Nando Metzger Releases Metal-Gauss for Mac Splatting

Nando Metzger posted on X that Metal-Gauss trains 3D Gaussian splats on Mac in minutes using Metal kernels that compile at runtime rather than CUDA. He claimed after benchmarking several projects that it may be the fastest Mac-native Gaussian optimizer to date. The GitHub repository describes a fused forward-plus-backward rasterizer on PyTorch MPS with no CUDA anywhere. Metzger invited users to try the code and contribute. Media attached to Metzger's X post shows a rapid slideshow of photorealistic 3D Gaussian splat renderings.