Matrix operations with Parallel Computing Toolbox

Hi all,

for a student project, I work with the "Parallel Computing Toolbox". To discover it's functionality I make simple matrix operations with distributed matrizes.

With an open matlapool (12 workers) I use (for example) the following code, on a computer with 16 cores:

    M1 = rand (i,i);
    M2 = rand (i,i); 

    tic
    M = M1 * M2;
    toc

    M1 = distributed(M1);
    M2 = distributed(M2);

    tic
    M = M1 * M2;
    toc

While execution of this code I run "ksysguard" to observe the cpu-usage. During the first multiplication (with "normal" matrizes), the cpu-usage of all 16 cores is at 100%. While the second multiplication runs, the usage of 12 cores is at 100% (I think this could be correct, cause of the 12 workers). I even notice that the first multiplication is much quicker (I tested it for values of n up to 20.000), and gets quicker with bigger matrizes.

Even when I close the matlab pool and do a matrix multiplication, all 16 cores have a usage of 100%. Does this mean that matlab does parallel computation at default?

I hope you understand, what I am wondering about. Many thanks in advance for your suggestions.

Answer 1

[Shashank Prasanna]

Most math operations are inherently multithreaded. There is a list you will find on our external page:

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

Jill will correct me if I am wrong, but each worker in the matlab pool is always single threaded. Which means the multithreaded operations will run single threaded on a worker but you have the advantage to scale to a large number of machines.

You may also find this link helpful which explains the difference.

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

Answer 2

[Jill Reese]

A number of MATLAB functions are multithreaded, and matrix multiplication (*) is one of them. This is what you are seeing during the first multiplication (with "normal" matrices).

[Sascha]

Thanks for your quick answer!

Is there a list of the yet multithreaded functions (without matlab pool using)? So that I can figure out in which functions I can get a benefit of the Parallel Computing Toolbox.

Answer 3

[Sascha]

Thank you both for your answers. You helped me very much!

[Shashank Prasanna]

Sascha, make sure you go ahead and accept an answer. This makes others who have questions similar to yours get their answer quickly by looking at the accepted answer.