Matlabpool Maximum Number of Local Workers on one computer - Parallel Computing

Couple of questions... 1) Can someone from Mathworks clarify exactly how many local workers one can enable on one single machine? I've read 8 or 12 workers, but not sure which is correct since I've seen users complaining about not being able to access 4 of their 12 cores, but I've also seen 12 as the answer (see http://www.mathworks.com/help/toolbox/distcomp/matlabpool.html)

2) If one were to have, say, a single machine with 16 cores on it with a Matlab Distributed Computing license, can I simply say... 'matlabpool open 16' to activate all 16 local cores?

I'm in the market for a new machine and need to know the answers to decide which machine to get.

Thanks in advance.

Answer 1

[Edric Ellis]

The number of local workers available with Parallel Computing Toolbox and no MDCS has changed. When introduced, the limit was 4; this changed to 8 in R2009a; and to 12 in R2011b.

If you want to use 16 workers, you will need a 16-node MDCS licence, and you'll also need to set up some sort of scheduler to manage those. There are detailed instructions about how to do this here. Once you've done that, yes, you'll be able to do "matlabpool open 16".

[Anastasios]

Hello,

Lets say that my machine has 16 cores. If I open, lets say, 4 multiple parallel jobs with 4 cores each will the workload be distributed equally?

thanks

[Walter Roberson]

If memory serves me correctly, that would depend on how you had configured your MDCS (which would be required for 16 workers.)

Or are you referring to 4 different MATLAB sessions, each with matpool 4 ?

[Matthew Phillips]

I realize this is an older post but could you explain the rationale for this? This seems crazy. I have access to a 48-core server but can only get Matlab to use 12 cores, and even going through the process described, the number only increases to 16. But when I code in C/C++, I get as many worker threads as I ask for. I can understand that Matlab would want to prevent the user for asking for more workers than there are cores, since that would reduce efficiency. But why should Matlab otherwise limit how many workers I have? From a software dev perspective, once the architecture for n>1 threads is there, it's trivial to invoke more, it's just changing a single runtime parameter! Please explain Matlab's system a little more, right now I worried that charging more for more cores is just pure gouging.

[Rafael J.]

You can use as many cores as you want, it's just a licensing issue. The Parallel Processing Toolbox lets you use matlabpool locally and maxes out at 12 cores. If you want more than that, you need to setup a distributed computing server (even if you run it locally, it will be the case where the server and client are the same machine).

There's probably no technical reason why the parallel computing toolbox is limited to 12, it seems to just be a licensing/monetary situation.

Answer 2

[Edric Ellis]

Please note that in R2014a the limit on the number of local workers was removed. See the release notes.

[Abdullah]

Although this is again a very old post, Can I ask, that now, if someone is using Matlab r2014a or latter on a server, it will be able to use all cores (like 96 cores) without distributed server computer license? Thank you.

[Edric Ellis]

Yes, that's correct.

[Tyler Warner]

Hello, The release note link is broken now. Is there an updated reference for this post? Good day.

[Philip Borghesani]

The above release notes link was fixed to this link: Parallel Computing Toolbox Release Notes R2014a-R2017a