Pith. sign in

REVIEW 4 major objections 4 minor 73 references

DeGS restructures 3DGS rendering into decoupled parse, reorganize, and blend stages, keeping processing-element utilization above 80% from 16 to 1024 units.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-04 15:09 UTC pith:E2AZZAUQ

load-bearing objection A genuinely new decoupled dataflow for 3DGS accelerators, with a credible scaling story that is somewhat oversold in the abstract; the core idea holds up, but the headline scalability claim depends heavily on Orin-class DRAM bandwidth and unpublished simulators. the 4 major comments →

arxiv 2608.02099 v1 pith:E2AZZAUQ submitted 2026-08-03 cs.AR cs.AIcs.CV

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

Minnan Pei , Gang Li , Zeyu Zhu , Siting Wang , Junwen Si , Zhuoran Song , Yu Feng , Fangxin Liu
show 2 more authors
Xiaoyao Liang Jian Cheng
This is my paper
classification cs.AR cs.AIcs.CV
keywords 3D Gaussian Splattingdomain-specific acceleratordecoupled dataflowspan parsingtask reorganizationPE utilizationrendering scalability3DGS inference
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that the poor scaling of 3D Gaussian Splatting (3DGS) accelerators is not caused by insufficient arithmetic throughput, but by a tightly coupled 'checking-while-blending' dataflow that wastes processing elements on spatially invalid pixels and temporarily stalled ones. DeGS restructures the standard blending recurrence into three consecutive stages—span parsing, task reorganization, and blending execution—so that only dense, conflict-free, valid workloads reach the arithmetic array. If correct, this means the rendering backend can be widened from 16 to 1024 PEs while retaining over 80% utilization at high resolutions, instead of the 10–35% utilization observed in prior accelerators. The result would make scalable 3DGS inference practical for high-resolution and edge real-time rendering without compromising image quality (worst-case 0.48 dB PSNR drop).

Core claim

The central claim is that the scaling wall of 3DGS accelerators stems from the coupled execution of alpha-checking, transmittance checking, and alpha-blending, which makes spatial redundancy (pixels outside Gaussian ellipses) and temporal redundancy (pixels terminated early waiting for long-tail pixels) unavoidable and increasingly costly as PE counts grow. DeGS eliminates these redundancies before blending: Span Parsing computes exact valid row intervals using the analytic ellipse of each projected Gaussian via a closed-form DDA recurrence, Task Reorganization partitions spans into conflict domains and packs dependence-safe fixed-width packets, and Blending Execution then only reconstructs

What carries the argument

The central object is the 'span vector'—a compact descriptor (y, x0, L, g0, d0, A) that fully encodes a valid row span of a projected Gaussian, derived by rewriting the quadratic form q(x,y) so that the span center is x_c = mu_x + m·dy and the half-width is w(dy) = r_x·sqrt(1 - dy^2·inv_r2_y). This reduces per-pixel boundary probing to a 1D scanline problem, and the closed-form q_n = g0 + n·d0 + A·n(n-1) lets blending reconstruct the quadratic value without loop-carried dependence. The Packing Scheduler's conflict-domain scheduling and packet compaction are the companion mechanism that converts length-variable, depth-dependent spans into dense, conflict-free fixed-width packets.

Load-bearing premise

The evaluation assumes that raising DRAM bandwidth makes the preprocessing stage (culling, projection, sorting) non-dominant, so the decoupled blending dataflow determines end-to-end speedup; at lower bandwidths the paper's own sensitivity analysis shows DeGS becomes memory-bound and the scaling advantage shrinks.

