From Splats to Silicon: Rethinking Computational Efficiency of 3DGS

Minnan Pei; Qiwei Dong; Yihan Zhou; Gang Li; Yuchen Zhu; Wenju Zhao; Zhongtian Long; Siting Wang

cs.ARSep 5, 2026

From Splats to Silicon: Rethinking Computational Efficiency of 3DGS

Authors: Minnan Pei, Qiwei Dong, Yihan Zhou, Gang Li, Yuchen Zhu, Wenju Zhao, Zhongtian Long, Siting Wang, +2 more

Organizations: Institute of Automation, Chinese Academy of Sciences, and also with the University of Chinese Academy of Sciences · Nanjing University · Eindhoven University of Technology · School of Computer Science and Technology, Huazhong University of Science and Technology

Abstract

3D Gaussian splatting (3DGS) represents scenes with explicit primitives and supports real-time novel-view synthesis, yet its system efficiency varies substantially across scenes, viewpoints, rendering paths, and platform constraints. Existing studies pursue efficiency through representation and algorithm design, GPU runtime optimization, and architectural support, but their reported gains correspond to different points along the rendering and update paths. Connecting these indicators to end-to-end system benefit requires tracing how each optimization changes Gaussian selection, screen-space work, data movement, and stage or frame time. We therefore use a workload-centric framework to connect representation and algorithm research, GPU runtimes, and hardware architectures and to identify recurring workload patterns. We complement literature analysis with reproduced measurements and controlled GPU profiling of selected implementations, relating workload counts to stage time and memory traffic. Together, these comparisons show that system gains depend on workload reductions reaching downstream execution, granularity matching each stage, and the cost of data transfers, synchronization, and cached results, gradients, and optimizer data. Building on these findings, we discuss more consistent evaluation under rendering-quality constraints and identify key directions for future system design.

Figures & tables

Fig. 1: Workload-centric map of efficient 3DGS systems. The upper panels connect Sections 2–7 to the analysis framework, and the lower timeline places representative algorithm, GPU runtime, and architecture studies in approximate chronological order within each year.

Fig. 2: 3DGS forward paths with the five counts used in the analysis. Forward rendering proceeds from Gaussians stored in rendering memory through visibility culling, projection, Gaussian–tile association generation, depth ordering, candidate evaluation, and alpha compositing.

Fig. 3: Representative algorithmic interventions in 3DGS, spanning representation compaction, primitive reduction, adaptive activation, screen-space work reduction, and temporal reuse.

Fig. 4: Rendering speedup versus (a) storage compression ratio and percentage reductions in (b) Gaussians, (c) active Gaussians, and (d) Gaussian–tile associations, matched within each source. Similar association reductions can yield different speedups. Bars span minima to maxima; hollow markers indicate contextual comparisons.

Category	Method	Mechanism	Workload	Condition
Representation compaction	Reducing Memory Footprint [ 37 ]	Sparse pruning + quantization	$N_{\mathrm{resident}}$ + storage bytes	Decoding and training cost, slight quality loss
	MEGS2 [ 39 ]	SG representation + unified pruning	$N_{\mathrm{resident}}$ + storage bytes	Fine-tuning, platforms unmatched
	Flux-GS [ 48 ]	Energy aggregation + mobile compression	$N_{\mathrm{resident}}$ + stored bytes and traffic	Preprocessing, synchronization, thermal context
	Provable Pruning via Coresets [ 42 ]	Coreset pruning	$N_{\mathrm{resident}}$	Theoretical guarantee, no FPS/traffic
	Compressed 3DGS [ 43 ]	Pruning + SH/VQ/entropy coding	$N_{\mathrm{resident}}$ + attribute bytes	Hardware rasterization, decoding, and QAT
	Compact 3DGS [ 22 ]	Masking + compact color + VQ	$N_{\mathrm{resident}}$ + attribute bytes	Small MLP, decoding, and training

TABLE I: Representative Algorithmic Methods and Workload Effects

Fig. 5: GPU runtime filtering, scheduling, and data placement. Gaussian-aware intersection reduces invalid evaluations, dynamic scheduling redistributes uneven tasks, and CPU–GPU tiering or multi-GPU sharding distributes Gaussian parameters and optimizer data.

