[Beowulf] Supercomputers - iPad versus Cray

Ellis H. Wilson III ellis at cse.psu.edu
Thu Mar 8 07:57:44 PST 2012


On 03/08/2012 08:43 AM, Hearns, John wrote:
> http://www.theregister.co.uk/2012/03/08/supercomputing_vs_home_usage/
>
> A rather nice Register article on costs for supercomputers, adjusted to
> 2010 dollars,

Without being overly pessimistic, I get a vibe of "contrived" throughout 
the article.  For instance, as you mention, the supers are CPI adjusted 
for present day worth, which is quite interesting to see the trend of 
increasing cost for supers.  However, it just ends there -- our 
curiosity as to the much more important question, why that trend exists, 
goes unsatisfied.  I have some guesses about why, but it would be far 
more interesting to see hard evidence (or even squishy evidence, really 
any evidence) as to why than to simply speculate.  For instance, I'm 
sure there are figures on the older systems at least that break down the 
overall cost into costs of development versus hardware, or perhaps even 
more interesting, the respective costs of different pieces of hardware 
in the system (how much is spent on I/O versus CPU versus Net, etc). 
Seeing a rapid growth in development costs or that new systems are 
spending way more on the interconnect than the CPU or vice-versa would 
shed much needed light on an otherwise unexplained finding.

> And a rather interesting cost per megaflop table on the second page.

I actually hate this table.  If we want to compare flops/cost between 
machines, we should at least level the playing field: find Linpack 
numbers on individual machines in the listed clusters and compare them 
to your desktop or iPad.  After all, in the latter cases, we have no 
interconnect issues to worry about (just imagine Linpack numbers over 3G 
or even Wifi to other iPads...probably worse for 2 than for 1).  I think 
this table does a disservice to all the complexities that 
super-designers have to deal with, dumbing down the great hurdles they 
must cope with to get Linpack to scale across the entire cluster.

Just my 2c,

ellis



More information about the Beowulf mailing list