Von-Neumann-centric Computing: unaffordable soon ?

KEYNOTE SPEAKER: Reiner Hartenstein, Professor, TU Kaiserslautern

With respect to the future affordability of our entire computing ecosystem, Reiner Hartenstein will address 3 different key issues. Faster than predicted years ago, we see the impact of climate change. A key factor seems to be carbon emission, primarily from power plants, so that we should reduce the number of plants required.

Another important issue is the high cost of energy. A study [Mark. P. Mills] estimates that almost 30% of all electricity consumed in the US goes into all kinds of computers, visible or embedded, and might go up to 50% some years later. The upward trend of the crude oil price (having reached 164 dollars by July 2008) reminds us, that the operating cost of
our entire computing ecosystem might become unaffordable within a few years: another reason for trying to reduce the number of power plants needed.

Again another important issue is the strategic change of the microprocessor industry by stopping the GHz race, kicking off the many-core programming crisis which threatens to stall further progress to affordable higher performance computing: „from growth industry to replacement industry“, [Dave Patterson]: „Methods for supporting many-core could reset microprocessor hardware and software roadmaps for the next 30 years.”

A fully von-Neumann-based approach can never be the solution to cope with the many-core programming crisis, nor with the growing energy cost. Low Power IC design techniques will play a role in this situation. However, methods to cope with the von Neumann syndrome by software to configware migration promise a drastically better reduction in power consumption and promise speed-ups to drastically higher performance – both promises
by up to several orders of magnitude.

Because of the enormous burden by legacy software, we cannot fully switch quickly to a disruptive paradigm shift. We need a twin paradigm approach, based on computing wisdom mostly having been ignored by our curricula for
decades - except by the scenes of Reconfigurable Computing offering the highly promising twin brother of the von Neumann paradigm.

The talk illustrates, why this approach has a drastically higher potential to solve the problems mentioned above, and proposes related models to cope with our CS education dilemma. This whistle blowing talk calls for a run-away computing revolution causing a quick and strong educational impact as known from the VLSI design revolution a la Mead & Conway. We all must stick together to avoid a disaster of historic dimensions.

Reiner Hartenstein received all his academic degrees from the Faculty of Electrical Engineering at the Karlsruhe Institute of Technology, Germany, his diploma in 1959 and his Dr.-Ing. degree in 1969.

He is professor at TU Kaiserslautern and was visiting professor at UC Berkeley. He is consultant and authorized expert and referee on  on Reconfigurable high Performance Computing. In research he is credited to be the father of HPC Reconfigurable Computing.

Hartenstein is pioneer of structured VLSI design and author of the hardware language KARL, the trailblazer of VHDL and Verilog. He is founder of three, and co-founder of two more successful international conference series, as well as founder of the Multi University VLSI Design „E.I.S. Projekt“, the German contribution to the Mead-&-Conway VLSI design revolution.

He is IEEE life fellow, SDPS fellow, FPL fellow, and recipient of several other awards He has published 14 books and more than 400 technical papers and has given numerous talks, including more than 200 invited talks and more than 20 keynote addresses.