What would settle it

Run the DeGS cycle-accurate simulator at 256 PEs on a 1080p scene with 51.2 GB/s DRAM bandwidth; if end-to-end speedup over the strongest prior accelerator falls below the reported 1.82x lower bound, the claim that decoupled workload parsing—rather than memory bandwidth—is the source of the scaling gain would be contradicted.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • DeGS preserves standard 3DGS blending semantics; image quality remains within 0.48 dB PSNR and 0.0038 SSIM of a GPU reference across 720p–8K.
  • Because redundancy is removed before blending, the blending array can be purely arithmetic; scaling PE count from 16 to 1024 keeps utilization above 80% at high resolutions.
  • The decoupled dataflow makes larger tile sizes beneficial rather than harmful, since tile growth no longer amplifies spatial and temporal redundancy.
  • On dynamic-scene 4DGS traces, DeGS improves blending PE utilization from 23.9% to 90.2% and achieves a 4.04x rendering speedup, showing applicability beyond static scenes.
  • The approach outperforms an NPU-based GEMM-reformulated baseline at smaller area and avoids the utilization collapse of wide systolic arrays.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the decoupled dataflow is the real driver of the reported gains, the same parse-reorganize-blend pattern could be applied to GPU software rasterizers for 3DGS, potentially improving utilization on commodity hardware without custom silicon.
  • The paper's sensitivity analysis implies that the design's advantage narrows at lower DRAM bandwidth; for edge devices without Orin-class memory, the decoupled frontend would likely need to be paired with bandwidth reduction techniques (e.g., Gaussian caching or compression) to retain the scaling benefit.
  • A fixed 4-cycle reuse distance for read-modify-write hazards may be an implementation artifact; an adaptive delay tuned to actual write-back latency could further reduce scheduling stalls, especially at smaller technology nodes.
  • Because the span descriptor depends only on the Gaussian's analytic ellipse, the same frontend could serve anti-aliasing or foveated rendering by adjusting the span half-width threshold per region.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper identifies poor scalability of existing 3DGS accelerators as a consequence of the coupled "checking-while-blending" dataflow, which exposes spatial and temporal workload irregularity to the PE array. It proposes DeGS, a decoupled three-stage dataflow (span parsing, task reorganization, blending execution) that precomputes valid row spans via an analytic ellipse solution, packs them into fixed-width conflict-free packets, and lets the blending array execute only dense, dependence-safe tasks. The authors derive a closed-form row-span quadratic reconstruction (Eq. 19), describe the microarchitecture (Span Engine, Packing Scheduler, Blending Array), synthesize the core logic in 28 nm, and evaluate against GSCore, GBU, GCC, and ORANGE using cycle-accurate simulators. The claimed results are 2.36x-7.25x throughput, 1.82x-6.02x end-to-end speedup, 1.59x-4.42x energy efficiency, and >80% PE utilization when scaling from 16 to 1024 PEs, with a worst-case PSNR gap of 0.478 dB.

Significance. The core idea is sound and relevant: decoupling validity/dependence handling from blending is a plausible way to prevent irregular Gaussian coverage from translating into backend underutilization. The analytic span derivation is explicit and internally consistent, and the numerical-error (Fig. 7) and end-to-end quality (Fig. 8) studies provide credible evidence that the reformulation preserves rendering semantics. The 4DGS applicability test is a useful robustness check, and the RTL synthesis of the core logic is a concrete implementation effort. If the performance claims survive scrutiny, this would be a meaningful step for scalable 3DGS inference. The main caveats concern the reporting and boundary conditions of the headline scaling claims, not the mathematical core.

