The purpose of this repo is to compare different options for creating Python modules and their efficiency, relative to each-other.
- Add more benchmarks
- Nuitka --
nuitka3 --lto --module /path/to/lib.py - Cython --
cythonize -a -i /path/to/lib.pyx - Nim --
nim c --app:lib -d:release --gc:markAndSweep --out:lib.[so|pyd] /path/to/lib.nim - V --
v -shared -prod /path/to/lib.v
$ ./benchmark_all.sh
Running benchmark 'base64'
---------------------------------------------------------
Python : 0.1405s 1.00x
Nuitka : 0.1410s 1.00x
Cython : 0.1441s 0.97x
Nim : 0.0845s 1.66x
Running benchmark 'fibonacci_iterative' to 100000 places.
---------------------------------------------------------
Python : 1.1194s 1.00x
Nuitka : 1.1471s 0.98x
Cython : 0.0007s 1,559.35x
Numba : 0.0004s 2,980.38x
Nim : 0.0007s 1,499.98x
V : 0.0029s 383.67x
Running benchmark 'fibonacci_recursive' to 30 places.
---------------------------------------------------------
Python : 1.6359s 1.00x
Nuitka : 1.0064s 1.63x
Cython : 0.3227s 5.07x
Numba : 0.0582s 28.13x
Nim : 0.0182s 89.79x
V : 0.0585s 27.97x
Running benchmark 'json_loads'.
_NOTE:_ JSON parsing within Nim is quite fast. I believe
there is extra overhead from converting each value to a
string. There is probably also more efficient way to do
pass this data back to Python. Suggestions welcome.
---------------------------------------------------------
Python : 0.1633s 1.00x
Nuitka : 0.1583s 1.03x
Cython : 0.1573s 1.04x
Numba : 0.5923s 0.28x
Nim : 0.5128s 0.32x
Running benchmark 'gen_range'.
---------------------------------------------------------
Python : 0.4937s 1.00x
Nuitka : 0.4567s 1.08x
Cython : 0.1428s 3.46x
Numba : 0.3121s 1.58x
Nim : 0.3362s 1.47xBenchmarks were ran in a VM with the following specs.
OS: Antergos
Kernel: x86_64 Linux
Packages: 908
Shell: zsh 5.7.1
CPU: Intel Core i7-6700
GPU: vboxdrmfb
RAM: 1115MiB / 3000MiB
Suggestions for more benchmarks are welcome, as well as PRs.