Rough benchmark to show Python multiprocessing overhead from inter-process communication

process_bench.py

import time
import threading
import multiprocessing
from numba import jit

import numpy as np
thing1 = np.empty((4000, 4000))
thing2 = np.empty((4000, 4000))


@jit(nogil=True)
def bench(thing1, thing2):
    return np.sum((thing1 - thing2)**2)

bench(thing1, thing2)

print('Threading')
for n_threads in range(1, 9):
    threads = [threading.Thread(target=bench, args=(thing1, thing2)) for _ in range(n_threads)]
    start = time.time()
    for thread in threads:
        thread.start()
    for thread in threads:
        thread.join()
    end = time.time()
    print(n_threads, end - start)


@jit
def bench_gil(thing1, thing2):
    return np.sum((thing1 - thing2)**2)


bench(thing1, thing2)

print('\nMultiprocessing:')
for n_threads in range(1, 9):
    threads = [multiprocessing.Process(target=bench_gil, args=(thing1, thing2)) for _ in range(n_threads)]
    start = time.time()
    for thread in threads:
        thread.start()
    for thread in threads:
        thread.join()
    end = time.time()
    print(n_threads, end - start)