Fig. 6: Runtime evidence under controlled tile size changes and matched source conditions. Panels (a)–(c) show workload, timing, and tile list imbalance. Panel (d) relates workload reduction to speedup under matched conditions within each source; both axes use

\operatorname{asinh}((r-1)/0.05)

with ticks labelled by actual ratios

r

. The Faster-GS point compares the official testbed’s basis and optimized rasterizers under matched inputs, measuring complete forward rendering rather than full training.

Category	Method	Mechanism	Work unit	Effect
Spatial organization	FlashGS [ 13 ]	Tighter tile overlap + compact lists	Gaussian–tile pairs and lists	Reduces $N_{\mathrm{pair}}$ , sorting, and list traffic
	Seele [ 10 ]	Preprocessing + contribution tests	Pixel candidates	Reduces low-contribution pixel work on mobile GPUs
	HiGS [ 16 ]	Macro-tile sort + fine-tile rasterization	Tiles, lists, and workers	Trades $N_{\mathrm{pair}}$ against $N_{\mathrm{eval}}$ and balances tasks
	LiteGS [ 69 ]	Block locality + optimized operators	Spatial blocks and operators	Improves locality in large-scene training
	LoBE-GS [ 70 ]	Balanced partitioning + visibility cropping	Scene partitions and visible Gaussians	Removes irrelevant work and balances partitions
Cooperative scheduling	Balanced 3DGS [ 15 ]	Dynamic tile scheduling + warp rendering	Tiles and warps	Reduces imbalance and unit cost

TABLE II: Representative GPU Runtime Methods and Workload Effects

Category	Design	Work object	Mechanism and direct effect	Coverage	Implementation	Reported resources/results
GPU extension	GauRast [ 25 ]	Candidates; pixel accumulators	Rasterizer extension accelerates evaluation/accumulation; Projection and Sorting remain on CUDA	Blending	RTL + post-layout; 28 nm	A: —; S: — 24 FPS ; 1.7 W
	GBU [ 11 ]	Ordered candidates; pixel accumulators	Row-wise engine caches Gaussian attributes and keeps pixel sums local	Blending	RTL + cycle model; 28 nm	A: 0.90 mm 2 ; S: 63 KB 92 FPS ; 0.22 W
Integrated GPU	Vorion [ 76 ]	Pixel accumulators; gradients	Rasterizer locally accumulates color/alpha gradients; remaining work runs on SIMT	Blending, Backward	Silicon + FPGA; 16 nm	A: 1.60 mm 2 ; S: — 6.4–19 FPS ; $<0.6$ W
Dedicated accelerator	GSCore [ 14 ]	Associations; subtile candidates	Shape overlap tests/subtile masks skip uncovered work; hierarchical sorting orders retained associations	Forward	RTL + cycle/memory; 28 nm	A: 3.95 mm 2 ; S: 272 KB 91.2 FPS ; 0.87 W
	GS-TG [ 72 ]	Neighboring tiles’ depth lists	Shared order reduces repeated sorting across tiles; rasterization remains per tile	Forward	RTL + cycle model; 28 nm	A: 3.984 mm 2 ; S: 336 KB Relative; 1.063 W
	FLICKER [ 77 ]	Pixel candidates; contributions	Hierarchical contribution tests and smaller tiles skip unnecessary pixel work	Forward	RTL + memory model; 28 nm	A: 3.47 mm 2 ; S: 288 KB Relative; Relative

TABLE III: Representative 3DGS Hardware Designs: Work Objects, Mechanisms, and Reported Results

Fig. 7: Workload and kernel-timing relationships under controlled conditions. The panels connect active Gaussians to associations, associations to sorting time, and pixel evaluations to rasterization time, an operation measured within Blending.

Fig. 8: Same input across 3DGS backends. Panel (a) shows active Gaussians and materialized pairs; panel (b) groups kernels by Projection, Sorting, and Blending, keeping cross-stage fused kernels joint. Panels (c)–(d) show independently measured complete forward latency and DRAM traffic. Markers are medians over views, and lines span minima to maxima.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Fig. 9: Evidence coverage in the audited subset. Quality and end-to-end performance are commonly reported together. Workload and stage timing are less often jointly available.