major comments (4)
  1. [Sec. V.A, V.E, and Abstract] The abstract's scalability claim ('scaling from 16 to 1024 PEs, DeGS maintains over 80% PE utilization at high resolutions') is evaluated only under the default Orin-class LPDDR5 configuration (204.8 GB/s). The paper itself states in Sec. V.E that lower-bandwidth settings 'would make DeGS appear memory-bound and thus obscure the scaling behavior of the blending backend.' Because the central thesis is that decoupling makes the blending array scalable, this is a load-bearing boundary condition rather than a peripheral sensitivity result. The abstract and Sec. V.D should qualify the claim, or the authors should provide the analogous PE-utilization and scaling study at 51.2 GB/s to show what happens to the headline metric outside the chosen regime.
  2. [Sec. V.C and Fig. 9] The headline throughput/speedup/energy-efficiency numbers in the Abstract and Contributions are introduced immediately after 'Figure 9 reports area-normalized improvement factors.' Table I shows DeGS total area of 2.72 mm² versus 3.95 mm² for GSCore and 4.44 mm² for GBU, so area-normalized gains may differ materially from raw gains. The abstract states '2.36x-7.25x throughput' and '1.82x-6.02x end-to-end speedup' without saying these are area-normalized. The authors should report raw metrics in addition to area-normalized ones, or explicitly label every headline number as area-normalized.
  3. [Sec. V.D and Table III] The 'scaling from 16 to 1024 PEs' result is obtained under a coordinated scaling recipe that also changes tile size (16 to 128), number of GSUs (1 to 8), Span Units (4 to 32), Issue Banks (4 to 32), and PEs per array. Thus the >80% utilization claim is not for a fixed architecture but for a specific provisioning rule. Because these frontend resources are exactly what remove spatial/temporal redundancy, the utilization could partly reflect the coordinated provisioning rather than the decoupled dataflow itself. Please provide a scaling curve where only the PE array is widened while frontend resources are held fixed (or otherwise explicitly state that the claim is for the Table III recipe).
  4. [Sec. II.B.3 and Abstract] PE utilization is defined in the paper as utilization of the pixel-wise calculation in Eq. 6 and Eq. 8 only. The abstract's 'over 80% PE utilization' should carry the same qualification; otherwise readers may infer overall system utilization, which is affected by memory-bound preprocessing. This is related to the bandwidth concern above and should be clarified with the headline numbers.
minor comments (4)
  1. [Eq. (15)] The parameter δ is described only as 'estimates the error based on dw/dy.' Please specify how δ is computed or bounded; this is not obvious from the surrounding equations.
  2. [Sec. V.A] The performance evaluation relies on in-house cycle-accurate simulators for DeGS and all baselines, but no simulator validation or artifact release is mentioned. A short validation subsection (e.g., comparing simulator predictions against published baseline numbers or RTL cycle counts) would substantially strengthen reproducibility.
  3. [Table II] Metrics such as 'Active-Gaussian overlap', 'tile-workload churn', and 'footprint Gini' are used without definitions or formulas. They are not standard terms in the paper and should be defined for the 4DGS applicability study to be interpretable.
  4. [References] References [57]-[61] appear to be listed but are not cited in the visible text. Please check the citation list and remove uncited entries or cite them appropriately.

Circularity Check

0 steps flagged

No material circularity: the speedup and utilization results come from a self-contained simulation; the one minor self-citation (GCC [30]) is used only as a baseline and for a bandwidth premise that the paper's own sensitivity analysis conditions.

full rationale

I walked the derivation chain from Eq. 7/8 to the Span Parsing/Task Reorganization/Blending Execution dataflow. The span formulas (Eqs. 9-19) are an algebraic re-expression of the opacity-ellipse test, and the paper verifies equivalence empirically (PSNR gap <=0.478 dB), so the 'prediction' of regularized blending work is not equivalent to its inputs by construction. The performance/scalability claims are produced by cycle-accurate trace-driven simulation (Sec. V.A) using identical projected Gaussian workloads for all architectures; no fitted parameter is relabeled as a prediction. The main self-citation, GCC [30], is used in Sec. II.B to argue that DRAM bandwidth can make preprocessing non-dominant; this premise is corroborated by [37] and is explicitly made conditional in Sec. V.E: 'lower-bandwidth settings would make DeGS appear memory-bound and thus obscure the scaling behavior of the blending backend. Therefore, our main evaluation uses an Orin-class LPDDR5 configuration...' That is a disclosed boundary condition, not a circular step. Sec. II.B.2's definition of PE utilization as blending-array utilization is also disclosed and does not define the claimed speedup into existence. Overall, the closest issue is the non-load-bearing self-citation of GCC as a baseline/reference; it does not reduce the central result to prior work by the same authors.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The design has no new physical entities. Its claims rest on hand-chosen scheduling parameters (reuse distance, segment length, frontend provisioning) and on assumptions inherited from prior 3DGS work; none of these are fitted to hit a target result, but the unverified baseline simulations and bandwidth choice are the main ledger burden.

