I was reading in a recent ACM journal a few days ago, and I came across an article entitled In Search of a Natural Gesture.  The article explores human input/output methods and devices for computing systems, and I noticed some indirect comparisons to HPC.  This quote caught my attention:

The average consumer’s demand for more powerful technology has simply not kept up with the exponentially increasing supply. Some have referred to this stall in performance demand as the era of “good enough computing.” While “good enough” might suggest even further reduction in device cost, what’s happening instead is it’s becoming economically sensible to manufacture a wider variety of increasingly special purpose computers rather than expensive general purpose machines.

As I read this, I could not help but note that the HPC performance demand model is the inverse of that in the consumer market.  On one hand, we have consumer systems which have plenty of spare cycles and lots of human interface features.  On the other hand, we have HPC systems with very few spare cycles, and few bells and whistles. (I will concede here that high-end HPC visualization systems have bells and whistles… I’m mostly comparing back-end HPC systems used for solving).

There is nothing real revelatory or new here, but is an interesting thought exercise to view HPC systems as the opposite of consumer systems, in terms of performance demand.

As for a stall in HPC performance demand, it seems there are never enough cycles to meet the increasing demand for greater fidelity in our simulations and models.  If anything, the demand for more cycles in HPC has increased over time, and I don’t think “good enough” HPC computing is anywhere on the horizon.

Advertisements