Using parfor on cluster, changing number of cores per worker

I am using parfor on a computing cluster that has 128 workers.

If I use parfor will it utilize multiple cores per worker automatically?

What command can I use to ensure that multiple cores per worker are used?

Currently I am using the following code, but have not made any settings changes.

c = parcluster;
Job = c.batch(@RunParallelForLoop, 2, {}, 'Pool', 127)

to utilize 127 workers to run this function.

Is there any way I can set the number of cores per worker?

Thank you in advance.

JB

Answer 1

[Edric Ellis]

You can set the NumThreads property of your cluster object. For example:

c = parcluster
c.NumThreads = 2;
j = batch(c, @eval, 0, {'spmd, maxNumCompThreads, end'}, 'Pool', 3);
wait(j), diary(j)

demonstrates that each worker in this case is using 2 computational threads.

[James Barnes]

Thank you, very helpful!

[Rony Reuveni]

Hi

I read this topic and found that I get an error with:

c.NumThreads = 2

No public property NumThreads exists for class parallel.cluster.Local.

I am using MATLAB 2016a.

Am I doing something wrong?

Thanks,