free parameters (4)
  • 4-cycle reuse distance = 4 cycles
    Fixed delay for overlapping segments to the same pixel region, matched to Blending Array writeback latency (Sec IV.B). If incorrect, read-modify-write hazards appear.
  • Segment length 1x16 = 16 pixels
    Segment granularity chosen in sensitivity study (Sec V.E) to keep redundancy low; not derived from first principles.
  • Coordinated scaling tile sizes = 16/32/64/128 for 16/64/256/1024 PEs
    Table III hand-provisions tile size and frontend resources; these choices directly affect the reported scaling utilization.
  • Frontend provisioning (GSUs, Span Units, Issue Banks) = 1/2/4/8, 4/8/16/32, 4/8/16/32
    Empirical sublinear scaling to avoid bottleneck; Fig. 13 shows frontend saturation in low-res large-array regime, so these choices bound the claimed scalability.
axioms (4)
  • domain assumption Gaussian projection follows the standard affine approximation using Jacobian J (Eq. 2)
    Inherited from original 3DGS paper [1]; not re-derived.
  • domain assumption Alpha threshold epsilon=1/255 and transmittance threshold tau=1e-4 define visible Gaussian support and early termination
    Standard 3DGS semantics [1]; DeGS relies on these thresholds to compute exact valid spans.
  • domain assumption Preprocessing (culling/projection/sorting) can be made non-dominant with sufficient DRAM bandwidth
    Sec II.B cites [30],[37]; enables the evaluation to focus on rendering dataflow. Sec V.E shows this fails at low bandwidth.
  • domain assumption Trace-driven cycle-accurate simulators faithfully model GSCore, GBU, GCC, and ORANGE
    Baselines are reimplemented from papers; no independent verification or released simulator.

pith-pipeline@v1.3.0-daily-deepseek · 29762 in / 13510 out tokens · 95025 ms · 2026-08-04T15:09:11.305225+00:00 · methodology

0 comments
read the original 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 $\alpha$-checking, transmittance checking, and $\alpha$-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.

Figures

Figures reproduced from arXiv: 2608.02099 by Fangxin Liu, Gang Li, Jian Cheng, Junwen Si, Minnan Pei, Siting Wang, Xiaoyao Liang, Yu Feng, Zeyu Zhu, Zhuoran Song.

Figure 1
Figure 1. Figure 1: Analysis of intrinsic spatial and temporal workload irregularity in Gaussian blending. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative comparison of workload regularity and PE utilization across 3DGS accelerators. DeGS reduces both spatially invalid work and temporal [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Decoupled DeGS dataflow. Span Parsing extracts valid row spans from projected Gaussians, Task Reorganization converts length-variable dependent spans into conflict-free packets, and Blending Execution operates on regularized packets. write dependence in Eq. 8. Consequently, irregular workloads across rows cannot be redistributed after assignment. Splat irregularity appears as unequal waiting windows across… view at source ↗
Figure 4
Figure 4. Figure 4: Overall design of DeGS architecture. C. Blending Execution After Span Parsing and Task Reorganization, the backend no longer receives raw candidate fragments. Instead, it receives fixed-width packets whose validity has already been resolved and whose update order has already been made dependence￾safe. Blending Execution therefore performs only three func￾tions: reconstruct qn in Eq. 19 from the span descri… view at source ↗
Figure 5
Figure 5. Figure 5: Architecture of Span Engine. after span generation, such as the coefficients and appearance fields later consumed by the Blending Array. This split reduces unnecessary parameter residency in the early pipeline stages and avoids transporting large Gaussian records through the Span Engine. B. Packing Scheduler The Packing Scheduler prevents variable span length and workload dependence from reaching the Blend… view at source ↗
Figure 7
Figure 7. Figure 7: Error of Single-Gaussian Parsing. 25 30 35 PSNR 720p 1080p 2K 4K 8K 0.85 0.90 0.95 SSIM 0.1 0.2 0.3 LPIPS DB T&T Mip DeGS GPU [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: End-to-End Rendering Quality. specialized architecture based on Gaussian-wise and cross￾stage dataflow reorganization. Table I further compares the area and power breakdowns of these designs. Across all architectures, area is mainly dominated by the projection and sorting (P&S) units, the Gaussian blending unit (Blending), and on-chip SRAM. The original GBU configuration uses only 8 blending PEs, which lea… view at source ↗
Figure 9
Figure 9. Figure 9: Area-normalized improvement factors across end-to-end latency [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Mechanism-level speed and energy ablation averaged across datasets. TABLE II TRACE-DERIVED 4DGS WORKLOAD EVIDENCE AND DEGS BACKEND BENEFIT AT 960 × 720, 16 × 16 TILE SIZE, AND 64 PES. Metric Coffee Spinach Beef Mean Dynamic and irregular 4DGS rendering traces Active-Gaussian overlap / tile-workload churn 0.450 / 4.6% 0.270 / 4.9% 0.266 / 5.7% 0.329 / 5.1% P99 tile candidates / footprint Gini 2.39K / 0.854… view at source ↗
Figure 12
Figure 12. Figure 12: Blending PE utilization of each accelerator. [PITH_FULL_IMAGE:figures/full_fig_p012_12.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

