[Beowulf] CCL:Question regarding Mac G5 performance (fwd from mmccallum at pacific.edu)

Greg Lindahl lindahl at pathscale.com
Fri May 21 13:56:48 PDT 2004


On Fri, May 21, 2004 at 09:14:09AM -0400, Mark Hahn wrote:

> right.  I did a little meta-analysis of specFP recently, trying to get at
> the fact that some machines owe much of their score to a small number of 
> spec components.  I simply sorted the individual spec scores for a machine,
> and recomputed specFP omitting the top 1, top 2, etc.

This is not a valid approach: The scores are not normalized in a
fashion that makes their absolute values useful.

A more valid approach is to look at how many components of SPECfp each
system wins on. That doesn't depend on the absolute values of the
scores, only the relative values.

> 	- it2 is great if you can afford them [...]

What does this mean? People say it all the time, but it doesn't mean
anything.  If we're talking clusters, and systems built with processor
X cost 3 times as much as systems built with processor Y, you can
"afford" X unless you're spending so little you can't get a whole
X. Do people commonly build clusters out of 2 machines? I don't think
so...

Conversely, even if I have a lot of money, I am still going to spend
it to get the most bang for the buck. So it doesn't matter if I have a
lot of money, it's my problem and the performance of systems on my
problem that dictate what I buy.

Funny that I'm complaining about the same issue twice over: absolute
values vs. relative ones ;-)

-- greg



More information about the Beowulf mailing list