In parallel processing, are multiple workers used for anything else than parfor loops?

When using the parallel computing toolbox, are parfor loops required for parallel processing? I plan to use parfor loops, but the most expensive part of my program is SVD, which is performed during each iteration of a loop. Will MATLAB break up the SVD into multiple workers?

Answer 1

[Shashank Prasanna]

Using SVD under a parfor will likely slow the performance of SVD. SVD is inherently multithreaded and make use of multiply cores calling it as it is. Here are a list of function that are inherently multithreaded:

http://www.mathworks.com/support/solutions/en/data/1-4PG4AN/

To comment further on how we could speed up your code, we may need to see more of what you have within the parfor loop.

[Clare]

Interesting... but you need the parallel processing toolbox to make SVD multithreaded? So if I call the toolbox (e.g. "matlabpool open 12"), then SVD will be multithreaded?

[Shashank Prasanna]

SVD is always multithreaded. This is a base MATLAB feature, you don't need parallel computing toolbox (PCT) at all. Using PCT however will make SVD on each worker single threaded and therefore not recommended.

[Clare]

Ok, so it is not at all possible to distribute SVD among several cores? Only to multithread it on one core? This is a bummer, because I have 16 cores to work with.

[Shashank Prasanna]

SVD is already multithreaded. It does make use of multiple cores.

Here is the difference between multicore (like svd) and multiprocessor (using PCT):

http://www.mathworks.com/discovery/matlab-multicore.html

[Clare]

EDIT: Never mind, I figured it out!

Ok, thanks for the response and link. I asked because I have experimented on my laptop, which has 4 cores and 8 threads. Yet a labor-intensive SVD computation hovers around 200% of CPU usage, suggesting it is using 2 threads in one core only. Any idea why this may be happening, and how to change it? Thanks again for your help.

p.s. On my laptop I have version r2011a. I tried changing the maximum # of threads to 8 using the "LASTN = maxNumCompThreads(8)" command. No luck.