73 extracted references · 7 linked inside Pith

  1. [1]

    3d gaussian splatting for real-time radiance field rendering

    B. Kerbl, G. Kopanas, T. Leimkühler, and G. Drettakis, “3d gaussian splatting for real-time radiance field rendering.”ACM Trans. Graph., vol. 42, no. 4, pp. 139–1, 2023

    2023

  2. [2]

    3d gaussian splatting: Survey, technologies, challenges, and opportunities,

    Y . Bao, T. Ding, J. Huo, Y . Liu, Y . Li, W. Li, Y . Gao, and J. Luo, “3d gaussian splatting: Survey, technologies, challenges, and opportunities,” IEEE Transactions on Circuits and Systems for Video Technology, 2025

    2025

  3. [3]

    3d gaussian splatting as new era: A survey,

    B. Fei, J. Xu, R. Zhang, Q. Zhou, W. Yang, and Y . He, “3d gaussian splatting as new era: A survey,”IEEE Transactions on Visualization and Computer Graphics, 2024

    2024

  4. [4]

    Recent advances in 3d gaussian splatting,

    T. Wu, Y .-J. Yuan, L.-X. Zhang, J. Yang, Y .-P. Cao, L.-Q. Yan, and L. Gao, “Recent advances in 3d gaussian splatting,”Computational Visual Media, vol. 10, no. 4, pp. 613–642, 2024

    2024

  5. [5]

    Gaussian splatting: 3d reconstruction and novel view synthesis, a review,

    A. Dalal, D. Hagen, K. G. Robbersmyr, and K. M. Knausgård, “Gaussian splatting: 3d reconstruction and novel view synthesis, a review,”IEEE Access, 2024

    2024

  6. [6]

    2d gaussian splatting for geometrically accurate radiance fields,

    B. Huang, Z. Yu, A. Chen, A. Geiger, and S. Gao, “2d gaussian splatting for geometrically accurate radiance fields,” inACM SIGGRAPH 2024 conference papers, 2024, pp. 1–11

    2024

  7. [7]

    3dgs. zip: A survey on 3d gaussian splatting compression methods,

    M. T. Bagdasarian, P. Knoll, Y . Li, F. Barthel, A. Hilsmann, P. Eisert, and W. Morgenstern, “3dgs. zip: A survey on 3d gaussian splatting compression methods,” inComputer Graphics Forum, vol. 44, no. 2. Wiley Online Library, 2025, p. e70078

    2025

  8. [8]

    Real-time photorealistic dynamic scene representation and rendering with 4d gaussian splatting,

    Z. Yang, H. Yang, Z. Pan, and L. Zhang, “Real-time photorealistic dynamic scene representation and rendering with 4d gaussian splatting,” arXiv preprint arXiv:2310.10642, 2023

    Pith/arXiv arXiv 2023

  9. [9]

    4d gaussian splatting for real-time dynamic scene rendering,

    G. Wu, T. Yi, J. Fang, L. Xie, X. Zhang, W. Wei, W. Liu, Q. Tian, and X. Wang, “4d gaussian splatting for real-time dynamic scene rendering,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 20 310–20 320

    2024

  10. [10]

