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

Re: Multi-core Support in Mathematica 5.2

• To: mathgroup at smc.vnet.net
• Subject: [mg74651] Re: Multi-core Support in Mathematica 5.2
• From: Antti Penttilä@smc.vnet.net
• Date: Thu, 29 Mar 2007 05:32:24 -0500 (EST)
• Organization: University of Helsinki
• References: <esom9f$eok$1@smc.vnet.net> <eufprp$7c8$1@smc.vnet.net>

Hi,

It's not rubbish - it uses multi-core via some linear algebra packages 
(from Intel?) where multi-core is "easy" and safe to implement. You can 
check if your system is properly configured by running e.g. 
SingularValueDecomposition[] on a large random square matrix. If 
multi-core support is on, MathKernel will reach ~100% CPU-load. If not, 
check that the OMP_NUM_THREADS environment variable is set to 2.

General multi-core support is very hard to implement, because inside 
loops (Do,Table,etc.) the code can change the values of some variables 
that affect later iterations. Still, I suppose that we will see more 
parallelized code in the future, since the processors are going into 
multi-core systems (>2 cores per system).

Antti

Daniel wrote:
> So basically the "touted" multi-core support is a load of rubbish.
> 
> Daniel
> 
> dwadding at atl.lmco.com wrote:
>> Hi,
>>
>> Wolfram advertises version 5.2 as being multi-core capable.  When I
>> use the standard vectorized operations I can only get things to run on
>> one core.  Can some one tell me exactly which Mathematica functions
>> can actually exploit multi-core?
>>
>> Thanks,
>> Daniel
> 
>