[Maxima] performance in 64 bits machines.
fateman at cs.berkeley.edu
Sun Sep 23 08:54:16 CDT 2007
Sorry to keep on asking the same question, (which may NOT be the question of
the original poster, I guess).
Building a maxima on your 64-bit FreeBSD machine is not especially
interesting unless you have a lisp that is different from the one on the
32-bit machine. Several common lisps (I just found that Scieneer does this
too), have special versions that take advantage of 64 bits. Comparing those
to 32-bit lisps running maxima would be interesting. It is not clear what
you have done.
Since you did not specify which version of CLISP (32 or 64), or which
version of SBCL (32 or 64),. I am assuming you used the same (32 bit)
version as other people, and that running in a 64-bit mode was mostly
irrelevant because no 64-bit instructions were used!
So maybe you could clarify what you did?
By the way, the scieneer common lisp
Says that for 64 bits it only goes to 64Gbyte of RAM and uses conservative
GC; both disappointing. On the plus side it appears to allow for
extended-double floating point in memory.
> -----Original Message-----
> From: maxima-bounces at math.utexas.edu
> [mailto:maxima-bounces at math.utexas.edu] On Behalf Of P.U.Kruppa
> Sent: Saturday, September 22, 2007 10:21 PM
> To: Richard Fateman
> Cc: 'Vadim V. Zhytnikov'; maxima at math.utexas.edu; 'Miguel
> Marco'; 'Raymond Toy (RT/EUS)'
> Subject: Re: [Maxima] performance in 64 bits machines.
> On Sat, 22 Sep 2007, Richard Fateman wrote:
> > To repeat the point made by Stavros:
> > To compare Maxima properly one needs to have two different Lisp
> > implementations, one using 32 bits (for addresses) and one
> using 64 bits.
> > Perhaps there is a working 64-bit SBCL as well as a 32-bit
> one. This will
> > require 2 Maxima binaries.
> > Taking the same binary and running it with the OS in 32 or
> 64 bit mode is
> > not such an interesting test because a 32-bit lisp
> implementation will not
> > use any 64-bit operations, and so could hardly benefit.
> > RJF
> Hm, ... , as far as I understood the original poster's
> (Miguel Marco) question, he wanted to know if he has to expect
> "performance impacts" after switching from 32-bit to 64-bit
> Raymond Toy's idea was to find someone with a running 64-bit
> maxima and compare results to a 32-bit machine.
> I built two maxima's on my 64-bit FreeBSD (AMD 64, 3 GHz, 2 GB
> RAM), and get for
> clisp 205.4 s
> sbcl 54.3 s
> gcl is not available for FreeBSD amd64 .
> If now we had results for a comparable 32-bit computer, this
> wouldn't be a highly elaborated benchmark test, but we could
> answer Miguel Marcos question - couldn't we?
> Peter Ulrich Kruppa
> Maxima mailing list
> Maxima at math.utexas.edu
More information about the Maxima