Determine Number of Tasks and Cores to Use with FSOLVE

I am using fsolve on a large system of nonlinear equations. To speed it up, I would like to submit the job to a computing cluster and take advantage of the "UseParallel" option of fsolve. However, (since I am new to parallel computing) I am not sure how many tasks and CPUs per task to request. I initially requested 1 task with 48 cores per task, and found it did not run significantly faster than my 8 core laptop (with "UseParallel" enabled). I then requested 8 tasks with 6 cores per task and found it ran moderately faster.

How can I determine the highest number of cores that my script can efficiently use, as well as what combination of tasks/cores per task I should request?

Answer 1

[Walter Roberson]

You pretty much need trial and error to determine that.