Why won't matlab use all logical cores?

I am running a highly vectorized code and have noticed that matlab will not use all of my logical cores. Many vector operations are computed on arrays which are on the order of 10^5-10^6 elements. I am running on an 6 physical, 12 logical core i7 4960x. I am using version 2013b.

When I run my code, windows CPU usage statistics indicate that only a maximum of 50% of CPU power is being used. It seems it is only using 6 cores even when forced to recognize 12. I have tried to set maxNumCompThreads=12 to force more core usage, but nothing changes.

It is trivial to show that this is not a fluke of hyperthreading inefficiency or anything of that sort. I can open up the same code in an additional matlab executable and the decrease in run speed is only 10% but with two codes running simultaneously, obviously demonstrating that the extra logical cores are simply not being used. Simultaneously, when hyperthreading is turned off, the increase in speed is a puny 10%. Thus hypertherthreading is a huge boost in my code. Moreover, my core temperatures are about 40C when one code is running, and 75C when two are. Obviously there is unused power here.

Regardless of the details, matlab is clearly failing to saturate 12 cores and thus my code is only running about 60% of the speed it would run at if all cores were used.

Is there a solution to this problem?

[Tai-Wu Chiang]

Virtual cores may modestly improve overall system performance, but they are likely to have little effect on the performance of MATLAB applications. Intel CPUs with hyper-threading give the appearance that a computer has twice as many cores than it actually has. When using a tool such as Windows Task Manager, MATLAB may appear to use only half of the CPU cores available on the computer, when in fact the "unused" half is actually the virtual cores created by hyper-threading.

[StefanoM64]

Finally, a smart observation about hyper-threading and Matlab. I think it's correct, the various managers give a misleading information about CPUs load ...

[Mehmet Selim Akay]

Dear Christopher, have you found a solution? In the previous years' Matlab versions I used to set the threads in the parallel preferences and it was working perfectly, but now it defaults to physical core count.

Answer 1

[ck4e]

I just ran into this and found a solution. You can set the number of worker threads (and so the number of "cores" used) using maxNumCompThreads.

See here: https://www.mathworks.com/help/matlab/ref/maxnumcompthreads.html

Answer 2

[per isakson]

• Matlab is single threaded. However, some functions use more than one core.
• Solution: Parallel Toolbox allows you to use "all" cores. (The Parallel Toolbox license sets an upper limit for the number of cores that may be used. The limit depend on the Matlab release.)

[Christopher]

The point is that matlab IS multithreading my vectorized code. I do have parallel toolbox but I'm not sure it is relevant here. Matlab is using half of my cores, but for some reason is failing to use the other half. The power is available, and it is using multitheaded techniques, so why won't it use the other half?

[per isakson]

Which release of Matlab? What does the Parallel Toolbox license say about number of "local workers"?

[Christopher]

I am currently using 2013b. In the Parallel computing toolbox preferences it says preferred number of workers is 12. Interestingly, I ran the 'cluster profile manager' and during the validation test CPU usage went up to almost 90% at one point.