[bkanuka]

I agree. This was honestly news to me - who very often uses Python for maths. However, I would never write the code as he did (instead I would rely on numpy/scipy). So I would also be intersted in a numpy version of the same test.

[ReflectedImage]

How exactly would you write a gameboy emulator in numpy/scipy?

It's sequential code with fiddly side effects. I know I've written one.

But I'm generally curious if this is in-fact possible in someway.

[adgjlsfhk1]

Numpy would probably be even slower here. Numpy is good when you have large arrays, but it adds roughly .1 to 1 us per call in overhead.

[bkanuka]

Without validating anything myself, I was able to find this post https://hilpisch.com/Continuum_N_Body_Simulation_Numba_27072... which showed a simple n-body program sped up by ~670 times when moving from pure python to numpy+numba.

[adgjlsfhk1]

2 things to notice: the first is that this is with 5 bodies while your link was with 1000. For 1000 bodies, numpy is a noticeable speedup (100x). For 5 particles (I used the same code as your article but adjusted the number of particles) numpy is 5x slower. Adding numba would make this fast again since it would remove the overhead, but at that point, just use a fast language in the first place.