help-octave
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: octave + mkl


From: Joshua Dillon
Subject: Re: octave + mkl
Date: Wed, 16 May 2012 13:28:00 -0700

It's no skin off my nose.  I just thought it would be helpful to
others who'd like to match Matlab's speed.

My *VERY* informal timing of MKL+octave shows it to be exactly the
same speed as Matlab for "standard" operations like eig and eigs.
Indeed, matlab uses MKL for these operations.

%Octave 3.7.0+ w/MKL:
>> X=rand(1000);
>> tic;eig(X);toc
Elapsed time is 0.967732 seconds.

%Octave 3.2.4 w/o MKL:
>> X=rand(1000);
>> tic;eig(X);toc
Elapsed time is 4.09906 seconds.

%Matlab r2011b:
>> X=rand(1000);
>> tic;eig(X);toc
Elapsed time is 0.986376 seconds.

I confirmed this over many runs and other operations but didn't save
the results.  As I said, I'm just trying to let others know how to do
this.

Cheers,

Josh


On Wed, May 16, 2012 at 1:16 PM, Sergei Steshenko <address@hidden> wrote:
>
>
>
>
> ----- Original Message -----
>> From: Jordi Gutiérrez Hermoso <address@hidden>
>> To: Joshua Dillon <address@hidden>
>> Cc: address@hidden
>> Sent: Wednesday, May 16, 2012 10:40 PM
>> Subject: Re: octave + mkl
>>
>> On 16 May 2012 15:12, Joshua Dillon <address@hidden> wrote:
>>>  I don't have time to disseminate this information, but it might be
>>>  nice to add a howto build octave with MKL support in the FAQ.
>>
>> We really don't want to be promoting non-free software in the Octave
>> mailing lists or wiki. What is MKL doing that we can't be doing
>> instead in ATLAS or OpenBLAS?
>>
>> - Jordi G. H.
>
> MKL probably works faster.
>
> If so, end users care much more about performance than about ideology.
>
> End users, as long as they don't distribute Octave, have full right to 
> combine free and non-free SW inside their version of Octave (of course, 
> provided non-free SW license doesn't prohibit using it with free SW).
>
> It's very good there are people who promote ways to build faster Octave - 
> despite your ideological objections.
>
> FWIW, 'ffmpeg' developers are much more tolerant - they explicitly allow in 
> 'configure' to choose non-free components, they just remind that such a 
> 'ffmpeg' built with non-free pieces can't be distributed.
>
> Regards,
>   Sergei.



-- 
Joshua V. Dillon, Ph.D.
Software Engineer, Google Inc.
650-404-7120 / address@hidden / http://www.almostsure.com
"If you work really hard and you're kind, amazing things will happen."
   -- Conan O'Brien


reply via email to

[Prev in Thread] Current Thread [Next in Thread]