[Date Index] [Thread Index] [Author Index]

Re: Mathematica 5.2: The 64-bit and multicore release

• To: mathgroup at smc.vnet.net
• Subject: [mg59045] Re: Mathematica 5.2: The 64-bit and multicore release
• From: "Nasser Abbasi" <nma at 12000.org>
• Date: Wed, 27 Jul 2005 01:26:00 -0400 (EDT)
• References: <dbnigj$hil$1@smc.vnet.net><dboum4$svm$1@smc.vnet.net> <dc4sgq$11r$1@smc.vnet.net>
• Sender: owner-wri-mathgroup at wolfram.com

 

"Vince Virgilio" <blueschi at gmail.com> wrote in message 
news:dc4sgq$11r$1 at smc.vnet.net...
>
> Hello David,
>
> I think Mathematica uses the ATLAS (Automatically Tuned Linear 
> Algebra
> System) libraries for a significant amount of its linear algebra. 
> And
> I think ATLAS has a compile-time option to enabled threading.  This
> might be what Wolfram Inc. means by "multicore" support.
>
> Corrections welcome.
>
> Vince Virgilio
>

But certainly one can have threaded libraries on a single core system. 
This has been the case for many years now. One can get some advantages 
by running threaded apps even on a single CPU, depending on the nature 
of the app. The only difference with multi-core is that one would get 
true concurrent executions of threads, which should results in much 
better theoretical performance advantages, everything else being 
equal. While on single core cpu the threads of course will have to 
take turns running on the single cpu.

i.e. on a single core, concurrent executions of threads is 
illusionary, but still can be more efficient than otherwise, while on 
multiple-core it is real.

The point is that, if there is such a flag on these ATLAS libraries, 
it could been turned on without the multiple-core being there. After 
all, one can install and run Mathematica 5.2 perbuild binaries on 
single core systems as is.

Nasser