matlab R2010b on windows server 2008 R2 Xeon 5130 using all cores without parallel specified code

Hi, I am running a program including a fminsearch and a fsolve.

The code is NOT written for parallel computing though. I am running the code on tow machines:

Machine A, Matlab R2007a on Windows server 2003 R2 @ Xeon E5410 with 4 cores and 8 threads.
Machine B, Matlab R2010b on Windows server 2008 R2 @ Xeon 5130 with 2 cores and 4 threads.

Wired thing is that on Machine A, one matlab process only use 1 thread, 1/8 of the total cpu usage. But on Machine B, one matlab process use all the 4 threads,100% of the cpu cores are used. But the performance is not improved at all.

Can somebody tell me how to solve the problem on Windows 2008?

Answer 1

[Walter Roberson]

Later MATLAB versions are better at automatically parallelizing some kinds of code. This is a built-in functionality that does not use the parallel toolbox.

[Yi]

hi, Walter, I was thinking about this possibility too. But the problem is that it use all the cpu but the performance does not improve at all. I also has another Matlab R2012b running the same code on my laptop i5-2410 2 core 4 threads, and it does not use up all the kernels.

[Walter Roberson]

There is a MATLAB command line option -singleCompThread that restricts to a single thread, in more modern versions. The earlier versions used http://www.mathworks.com/help/matlab/ref/maxnumcompthreads.html

[Yi]

Thanks a lot. Walter. It solves the problem.

Still wondering why the automatically parallel does not improve the performance.