[Beowulf] Definition of HPC

James Cuff james.cuff at cyclecomputing.com
Mon Apr 15 11:46:33 PDT 2013


> In response to Jeff Layton's article on using the cloud for HPC (link
> below), there's been a discussion on whether there should be a new term
> used in this case to replace 'HPC', such as 'research computing' which
> Jeff used in his articles. I say no. We just need to reconsider how we
> define HPC. For example, my definition of HPC is as follows, which I
> often use when giving introductory presentations on HPC.

Interestingly I caught up for coffee with Miron on Friday.  He and I
would argue "throughput"

http://www.hpcwire.com/hpcwire/2013-03-29/htc_big_data_and_god_particle.html

I'm with him, you and heck anyone that says we need to do *more*
computing. More is the important thing, and the more you do, the
better you need to do it.

I agree with all comers on this topic for sure!

The semantics put about by the "weirdo beardo's" back when it was cool
to have a monster expensive single system image running some sort of
proprietary pseudo unix have all but gone thankfully.  We live in a
new world now, and the "machines" folks are putting together look very
different.  Take TACC for example, Stampede is a wonderful example of
throughput, performance and vectorization though Phi all wrapped up
with a bow around it.  Compare that with what Vogels' team are doing,
and that with Zuckerberg's folks - it's all just computing at an
amazingly monster scale to me.

So here it is, a new name for what we do....

H{P,T}(R,A,T)C

Or maybe just "C" - it's all just *Computing* after all 'innit?  ;-)

j.



More information about the Beowulf mailing list