← All resources

CuRast: Cuda-Based Software Rasterization for Billions of Triangles

Auto-discovered

Library / Codebase Advanced

Markus Schütz, Lukas Lipp, Elias Kristmann, Michael Wimmer

Visit resource ↗

Description

CuRast: Cuda-Based Software Rasterization for Billions of Triangles is an open-source CUDA library and demo by Markus Schütz and collaborators at TU Wien, hosted on GitHub with an accompanying Computer Graphics Forum 2026 paper. It rasterizes extremely dense opaque meshes without precomputed LODs or hardware acceleration structures.

The pipeline assigns one thread to tiny triangles, a warp cooperative path to medium triangles, and a third stage that splits oversized triangles, reporting large speedups over Vulkan baselines on hundreds of millions to billions of triangles in unique and instanced scenes. A demo application loads glTF and GLB assets for interactive inspection.

Advanced GPU programmers researching alternative rasterization paths for massive static geometry are the audience. CUDA and basic rasterization algorithm knowledge is required.

The repository and paper PDF are linked from the GitHub README under open licenses. Software rasterization research from the Vienna group is cited in massive-model visualization literature.