[Beowulf] Is there really a need for Exascale?

Lux, Jim (337C) james.p.lux at jpl.nasa.gov
Wed Nov 28 20:57:15 PST 2012



On 11/28/12 8:18 AM, "Alan Louis Scheinine" <alscheinine at tuffmail.us>
wrote:

>
>Prentice writes:
>> An even more cynical view say that the HPC vendors lobby the government
>> to believe exascale is important so the government invests in it and
>> subsidizes their R&D.
>
>Whether a few big exaflops computer or many teraflop computers,
>the computational needs exceed what is now available, yet more
>computational capability runs into the problem of extreme scale
>electrical power usage.  Every aspect of computers and interconnects
>needs drastic reductions in power usage, so government subsidies
>would be useful.  Calling this "an even more cynical view" seems
>a little harsh.
>
>Commodity mother boards are similar or equal to supercomputer
>hardware.  But I wonder what will drive further improvements in
>reducing power usage by several orders of magnitude.  I've heard
>the suggestion that computers in cell phones will be the mass
>market that leads to low-power hardware suitable for supercomputers.
>But the cell phone components do not cover the same range as
>supercomputer components.

But is that really true.   Sure, the processor in a cellphone is slower
than say a typical modern PC CPU.. But, given appropriate software, is it
a better $/FLOPS or W/FLOPS deal to get 100 cellphone CPUs or 1 superduper
PC CPU?


>
>What do others in the mailing list see as the trend?  Does the
>development of mass-market consumer products suffice for meeting
>the needs of the HPC community during this decade and the next?

At some point, light speed becomes the limiting factor, and for that,
reducing physical size is important.  Consumer gear is heading smaller, in
general (viz PC mobos getting smaller over the years), and certainly,
cellphones and their ilk are sort of at the limit of practical mass
production density.  Consumer markets have the advantage of enormous
volumes to spread the very high non-recurring-engineering cost over.

So, harkening back to the original Beowulf idea, you can get more FLOPS
per unit whatever (cubic meter/kW) for less capital outlay by leveraging
consumer development.  OTOH, if cost is no object (national pride, etc.)
then a case can be made that a custom design might do better.  But "custom
design" in this space is pretty expensive.. Say you wanted to design
something with 10 times the density of the densest consumer gear (for the
factor of 10 potential latency reduction due to shorter light time
delays).  How much is that worth?  10 million? 100 million?  (engineering
a new cellphone probably costs in the area of 10M, for context..  The
recent Apple/Samsung lawsuit showed how many prototypes they went
through.. That's not cheap)
>



More information about the Beowulf mailing list