performance in 64 bits machines.



I think you would find it difficult to match two computers that were in all
ways identical except that one was 64-bit and the other 32-bit. Maybe you
can run the same benchmark on one machine once using 32-bit and once using
64-bit versions of the software.


Speculation:

If you have more that 4 gigabytes of RAM and also use more than 4 gigabytes
of virtual memory for your job, it seems to me that the 64-bit addressing
could provide an advantage.

If you do lots of fixnum arithmetic of numbers whose absolute value is
larger than 2^(29) but smaller than 2^(61) or so, then 64-bit addressing
could be an advantage.

 There might be some disadvantages if all the data objects were twice as
large.

RJF

> -----Original Message-----
> From: maxima-bounces at math.utexas.edu 
> [mailto:maxima-bounces at math.utexas.edu] On Behalf Of Raymond 
> Toy (RT/EUS)
> Sent: Friday, September 21, 2007 8:21 AM
> To: P.U.Kruppa
> Cc: maxima at math.utexas.edu; Miguel Marco
> Subject: Re: [Maxima] performance in 64 bits machines.
> 
> P.U.Kruppa wrote:
> > Hello,
> > 
> > On Thu, 20 Sep 2007, Raymond Toy (RT/EUS) wrote:
> > 
> >>>>>>> "Miguel" == Miguel Marco <mmarco at unizar.es> writes:
> >>
> >>    Miguel> I guess that numerical routine should take 
> advance of the
> >>    Miguel> extra precission easily, but i have no clue about the
> >>    Miguel> internals of symbolic manipulation algorithms.
> > 
> >> All of these could be tested fairly easily by someone 
> running maxima
> >> on a 64-bit machine with a 64-bit Lisp. The timings for 
> the test suite
> >> would give some indication.
> >>
> >> Ray
> > 
> > I am running maxima-5.12.0 based on 64 bit CLISP 2.41 on an 
> AMD Athlon 
> > 64 (3 GHz, 2 GB RAM). My OS is FreeBSD-7.0 .
> > Problem is: what and how could we test, and what should it 
> be compared to?
> > Also I have to admit I don't do any "serious" computations. 
> I merely use 
> > maxima to prepare my students homeworks and examinations.
> > Though - tell me, if I can be of any help.
> 
> One option is to just run the test suite with a 64-bit Clisp 
> and and a 
> 32-bit Clisp and see how long each takes.
> 
> Ray
> _______________________________________________
> Maxima mailing list
> Maxima at math.utexas.edu
> http://www.math.utexas.edu/mailman/listinfo/maxima
>