platform for number crunching---resume

[Nadav Har'El]

On Sat, Jun 19, 2010, Shimon Panfil wrote about "platform for number crunching---resume":
> b) nobody provides example of heavy *numerical* load without  
> overheating, kernel compilation for example is not relevant becouse  
> AFAIK compiler does not use floating point calculations and power  
> consumption and heating may be essentially different;

The key phrase in the above sentence is "may be". Do you have any reason
to suspect why floating-point calculation take significantly more energy
than integer calculations (and all the other things that go on, including
memory refresh, and what have you)? Have you actually measured such a
difference?

I didn't, so I'm only guessing here, but my guess is that there may be a
difference between the exact energy use of a tight FPU calculation loop and
a tight integer calculation (or kernel compilation, or whatever), but I doubt
it's a really significant different to the point of making a difference
between a perfectly functioning 8-CPU machine and the heat-death of a 2-CPU
machine...


-- 
Nadav Har'El                        |       Sunday, Jun 20 2010, 9 Tammuz 5770
nyh at math.technion.ac.il             |-----------------------------------------
Phone +972-523-790466, ICQ 13349191 |This message contains 100% recycled
http://nadav.harel.org.il           |characters.