# dendenxu/fast-gaussian-rasterization

A geometry-shader-based, global CUDA sorted high-performance 3D Gaussian Splatting rasterizer. Can achieve a 5-10x speedup in rendering compared to the vanialla diff-gaussian-rasterization.

Repository: https://github.com/dendenxu/fast-gaussian-rasterization
Canonical: https://ross.abutalabs.com/products/fast-gaussian-rasterization
Language: Python
License: MIT
License Family: permissive
Topics: 3dgs, nerf, rasterization, shaders, 4dgs
Last push: 2025-02-26T01:14:43+00:00

## Health v2 (maintenance only)
Score: 19/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 8, release rhythm 8, longevity 62
- inputs: {"age_days": 876, "days_push": 554, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1202, forks 65 (observed 2026-08-28T04:03:58.365107+00:00)

## What it is
A drop-in replacement for diff-gaussian-rasterization that renders 3D Gaussian Splatting scenes using a geometry-shader-based GPU pipeline with global CUDA sorting. It achieves 5-10x faster rendering than the vanilla CUDA software rasterizer, though it does not yet support a backward pass for training.

## Use cases
- render 3d gaussian splatting scenes faster
- speed up 3dgs rendering in a viewer or gui
- replace diff_gaussian_rasterization with a faster drop-in
- real-time high-resolution gaussian splatting rendering
- render large gaussians at high pixel-to-point ratios
- offline rendering of gaussian splat scenes

## When to choose
- you need fast forward-only 3DGS rendering, e.g. in a real-time viewer
- your scenes have large Gaussians or high-resolution output
- you want a drop-in import swap with no CUDA compilation step
- you render in an OpenGL-based GUI and can write directly to the framebuffer

## When to avoid
- you need gradient/backward passes for training Gaussian Splatting models
- your point clouds are extremely dense (>1M points) with small Gaussians, where the CUDA implementation may be faster
- you rely on features like depth-peeling not supported here

## Facets
- artifact type: library
- maturity: active
- function: graphics, image-processing, machine-learning, gpu-computing
- domain: computer-vision, graphics, machine-learning, gpu-computing
- platform: python, windows, cross-platform
- tags: gaussian-splatting, 3dgs, rasterization, cuda, geometry-shader, opengl, nerf, rendering, drop-in-replacement, gpu, linux

## Member repositories
- dendenxu/fast-gaussian-rasterization (main) score 19

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:58.365107+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:20:35.305903+00:00, confidence not recorded.
  - readme: https://github.com/dendenxu/fast-gaussian-rasterization (fetched 2026-08-28T04:03:58.365107+00:00, sha b8193dd4a3b9)
- Data as of 2026-08-30T08:39:29.467469+00:00.