Re: Mathematica 7.0 slow on OS X
- To: mathgroup at smc.vnet.net
- Subject: [mg93834] Re: Mathematica 7.0 slow on OS X
- From: Antti Penttilä@smc.vnet.net
- Date: Wed, 26 Nov 2008 07:23:04 -0500 (EST)
- Organization: University of Helsinki
- References: <ggj7j1$j8a$1@smc.vnet.net>
Bill Rowe wrote:
> On 11/25/08 at 7:18 AM, maderri2 at gmail.com (magma) wrote:
>
>> Does the Mathematica 7.0 own benchmarkreport take advantage of
>> parallel computing? There should be an improvement at least there.
>> Or not?
>
> Yes. That is looking at the code in Benchmark.m, there are
> points where the code is set to use parallel processing when
> possible. But this is also true for the code in Version 6. At
> the momement, I don't have a setup that allows me to see what
> effect this has.
With version 6 the parallel computing happened inside some numerical
matrix library functions (provided by Intel). In version 7 the parallel
computing can, in addition to that, use several Mathematica kernels in a
totally new way.
With Mathematica benchmarking in my platform, however, the use of
parallel subkernels with Math 7.0 seems to be very bad idea:
Needs[ "Benchmarking`"]
Kernels[]
{}
(* Only one kernel *)
Benchmark[]
..."System" -> "Microsoft Windows (64-bit)",
"BenchmarkName" -> "MathematicaMark7", "FullVersionNumber" -> "7.0.0",
"Date" -> "November 26, 2008", "BenchmarkResult" -> 1.699,
"TotalTime" -> 50.72...
LaunchKernels[]
KernelObject[1, "local"], KernelObject[2, "local"]}
(* two parallel subkernels *)
Benchmark[]
...{"MachineName" -> "2-node homogeneous cluster",
"System" -> "Windows-x86-64", "BenchmarkName" -> "MathematicaMark7",
"FullVersionNumber" -> "7.0.0", "Date" -> "November 26, 2008",
"BenchmarkResult" -> 1.97, "TotalTime" -> 262.494}
So, TotalTime went from 51 secs to 262 secs. Not very nice example of
parallel advantages.
Antti