While for this particular case Yeppp does improve the performance (at the cost of almost a gig of memory usage), it was a simplified minimal example. The real target function is much more complicated, uses complex exponentials, and is not easily vectorizable. It sounds like Julia should be well-suited for such applications, but as I see here it performs significantly slower than Python. As I understand, Julia uses a custom implementation of trig functions instead of optimized libraries, can it be the reason?