A Scalable PyTorch Abstraction for Multi-GPU Gaussian Splatting

2026-06-09

AI Summary

This work addresses the computational and memory bottlenecks that limit Gaussian splatting in large-scale, high-resolution neural reconstruction. The authors propose a multi-GPU Gaussian splatting framework that, for the first time, integrates CUDA Unified Memory with NVLink to enable transparent multi-GPU parallelism at the operator level. By abstracting multiple GPUs as a single PyTorch device, the framework automatically distributes Gaussian parameters and rendering operations without requiring any modifications to the model code. This approach substantially reduces the complexity of distributed implementation and successfully reconstructs city-scale street-level scenes containing over one billion Gaussians—surpassing the scale of current state-of-the-art methods by more than 25×.

Abstract

Gaussian splatting methods have become increasingly popular for neural reconstruction of the real world. However, they are often limited in scale and resolution due to compute and memory constraints. We present a multi-GPU Gaussian splatting approach that scales reconstruction to higher resolutions and larger scenes while abstracting away the code complexity typically associated with distributing a model. To accomplish this, we propose a PyTorch backend that distributes the Gaussian parameters and splatting operators across GPUs via CUDA unified memory and NVLink. Because distribution occurs at the operator level, the model code requires no explicit cross-device communication. More broadly, the backend exposes multiple GPUs as an aggregate PyTorch device and supports other PyTorch operators. We demonstrate city-scale reconstructions with street-level detail consisting of over 1 billion Gaussian splats, more than 25 times as many as the current state of the art.

Problem

• Gaussian splatting
• multi-GPU
• scalability
• neural reconstruction
• large-scale scenes

Innovation

• multi-GPU
• Gaussian splatting
• PyTorch abstraction
• CUDA unified memory
• scalable rendering