I've noticed that when cvx is "compiling" the model to the solver it is not using all of my cores. The solver (Mosek) does use them, it's very noticeable as it spikes the cpu usage. Is there a way to make it use more cpu in that step?
I'm afraid not. Please understand that CVX is not designed for speed of solving; it is designed for the convenience of the user. And of course it's limited by what MATLAB can do.
To be fair to CVX. It is not common that multiple threads can used during model building. For instance AMPL i.e.
can not do it in parallel.