break-throughs mainly stem from the progress of semiconductor technology but have hardly affected the blinders limiting the scope of conference series like ISCA or MICRO, which remained almost 100% von- Neumann-based all the time. Dozens of expired supercomputing companies illustrate, how decades of searching for the universal massively parallel computer architecture finally failed. Von-Neumann-based parallel computing really succeeded only in some special application areas. Not only the area of embedded computing systems, but also PCs demonstrate, that more and more most silicon real estate is occupied by accelerators, prostheses needed by the aging von Neumann processor, now being a methusela after surviving so many technology generations.