poisson

Poisson Disk Sampling Benchmark

Introduction

In this benchmark, we compare three Poisson Disk Sampling implementations among Taichi, Numpy, and Numba. The algorithm is suitable for single-threaded performance comparisons.

Evaluation

We conduct performance evaluation on the following device.

Device	Nvidia RTX 3080 (10GB)
FP32 performance	29700 GFLOPS
Memory bandwidth	760 GB/s
L2 cache capacity	5 MB
Driver version	470.57.02
CUDA version	11.4

Performance is measured in milliseconds (ms), we run over different number of samples. The reported times are measured using a 400 x 400 grid. The employed Taichi version is 1.0.0, the Numpy version is 1.21.5, the Numba version is 1.55.1, and the python version is 3.9.

Reproduction Steps

• Pre-requisites

python3 -m pip install numpy numba
python3 -m pip install taichi
python3 -m pip install matplotlib

• Run the benchmark and draw the plots

python3 plot_benchmark.py