[Beowulf] 90nm opteron power usage

Mark Hahn hahn at physics.mcmaster.ca
Sat Aug 13 11:31:25 PDT 2005


> Mark, in your experience how big is the difference between 
> one- and two-sided modules?

I'm just reading the specs.  given the same generation/density of chips,
a 2-sided module will have twice as many, and dissipate up to twice as 
much power.  chances are excellent that real dissipation will be lower
than this, because you probably won't drive both sides as fast as one side.

my experience is that quite a large number of HPC jobs do not saturate
the memory interface (some barely use it at all!).  certainly for a
montecarlo like simulation, I'd expect dimms to dissipate noticably less
than someone who's doing gigabyte daxpy or random accesses.




More information about the Beowulf mailing list