DeGS: A Scalable 3DGS Architecture via Decoupled Workload Parsing and Reorganization

2026-08-03

AI Summary

This work addresses the low processing element (PE) utilization and poor scalability of existing 3D Gaussian Splatting (3DGS) accelerators, which stem from a tightly coupled “sort-blend” dataflow. To overcome these limitations, the authors propose the first decoupled and scalable architecture tailored for 3DGS, restructuring the rendering pipeline into three stages: workload parsing, reorganization, and parallel blending. This approach eliminates spatial and temporal redundancies, transforming fragmented, temporally dependent operations into compact, dense computations. Implemented in 28 nm technology, the design achieves 2.36–7.25× higher throughput, 1.82–6.02× end-to-end speedup, and 1.59–4.42× better energy efficiency compared to state-of-the-art accelerators such as GSCore, GBU, and GCC across resolutions from 720p to 8K. Moreover, PE utilization remains above 80% even at high resolutions when scaling from 16 to 1024 PEs.

Abstract

3D Gaussian Splatting (3DGS) has emerged as a leading technique for real-time novel view synthesis, yet existing 3DGS accelerators suffer from poor architectural scalability: increasing the number of PEs leads to marginal performance improvement during rendering. We identify that the root cause is the tightly coupled ``checking-while-blending'' dataflow, which exacerbates PE underutilization caused by spatial redundancy from irregular Gaussian coverage and temporal redundancy from asynchronous pixel-wise termination under parallel execution. To address this issue, we propose DeGS, a scalable architecture for efficient 3DGS inference. To systematically eliminate the redundancies inherent in rendering, DeGS exploits a decoupled dataflow, restructuring the coupled $α$-checking, transmittance checking, and $α$-blending of the standard rendering process into consecutive workload parsing, reorganization, and blending stages. This allows the fragmented, length-variable, and temporal-dependent workloads to be reorganized into compact, conflict-free, and dense workloads prior to blending, thereby significantly improving PE utilization during parallel blending. Implemented in 28 nm technology, DeGS achieves 2.36$\times$--7.25$\times$ throughput, 1.82$\times$--6.02$\times$ end-to-end speedup, and 1.59$\times$--4.42$\times$ energy efficiency over state-of-the-art 3DGS accelerators (GSCore, GBU, GCC) across diverse scenes and resolutions (720p to 8K). Moreover, scaling from 16 to 1024 PEs, DeGS maintains over 80\% PE utilization at high resolutions, significantly outperforming existing accelerators.

Problem

• 3D Gaussian Splatting
• architectural scalability
• PE underutilization
• spatial redundancy
• temporal redundancy

Innovation

• Decoupled Dataflow
• 3D Gaussian Splatting
• Workload Reorganization
• Scalable Architecture
• PE Utilization