Python 3.13 is due to be released in October, yet the first release candidate was published earlier in August. Last year, I did a quick CPU bound benchmark of version 3.12 using Mandelbrot set calculation.
With all files in place, I downloaded version 3.13RC from the official web site and tested 3 versions of Python on my M1 Mac Book Pro:
user@users-MacBook-Pro py_chat_ui % python --version
Python 3.11.9
user@users-MacBook-Pro py_chat_ui % python3 --version
Python 3.12.5
user@users-MacBook-Pro py_chat_ui % python3.13 --version
Python 3.13.0rc1
Here're the execution time (in seconds) for the implementation relying on NumPy (versions 3.11, 3.12 and 3.13RC1 correspondingly):
user@users-MacBook-Pro mandelbrot % python mandelbrot.py
1 Execution Time: 6.305059909820557 78513419
2 Execution Time: 6.282307863235474 78513419
3 Execution Time: 6.473501920700073 78513419
user@users-MacBook-Pro mandelbrot % python3 mandelbrot.py
1 Execution Time: 5.418003082275391 78513419
2 Execution Time: 5.412122011184692 78513419
3 Execution Time: 5.434246778488159 78513419
user@users-MacBook-Pro mandelbrot % python3.13 mandelbrot.py
1 Execution Time: 7.197513818740845 78513419
2 Execution Time: 7.212265968322754 78513419
3 Execution Time: 7.200297832489014 78513419
And the results for the pure Python implementation:
user@users-MacBook-Pro mandelbrot % python mandelbrot_pure.py
1 Execution Time: 41.18416976928711 78513425
2 Execution Time: 41.16466403007507 78513425
3 Execution Time: 41.148504972457886 78513425
4 Execution Time: 41.55486297607422 78513425
user@users-MacBook-Pro mandelbrot % python3 mandelbrot_pure.py
1 Execution Time: 49.806406021118164 78513425
2 Execution Time: 49.485753774642944 78513425
3 Execution Time: 49.52305006980896 78513425
4 Execution Time: 49.57118225097656 78513425
user@users-MacBook-Pro mandelbrot % python3.13 mandelbrot_pure.py
1 Execution Time: 41.07340693473816 78513425
2 Execution Time: 41.08624267578125 78513425
3 Execution Time: 41.09266400337219 78513425
4 Execution Time: 41.1431610584259 78513425
Frankly speaking, I am confused with the inconsistency in the results - newer doesn't mean better :)
P.S. >
For the reference, running Mandelbrot calculation with JiT compiled Dart program and AoT compiled C version:
user@users-MacBook-Pro mandelbrot % dart mandelbrot.dart
1 Execution Time: 0.476 78513425
2 Execution Time: 0.484 78513425
3 Execution Time: 0.475 78513425
user@users-MacBook-Pro mandelbrot % gcc -o mandelbrot mandelbrot.c -Ofast
user@users-MacBook-Pro mandelbrot % ./mandelbrot
1 Execution Time: 0.256706 79394433
2 Execution Time: 0.234396 79394433
3 Execution Time: 0.234862 79394433
P.P.S. > Here's the GitHub repo with the same benchmark implemented in different languages.
P.P.P.S. > Here's last year's post comparing Python, Numba and Mojo.
Top comments (1)
Modifying mandelbrot_0() to the following python aware code, cuts down running time to 6.8 seconds per run. decorating both functions with @numba.jit cuts it down to 0.26 seconds:
def mandelbrot_0(c: complex) -> int:
... z = 0j
... for i in range(MAX_ITERS):
....... z = z * z + c
....... if (z * z.conjugate()).real > 4.0:
........... return i
... return i
Runtime with python aware code decorated with @numba.jit:
1 Execution Time: 0.3756251335144043 78513425
2 Execution Time: 0.2627451419830322 78513425
3 Execution Time: 0.26718902587890625 78513425
4 Execution Time: 0.2661020755767822 78513425
Runtime with python aware code:
1 Execution Time: 6.8413779735565186 78513419
2 Execution Time: 6.812233924865723 78513419
3 Execution Time: 6.828969955444336 78513419
4 Execution Time: 6.858006000518799 78513419
Runtime of original code:
1 Execution Time: 41.62785005569458 78513425
2 Execution Time: 41.481120109558105 78513425
3 Execution Time: 41.68783092498779 78513425
4 Execution Time: 41.20909810066223 78513425