FastSimplex

GPU simplex noise for Python. Provides significant performance improvements over CPU-based implementations, especially for 3D noise generation.

Uses custom Metal/CUDA kernels for extremely fast generation on most devices.

Installation

pip install git+https://github.com/Greg-Tarr/fastsimplex.git

Usage

import torch
from fastsimplex import noise2, noise3

# 2D noise
x = torch.linspace(-1, 1, 256)
y = torch.linspace(-1, 1, 256)
X, Y = torch.meshgrid(x, y, indexing="ij")
noise_2d = noise2(X, Y, octaves=4, persistence=0.5, lacunarity=2.0)

# 3D noise
z = torch.linspace(-1, 1, 256)
X, Y, Z = torch.meshgrid(x, y, z, indexing="ij")
noise_3d = noise3(X, Y, Z, octaves=4, persistence=0.5, lacunarity=2.0)

Performance

See scripts/benchmark.py for detailed benchmarks. Generally achieves:

• 4-5x speedup for 2D noise generation
• 80-100x speedup for 3D noise generation

Gallery

2D Noise Examples

CUDA Backend

Scale 25.0:

Scale 50.0:

MPS Backend (Apple Silicon)

Scale 25.0:

Scale 50.0:

3D Noise Examples

CUDA Backend

Scale 25.0:

Scale 50.0:

MPS Backend (Apple Silicon)

Scale 25.0:

Scale 50.0:

Requirements

• Python 3.12 or later
• PyTorch 2.6.0 or later
• CUDA-capable GPU (for CUDA backend) or Apple Silicon (for MPS backend)