The MATLAB 2017a identifies half number of core on Google Cloud virtual machine.

The MATLAB 2017a identifies half number of core on Google Cloud virtual machine. This could be a bug of MATLAB 2017a. When you start parallel pool of MATLAB 2017a with defaults set, one always gets the number of worker is equal to the half of cores. A virtual machine with eight cores starts with four workers, and sixteen cores with eight workers. When all these workers fully run, fifty percent CUP is used. Manually set the number of worker to the number of cores can fix this problem.

Answer 1

[Edric Ellis]

What happens when you run

>> feature('numcores')

on one of those machines? By default, MATLAB (and Parallel Computing Toolbox) considers only physical cores.

[Blue Bird]

I have down this! It was the half of the core of the virtual machine. By the way, we pay every core in Google cloud. They are definitely real CPUs and not Hyper-Threading.