Jiahao,
thanks for your comment, I restarted Matlab with the -singleCompThreadoption
and I got similar results (0.045 seconds on average)
Eric,
yes, I run it once to let Julia compile it, and btw, Matlab is also much
slower the first time you run it (~ 0.3 secs the first time and than ~
What do you get from versioninfo().
2014/1/18 Andrea Vigliotti andrea.viglio...@gmail.com
p.s.
I am running:
- MATLAB Version: 8.1.0.604 (R2013a)
- Julia Version 0.2.0 (2013-11-16 23:48 UTC)
on a DELL with Intel Core i7 and Kubuntu 13.10
cheers,
andrea
On Friday, 17 January 2014
*What do you get from versioninfo().*
Julia Version
0.2.0
Commit 05c6461 (2013-11-16 23:48
UTC)
I think we have the answer there. You are using system BLAS and LAPACK
which are likely to be much slower than OpenBLAS which is shipped with
julia.
2014/1/18 Andrea Vigliotti andrea.viglio...@gmail.com
*What do you get from versioninfo().*
Julia Version
0.2.0
Commit 05c6461 (2013-11-16
Andreas,
you were just right!
I downloaded and compiled the latest julia and now it's faster than Matlab
here's what I get now:
@time A\x;
elapsed time: 0.02576362 seconds (8016560 bytes allocated)
instead of 0.03 - 0.04 from Matlab
my versioninfo() now is
Julia Version
If you look at the bytes allocated, it seems that Andrea did that.
What version of Julia are you running, Andrea?
Kevin
On Fri, Jan 17, 2014 at 11:31 AM, Eric Davies iam...@gmail.com wrote:
Running once before will cause compilation of the called functions for the
given types.
julia A =
You can make sure to run MATLAB without parallelization, e.g. by
starting with the -singleCompThread option to turn off
parallelization. MATLAB's linear algebra library is multithreaded by
default.
http://www.mathworks.com/help/matlab/ref/maxnumcompthreads.html
Julia's linear algebra routines
Sorry; I have an older version of MATLAB (2010a) and saw the comparison on
my machine and jumped to conclusions. I guess if I had 2012+ I would see
MATLAB still being faster.
I'll pay attention to allocated bytes in the future.
On Friday, 17 January 2014 13:33:15 UTC-6, Kevin Squire wrote:
On my old MacBook with MATLAB 2013b, julia is the faster of the two for
this problem, but the exercise is mainly a comparison of OpenBLAS and MKL,
I think.
2014/1/17 Eric Davies iam...@gmail.com
Sorry; I have an older version of MATLAB (2010a) and saw the comparison on
my machine and jumped