Core 2 Duo and speed of computations

[Andrew Plewe]

I'm curious -- are there any implementations of ECM out there for GPUs? I'm
not familiar enough with the math (yet) to do something like that myself,
but I'd think the parallel processing abilities of a GPU would make it a
natural choice for algorithms which require lots of sieving.

	-Andrew Plewe-

-----Original Message-----
From: Ralf Stephan [mailto:ralf at ark.in-berlin.de]
Sent: Thursday, October 12, 2006 10:04 AM
To: seqfan at ext.jussieu.fr
Subject: Re: Core 2 Duo and speed of computations

An obvious winner with multicore processors is factorization
using elliptic curves, also sieving is done in parallel mostly.
This will also work with pari if different instances of pari
have different random numbers. As you can see, it means you
must start one program for each processor core you have. Using
ecm is better here, however, as you can tune its parameters to
take advantage of multiple cores.

Another winner is tasks where each number has to be tested for
a property, and this test is independent of results of other numbers,
and it takes time. Here, you just start two instances of the same
program, working on different ranges.


Regards,
ralf