Can I have 8 workers on a quad-core processor?

1 visualizzazione (ultimi 30 giorni)
I use PCT to create a "local cluster":
parallel.defaultClusterProfile('local');
c = parcluster();
and create and submit a job in the following way:
j=createJob(c);
createTask(j,...);
createTask(j,...);
...
submit(j);
This was done on a PC desktop with a i7 quad-core processor (Matlab2013b). I noticed that even after I created 20 tasks, after the job was submitted, only 4 workers were working simultaneously. Correct me if I'm wrong, I expect 8 workers working simultaneously. Can I have more than 4 workers working simultaneously?

Risposta accettata

Jill Reese
Jill Reese il 15 Ago 2014
When determining the default size of the local cluster, MATLAB uses the number of true cores you have (hyperthreading is ignored). You can modify the size of the local cluster yourself.
1. On the Home tab, open the Parallel menu.
2. Choose "Manage Cluster Profiles...".
3. Select the local Cluster Profile and click the Edit button. At that point you can set the NumWorkers field to whatever you like (8 or even more).
It is generally best practice to have the same number of workers as true cores on your machine. Opening more workers than cores doesn't usually improve performance.

Più risposte (1)

Matt J
Matt J il 15 Ago 2014
Maybe, if you have hyperthreading enabled.
  1 Commento
Degang
Degang il 15 Ago 2014
I guess hyperthreading has been enabled on my PC. The Ubuntu system monitor shows that there are 8 CPUs. Also, C++ program with openmp can use up to 78% CPU while a single Matlab worker can take only 12%.

Accedi per commentare.

Categorie

Scopri di più su Cluster Configuration in Help Center e File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by