    Ges: Generalized exponential splatting for efficient radiance field rendering,

    A. Hamdi, L. Melas-Kyriazi, J. Mai, G. Qian, R. Liu, C. V ondrick, B. Ghanem, and A. Vedaldi, “Ges: Generalized exponential splatting for efficient radiance field rendering,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 812–19 822

    2024

  11. [11]

    Gs-scale: Unlocking large- scale 3d gaussian splatting training via host offloading,

    D. Lee, D. Jeong, J. W. Lee, and H. Yoon, “Gs-scale: Unlocking large- scale 3d gaussian splatting training via host offloading,”arXiv preprint arXiv:2509.15645, 2025

    arXiv 2025

  12. [12]

    Adr-gaussian: Accelerating gaussian splatting with adaptive radius,

    X. Wang, R. Yi, and L. Ma, “Adr-gaussian: Accelerating gaussian splatting with adaptive radius,” inSIGGRAPH Asia 2024 Conference Papers, 2024, pp. 1–10

    2024

  13. [13]

    Vr-gs: A physical dynamics-aware interactive gaussian splatting system in virtual reality,

    Y . Jiang, C. Yu, T. Xie, X. Li, Y . Feng, H. Wang, M. Li, H. Lau, F. Gao, Y . Yanget al., “Vr-gs: A physical dynamics-aware interactive gaussian splatting system in virtual reality,” inACM SIGGRAPH 2024 Conference Papers, 2024, pp. 1–1

    2024

  14. [14]

    Vrsplat: Fast and robust gaussian splatting for virtual reality,

    X. Tu, L. Radl, M. Steiner, M. Steinberger, B. Kerbl, and F. De la Torre, “Vrsplat: Fast and robust gaussian splatting for virtual reality,” Proceedings of the ACM on Computer Graphics and Interactive Tech- niques, vol. 8, no. 1, pp. 1–22, 2025

    2025

  15. [15]

    Taoavatar: Real-time lifelike full-body talking avatars for augmented reality via 3d gaussian splatting,

    J. Chen, J. Hu, G. Wang, Z. Jiang, T. Zhou, Z. Chen, and C. Lv, “Taoavatar: Real-time lifelike full-body talking avatars for augmented reality via 3d gaussian splatting,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 10 723–10 734

    2025

  16. [16]

    Gaussianshopvr: Facilitating immersive 3d authoring using gaussian splatting in vr,

    Y . Shen, B. Li, J. Huang, D. Yip, and Z. Wang, “Gaussianshopvr: Facilitating immersive 3d authoring using gaussian splatting in vr,” in Proceedings of the 38th Annual ACM Symposium on User Interface Software and Technology, 2025, pp. 1–14

    2025

  17. [17]

    Vr-splatting: Foveated radiance field rendering via 3d gaussian splatting and neural points,

    L. Franke, L. Fink, and M. Stamminger, “Vr-splatting: Foveated radiance field rendering via 3d gaussian splatting and neural points,”Proceedings of the ACM on Computer Graphics and Interactive Techniques, vol. 8, no. 1, pp. 1–21, 2025

    2025

  18. [18]

    Wildgs-slam: Monocular gaussian splatting slam in dynamic environ- ments,

    J. Zheng, Z. Zhu, V . Bieri, M. Pollefeys, S. Peng, and I. Armeni, “Wildgs-slam: Monocular gaussian splatting slam in dynamic environ- ments,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 11 461–11 471

    2025

  19. [19]

    Cg-slam: Efficient dense rgb-d slam in a consistent uncertainty- aware 3d gaussian field,

    J. Hu, X. Chen, B. Feng, G. Li, L. Yang, H. Bao, G. Zhang, and Z. Cui, “Cg-slam: Efficient dense rgb-d slam in a consistent uncertainty- aware 3d gaussian field,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 93–112

    2024

  20. [20]

    Rtg- slam: Real-time 3d reconstruction at scale using gaussian splatting,

    Z. Peng, T. Shao, Y . Liu, J. Zhou, Y . Yang, J. Wang, and K. Zhou, “Rtg- slam: Real-time 3d reconstruction at scale using gaussian splatting,” in ACM SIGGRAPH 2024 conference papers, 2024, pp. 1–11

