一句话：GS-Scale通过将高斯参数卸载到主机内存并按需传输，在降低GPU显存的同时保持训练速度。
大规模高质量 3D Gaussian Splatting 训练需存储大量参数、梯度和优化器状态，易超出 GPU 显存。GS-Scale 把所有高斯保存在主机内存，仅按需将子集传到 GPU 做前向和反向传播；为缓解 CPU 端视锥剔除与优化器更新的瓶颈，它采用选择性卸载几何参数、参数转发流水化 CPU 优化器更新、延迟更新零梯度高斯三项优化。评估显示 GPU 显存降低 3.3-5.6 倍，训练速度接近无卸载方案；在 RTX 4070 Mobile 上高斯数可从 400 万扩展到 1800 万，LPIPS 改善 23-35%。
The advent of 3D Gaussian Splatting has revolutionized graphics rendering by delivering high visual quality and fast rendering speeds. However, training large-scale scenes at high quality remains challenging due to the substantial memory demands required to store parameters, gradients, and optimizer states, which can quickly overwhelm GPU memory. To address these limitations, we propose GS-Scale, a fast and memory-efficient training system for 3D Gaussian Splatting. GS-Scale stores all Gaussians in host memory, transferring only a subset to the GPU on demand for each forward and backward pass. While this dramatically reduces GPU memory usage, it requires frustum culling and optimizer updates to be executed on the CPU, introducing slowdowns due to CPU's limited compute and memory bandwidth. To mitigate this, GS-Scale employs three system-level optimizations: (1) selective offloading of geometric parameters for fast frustum culling, (2) parameter forwarding to pipeline CPU optimizer updates with GPU computation, and (3) deferred optimizer update to minimize unnecessary memory accesses for Gaussians with zero gradients. Our extensive evaluations on large-scale datasets demonstrate that GS-Scale significantly lowers GPU memory demands by 3.3-5.6x, while achieving training speeds comparable to GPU without host offloading. This enables large-scale 3D Gaussian Splatting training on consumer-grade GPUs; for instance, GS-Scale can scale the number of Gaussians from 4 million to 18 million on an RTX 4070 Mobile GPU, leading to 23-35% LPIPS (learned perceptual image patch similarity) improvement.
DOI 原文 · arXiv · PDF（开放获取） · DBLP
@inproceedings{abs-2509-15645,
title = {GS-Scale: Unlocking Large-Scale 3D Gaussian Splatting Training via Host Offloading},
author = {Donghyun Lee and Dawoon Jeong and Jae W. Lee and Hongil Yoon},
booktitle = {ASPLOS},
year = {2026},
doi = {10.1145/3779212.3790167}
}