HPC: The Onset of Massively Nano-scopic Parallelism

We are still in the early days of a new era, an era where parallelism is becoming the dominant means to accomplishing year-over-year increase in performance and price/performance.  In this talk, I will review the recent history and trends of dual-, quad-, multi-, and many-core processor adoption in the HPC market segment, both in the volume and high-end spaces.  Then I will move to a macro perspective of the HPC market segment, where, of particular interest is the fact that the Top500 list continues to record a track record where the performance of deployed supercomputers is outpacing Moore’s Law by no less than a factor of four and by as much as a factor of 15.  Moore ’s Law is looking healthy for the foreseeable future, but what about the “Unnamed Law” of #1 System performance going up by a factor of 1000 every 12-13 years?  As the promise of bringing about supercomputing elements of increasing performance continues, the challenge will be how to continue the rate of meeting performance requirements of an insatiable market segment at the current rate.  Words we have used before, such as Massive and Hyper are now too small, Micro and Nano are too big, and while Revitalization isn’t quantitative, its context is also requiring a fresh visit.  I will review a series of challenges that threaten the continued vibrancy of this market segment and outline a suggested course of action to mitigate the risks our HPC community is presently facing.