Re: Announcing Intel Fortran Compilers 9.0

Richard E Maine wrote:

In article <1119804438.088479.252910@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
 hchu@xxxxxxxxxxxxxx wrote:


I know perfectly well about parallel computng. I brought double
precision up only because it's natural for 64 bit computing.

This is a common misconception. There is little relationship between so-called 64-bit systems and 64-bit floating point. Terms like 64-bit systems have been largely used as marketing buzz words to the extent that they mean different things at different times. You need to ask precisely what the the term means in order to get any technical use out of it.

Intel processors starting with the first Pentium have a 64 bit data bus, so that 64 bit floating point data can be loaded or stored in one bus cycle.

But faster 64-bit floating-point arithmetic is *NOT* directly related to so-called "64-bit computing". If you have been lead to think otherwise, then you are headed for disappointment.

It is not so obvious what determines 64 bit floating point performance.
It seems to me that scientific computing is gaining a lot from the economy of scale of floating point processor built for people who don't really need them. That is also likely true for 64 bit addressing.

Still, I have been told that the AMD x86-64 processors have better floating point than comparable IA32 processors. It is also possible that 64 bit processors are slightly more efficient moving data to and from memory, which can also help 64 bit floating point.

With Athlon 64 below $200, it is hard to argue against 64 bit processors! (Fry's price for processor, fan, and motherboard
in recent weeks.)

-- glen

.

Relevant Pages

  • Re: Floating point in Bash scripts??
    ... Arbitrary precision is suitable for that. ... Normal floating point (IEEE ... The sort that would be suitable for computing 2^65536. ...
    (comp.os.linux.development.apps)
  • Re: Sine code for ANSI C
    ... >representation arose. ... >that there's a corresponding sine that is worth computing ... Floating point numbers *are* fuzzy. ... Dan Pop ...
    (comp.lang.c)
  • Re: Adding floating point numbers
    ... I use it when I'm feeling lazy, such as for computing the correct font height ... certainly see floating point being handy for all kinds of graphical ... precision is necessary. ...
    (microsoft.public.vc.mfc)
  • Re: Help! Im new to Fortran 77
    ... > First is representation of value in floating point. ... To elaborate just slightly (a full elaboration would take too long and ... technical computing - not just Fortran. ... writing a value requires computations and approximations. ...
    (comp.lang.fortran)
  • Re: Precision issue with Float and Double (C++)
    ... If you are having problems with "decimal precision", ... You can download my Floating Point Explorer from my MVP Tips site and prove that it is not ... One of the first things we teach students in computing is that you ... My first programming exercise in the course I took in numeric programming in the summer of ...
    (microsoft.public.vc.mfc)