    2024

  21. [21]

    Sgs- slam: Semantic gaussian splatting for neural dense slam,

    M. Li, S. Liu, H. Zhou, G. Zhu, N. Cheng, T. Deng, and H. Wang, “Sgs- slam: Semantic gaussian splatting for neural dense slam,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 163–179

    2024

  22. [22]

    Gaussian-lic: Real-time photo-realistic slam with gaussian splatting and lidar-inertial-camera fusion,

    X. Lang, L. Li, C. Wu, C. Zhao, L. Liu, Y . Liu, J. Lv, and X. Zuo, “Gaussian-lic: Real-time photo-realistic slam with gaussian splatting and lidar-inertial-camera fusion,” in2025 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2025, pp. 8500–8507

    2025

  23. [23]

    Splatam: Splat track & map 3d gaussians for dense rgb-d slam,

    N. Keetha, J. Karhade, K. M. Jatavallabhula, G. Yang, S. Scherer, D. Ramanan, and J. Luiten, “Splatam: Splat track & map 3d gaussians for dense rgb-d slam,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 21 357–21 366

    2024

  24. [24]

    Gaussianproperty: Integrating physical prop- erties to 3d gaussians with lmms,

    X. Xu, W. Ge, D. Qiu, Z. Chen, D. Yan, Z. Liu, H. Zhao, H. Zhao, S. Zhang, J. Lianget al., “Gaussianproperty: Integrating physical prop- erties to 3d gaussians with lmms,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 7231–7240

    2025

  25. [25]

    Phys- gaussian: Physics-integrated 3d gaussians for generative dynamics,

    T. Xie, Z. Zong, Y . Qiu, X. Li, Y . Feng, Y . Yang, and C. Jiang, “Phys- gaussian: Physics-integrated 3d gaussians for generative dynamics,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 4389–4398

    2024

  26. [26]

    Physics3d: Learning physical properties of 3d gaussians via video diffusion,

    F. Liu, H. Wang, S. Yao, S. Zhang, J. Zhou, and Y . Duan, “Physics3d: Learning physical properties of 3d gaussians via video diffusion,”arXiv preprint arXiv:2406.04338, 2024

    Pith/arXiv arXiv 2024

  27. [27]

    Ready-go: Real-to-sim dynamic 3d gaussian splatting simulation for environment- specific visual navigation with moving obstacles,

    S. Yoo, Y . Jang, D. Kim, Y . Han, S. Jung, and H. J. Kim, “Ready-go: Real-to-sim dynamic 3d gaussian splatting simulation for environment- specific visual navigation with moving obstacles,”arXiv preprint arXiv:2602.11575, 2026

    Pith/arXiv arXiv 2026

  28. [28]

    Splatad: Real-time lidar and camera rendering with 3d gaussian splat- ting for autonomous driving,

    G. Hess, C. Lindström, M. Fatemi, C. Petersson, and L. Svensson, “Splatad: Real-time lidar and camera rendering with 3d gaussian splat- ting for autonomous driving,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 11 982–11 992

    2025

  29. [29]

    Metasapiens: Real-time neural rendering with efficiency-aware pruning and accelerated foveated rendering,

    W. Lin, Y . Feng, and Y . Zhu, “Metasapiens: Real-time neural rendering with efficiency-aware pruning and accelerated foveated rendering,” in Proceedings of the 30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 1, 2025, pp. 669–682

    2025

  30. [30]

    Gcc: A 3dgs inference architecture with gaussian-wise and cross-stage conditional processing,

    M. Pei, G. Li, J. Si, Z. Zhu, Z. Mo, P. Wang, Z. Song, X. Liang, and J. Cheng, “Gcc: A 3dgs inference architecture with gaussian-wise and cross-stage conditional processing,” inProceedings of the 58th IEEE/ACM International Symposium on Microarchitecture, 2025, pp. 1824–1837

    2025

  31. [31]

    Gsnorm: An efficient 3d gaussian rendering accelerator with splat normalization and lut-assist rasterization,

