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

Re: Parallel computations

• To: mathgroup at smc.vnet.net
• Subject: [mg82919] Re: Parallel computations
• From: Jens-Peer Kuska <kuska at informatik.uni-leipzig.de>
• Date: Sat, 3 Nov 2007 03:28:05 -0500 (EST)
• References: <fg4coq$5sf$1@smc.vnet.net> <fg6rc6$e3f$1@smc.vnet.net> <200710311124.GAA22911@smc.vnet.net> <fgemrn$ffv$1@smc.vnet.net>

Hi,

it is not clear *what* you wish to do with your Mathematica.
If you have a lot self programmed symbolic stuff you should
use the Personal Grid Edition and the Parallel Computing Toolkit.

If you have a lot of numeric stuff with matrix operations or
a lot of interactive or graphic applications the SMP support
of Mathematica would do be sufficient.

I'm sure, that with Mathematica, a shared cache is the very last
thing you should think about ... that is a question if you
have a C/C++ program that should run parallel and not if you have
a interpreter like Mathematica that pack every data into a buffered
(Math)Link and send it to an second and third kernel -- this is
all light years slower than a shared/not-shared cache.

BTW I have a large image processing toolkit that run parallel
on SMP machines and the most of it, is written in C++ and
the Parallel Computing Toolkit was never an option, because the
data exchange via a shared memory link was always to slow ...

Regards
   Jens

Stefan Porubsky wrote:
> Jens-Peer Kuska wrote:
>> Hi,
>>
>> since when a C/C++ compiler generate parallel code with out
>> any help and with out an extension like OpenMP or TBB ??
>>
>> Regards
>>    Jens
>>
>> linuxl4 at sohu.com wrote:
>>   
>>> if you are a fortran or C programmer, a multi-core processor will
>>> benefit you. so does mathematica.
>>>
>>>
>>>     
>>
>>   
> Hi all,
> 
> thanks for the answers. Unfortunately, I am neither  a Fortran nor a C 
> programmer. To be more precise, it is not clear to me, what is better 
> for Mathematica (Personal Grid) parallel computations, a multi-core 
> processor machine or one with more processors?  And what about the 
> question of  the relation to shared cache?
> 
> Thanks and regards,
> 
> Stefan
>