Backend	Peak memory (GiB) median [min–max]	Max. absolute score difference
Backend	Peak memory (GiB) median [min–max]	PSNR	SSIM	LPIPS
Graphdeco	1.92 [1.88, 1.94]	0.000	0.0000	0.0000
gsplat	0.97 [0.81, 0.99]	0.036	0.0005	0.0003
FlashGS	0.94 [0.94, 0.94]	0.076	0.0028	0.0043
StopThePop	1.96 [1.92, 1.98]	0.001	0.0000	0.0000
Speedy-Splat	1.79 [1.75, 1.79]	0.001	0.0000	0.0000

Appendix

TABLE IV: Backend Memory and Reconstruction Quality

Explore similar work

Local-GS: Accelerating 3D Gaussian Splatting via Tile-Local Warp Coherence
Jun 15, 2026Yang Luo, Yan Gong, Yongsheng Gao +33D Gaussian Splatting 3DGS Acceleration
State Key Laboratory of Robotics and Systems, Harbin Institute of Technology, Harbin 150001, China · State Key Laboratory of Intelligent Green Vehicle and Mobility, the School of Vehicle and Mobility, Tsinghua University, Beijing 100084, China · Department of Computer Science and Technology, Tsinghua University, Beijing 100084, China
DeGS: A Scalable 3DGS Architecture via Decoupled Workload Parsing and Reorganization
Aug 3, 2026Minnan Pei, Gang Li, Zeyu Zhu +73D Gaussian Splatting 3DGS Acceleration
Institute of Automation, Chinese Academy of Sciences, C2DL, Beijing, China · University of Chinese Academy of Sciences, Beijing, China · Shanghai Jiao Tong University, Shanghai, China
SPARE-GS: Structural Parsimony and Resource Efficiency for 3D Gaussian Splatting
Jul 18, 2026Zhang Chen, Shuai Wan, Fuzheng Yang +33D Gaussian Splatting 3D Gaussian Pruning
School of Electronics and Information, Northwestern Polytechnical University, Xi’an 710129, China · Department of Computer Science, City University of Hong Kong, Hong Kong SAR, China · School of Telecommunication Engineering, Xidian University, Xi’an 710071, China +2

Jun 15, 2026J/K move · Enter open · S save

Local-GS: Accelerating 3D Gaussian Splatting via Tile-Local Warp Coherence

Yang Luo, Yan Gong, Yongsheng Gao, Jie Zhao+2

State Key Laboratory of Robotics and Systems, Harbin Institute of Technology, Harbin 150001, China · State Key Laboratory of Intelligent Green Vehicle and Mobility, the School of Vehicle and Mobility, Tsinghua University, Beijing 100084, China · Department of Computer Science and Technology, Tsinghua University, Beijing 100084, China

3D Gaussian Splatting 3DGS Acceleration GPU Kernel Optimization GPU Acceleration

Open paper AskPDF arXiv

3D Gaussian Splatting (3DGS) has significantly advanced real-time novel view synthesis by representing scenes as dense collections of anisotropic 3D Gaussian primitives. However, the irregular spatial distribution of Gaussians often leads to poor GPU utilization, as warp divergence and redundant computation degrade rendering performance. To address this, we present Local-GS, a warp-coherent rendering paradigm that, organizes Gaussian primitives with respect to SIMT (Single Instruction, Multiple Threads) execution boundaries rather than scene geometry. Specifically, we propose three warp-coherent stages: a hoisting stage that precomputes shared parameters at tile level, a culling stage that discards warps with no contribution, and a blending stage that replaces per-pixel branching with a uniform instruction stream. Across extensive benchmarks on multiple datasets, Local-GS improves efficiency without compromising quality. As a plug-and-play optimization, it provides additional performance gains to all tested baselines, culminating in a

7.76\times

speedup on Deep Blending scenes.

Abstract

Explore similar work

Local-GS: Accelerating 3D Gaussian Splatting via Tile-Local Warp Coherence

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

SPARE-GS: Structural Parsimony and Resource Efficiency for 3D Gaussian Splatting