processor use only 13% when excuting function/script

Hello,

I am calculating spread pattern of quite a large amount of projectiles. When I run the full model it take more than 1600 sec cpu time ( profiler info) This is actually quite long since I need to test a lot different setting.

I am using R2012a win 64 version and I have a 8GB ram with Intel® Core™i7QM CPU @2.2Ghz with window 7 professional.(HP Elitebook 8560w)

Now I don't know much about 'using different cores of the CPU' but I can see that there are 8 cores and the total CPU use does not exceed 12-13%, since 100%/8 is 13% it seems like only 1 core is used? Is that correct and if yes, can I do something to make all cores working? or any other action? It seems lile only 1/8 of my computer is used

thanks for any help regards,Jürgen

Answer 1

[Walter Roberson]

Some call patterns, when operating on "large enough" data (size can depend on the exact pattern), will call LAPACK or BLAS to do the work. Both of those libraries are highly optimized and multithreaded even without the Parallel Processing Toolbox.

Call patterns that do not happen to match the ones handled by the libraries, or if the data is not "big enough", are handled single processor unless the Parallel Processing Toolbox is used (and parallel processing is requested.)

The Parallel Processing Toolbox allows explicit requests for parallelization; code (at some level!) has to be written specially to use it. There are several facilities in the PCT, to allow the code to be broken up in different ways. But MATLAB will not just "figure out for itself" which parallel processing to use: it has to be coded in. Sometimes the changes might be quite small, but sometimes the code changes might be extensive.

[Jürgen]

Thanks for the answer, So actually there is very little I can do: like I unbderstood LAPACK and BLAS are libraries integrated in matlab and are not user controlled?

so unless I purchase the PCT I can not fully use the capapcities of my pc? rgardsJ

[Titus Edelhofer]

Hi Jürgen,

yes, generally speaking that's right. In your case ("since I need to test a lot different setting") the PCT would help you speed up the whole simulation both easily and significantly.

Titus

[Walter Roberson]

It is possible to explicitly call those two libraries. It is, I think, also possible to disable calling those two libraries.

Whether you will be able to use the full capacities of your computer by using PCT will depend a lot on how your code is written and what it does. Plain CPU-bound calculations can often be improved, but that depends on how much the calculations must interact. For example if the calculations for the projectiles involved details of the turbulent air flow around each projectile for projectiles flying near each other, then there could be difficulty in improving performance.

[Jürgen]

there you have it : yes we insert drag influences and spin, at the moment just top spin and back spin but later also side spin effect and thus curved trajectories,

[Walter Roberson]

Does the flight of one projectile influence the flight of another? The less they interact, the more easily you can divide the work evenly amongst your processors.

If the flights do interact, are you using something equivalent to Finite Element Mesh (FEM) to model the air flows?

[Jürgen]

Hi, until now the flights are actually not interacting, for each object or grain we calculate the trajectory and air influence seperately so I imagine that e.g. 1000 grains each core could calculate half. If you want to make it very complex you can use techniques like Computational Fluid Dynamics( CFD) but that is a next step and not necessary for the moment

[Walter Roberson]

In such a case, you should be able to use parfor() or distributed arrays, I suspect.

[Jürgen]

okay thanks, I'll take a look at the price :-) but anyway for the coming period all except that I will have to wait while the programm is executing, one way I tried is to open a second time matlab and write new code and testing while the other application is do his thing.

Answer 2

[per isakson]

Basically, Matlab uses one core, which explains 12-13%. (Single threaded.)

Some Matlab functions use several cores.

Parallel Toolbox is needed to use all 8 cores.

[Jürgen]

Hi, so that means that I am better of with 4 core processor, because then I will be able to use at least 25% of my capacity?? That's really strange thanks for your answer regards,Jürgen

[Friedrich]

No it isn’t. It’s the same.

Consider I am offer you:

a.) 10% of 100 dollar

b.) 25% of 40 dollar

Which one would you choose? It doesn’t matter because it’s the same(10 dollar in both cases)

It doesn’t matter if you use 25% of 4 Cores (which is basically one core) or if you use 12-13% of 8 cores (which is also one core)

If one would follow your logic, the best would be using 1 Core CPU because then you use 100% of your capacity. But thats not what you want.

[Jürgen]

interesting, okay that way you are right but I saw it different: I know have a quite new laptop with a good processor, let's say a 100 dollar processor to stay in your example, but I can use only 12.5% so 12.5 dollar is used, but my older computer with a 50 dollar processor and only dual core will use 50% or 25 dollar, I mean, habing a super processor that has only 1/8 use, can by less effective than a poor pr0cessor used on 50%, no??