Boost Python Performance 100× with Taichi: Real‑World Speedup Examples

It is well known that Python’s simplicity and readability come at the cost of performance, especially in compute‑intensive cases such as nested for loops. According to Hu Yuanming, importing the Taichi library can increase execution speed by up to 100×.

Just import a library called "Taichi" and you can boost code speed 100 times !

Prime counting speed ×120

The first example counts all prime numbers less than a given positive integer N. The standard Python implementation takes 2.235 seconds for N = 1,000,000:

After importing Taichi and adding two decorators (no changes to the function body), the same task finishes in 0.363 seconds, a nearly 6× speedup. For N = 10,000,000 the runtime drops from 55 seconds to 0.8 seconds, a 70× improvement. Enabling GPU execution with ti.init(arch=ti.gpu) yields a 120× speedup over the original Python code.

Dynamic programming (LCS) speed ×500

Using the classic longest common subsequence (LCS) problem from "Introduction to Algorithms", the Taichi implementation solves the problem in 0.9 seconds, whereas a NumPy version requires 476 seconds—over 500× slower.

Reaction‑diffusion equation: impressive results

The reaction‑diffusion model simulates pattern formation using two chemicals U and V. The governing equations involve diffusion rates Du, Dv, feed rate f, and kill rate k. Taichi implements the model on a grid, updating concentrations with a kernel that accesses neighboring cells via a Laplacian operator.

The kernel code (shown below) runs in fewer than ten lines and can be executed on the GPU, achieving over 300 fps, far surpassing a Numba implementation that caps at around 30 fps.

@ti.kernel
def compute(phase: int):
    for i, j in ti.ndrange(W, H):
        cen = uv[phase, i, j]
        lapl = uv[phase, i+1, j] + uv[phase, i, j+1] + uv[phase, i-1, j] + uv[phase, i, j-1] - 4.0 * cen
        du = Du * lapl[0] - cen[0] * cen[1] * cen[1] + feed * (1 - cen[0])
        dv = Dv * lapl[1] + cen[0] * cen[1] * cen[1] - (feed + kill) * cen[1]
        val = cen + 0.5 * tm.vec2(du, dv)
        uv[1 - phase, i, j] = val

Installation

Taichi can be installed with a single command: pip install taichi Taichi is a domain‑specific language embedded in Python that compiles @ti.kernel -decorated functions to run on CPUs or GPUs, providing high‑performance computing without the need to write C++/CUDA code. It interoperates with popular libraries such as NumPy, Matplotlib, and PyTorch.

For a detailed comparison of Taichi with other acceleration methods, see the original documentation.