    Y . Sun, P. Yan, Y . Jing, L. Ye, and T. Jia, “Gsnorm: An efficient 3d gaussian rendering accelerator with splat normalization and lut-assist rasterization,” inProceedings of the 30th Asia and South Pacific Design Automation Conference, 2025, pp. 1379–1385

    2025

  32. [32]

    Gscore: Efficient radiance field rendering via architectural support for 3d gaussian splatting,

    J. Lee, S. Lee, J. Lee, J. Park, and J. Sim, “Gscore: Efficient radiance field rendering via architectural support for 3d gaussian splatting,” in Proceedings of the 29th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 3, 2024, pp. 497–511

    2024

  33. [33]

    Gaussian blending unit: An edge gpu plug-in for real-time gaussian-based rendering in ar/vr,

    Z. Ye, Y . Fu, J. Zhang, L. Li, Y . Zhang, S. Li, C. Wan, C. Wan, C. Li, S. Prathipatiet al., “Gaussian blending unit: An edge gpu plug-in for real-time gaussian-based rendering in ar/vr,” in2025 IEEE International Symposium on High Performance Computer Architecture (HPCA). IEEE, 2025, pp. 353–365

    2025

  34. [34]

    Flashgs: Efficient 3d gaussian splatting for large- scale and high-resolution rendering,

    G. Feng, S. Chen, R. Fu, Z. Liao, Y . Wang, T. Liu, Z. Pei, H. Li, X. Zhang, and B. Dai, “Flashgs: Efficient 3d gaussian splatting for large- scale and high-resolution rendering,”arXiv preprint arXiv:2408.07967, 2024

    Pith/arXiv arXiv 2024

  35. [35]

    Characterization and analysis of the 3d gaussian splatting rendering pipeline,

    J. Lee, Y . Lee, Y . Kwon, and M. Rhu, “Characterization and analysis of the 3d gaussian splatting rendering pipeline,”IEEE Computer Architec- ture Letters, vol. 24, no. 1, pp. 13–16, 2024

    2024

  36. [36]

    Speedy-splat: Fast 3d gaussian splatting with sparse pixels and sparse primitives,

    A. Hanson, A. Tu, G. Lin, V . Singla, M. Zwicker, and T. Goldstein, “Speedy-splat: Fast 3d gaussian splatting with sparse pixels and sparse primitives,” inProceedings of the Computer Vision and Pattern Recog- nition Conference, 2025, pp. 21 537–21 546

    2025

  37. [37]

    Neo: Real-time on-device 3d gaussian splatting with reuse-and-update sorting acceleration,

    C. Oh, S. Oh, J. Hwang, Y . Kim, H. Sharma, and J. Park, “Neo: Real-time on-device 3d gaussian splatting with reuse-and-update sorting acceleration,”arXiv preprint arXiv:2511.12930, 2025

    arXiv 2025

  38. [38]

    Nebula: Enable city-scale 3d gaussian splatting in virtual reality via collaborative rendering and accelerated stereo rasterization,

    H. Zhu, Z. Liu, X. Li, A. Wu, J. Zhao, F. Liu, Y . Gan, J. Leng, and Y . Feng, “Nebula: Enable city-scale 3d gaussian splatting in virtual reality via collaborative rendering and accelerated stereo rasterization,” arXiv preprint arXiv:2512.20495, 2025

    arXiv 2025

  39. [39]

    Orange: Exploring ockham’s razor for neural rendering by accelerating 3dgs on npus with gemm-friendly blending and balanced workloads,

    H. Li, Y . Liang, F. Liu, B. Zhu, Z. Wang, Y . Feng, L. Lu, L. Jiang, and H. Guan, “Orange: Exploring ockham’s razor for neural rendering by accelerating 3dgs on npus with gemm-friendly blending and balanced workloads,” in2026 IEEE International Symposium on High Perfor- mance Computer Architecture (HPCA). IEEE, 2026, pp. 1–15

    2026

  40. [40]

    Gaurast: Enhancing gpu triangle rasterizers to accelerate 3d gaussian splatting,

    S. Li, B. Keller, Y . C. Lin, and B. Khailany, “Gaurast: Enhancing gpu triangle rasterizers to accelerate 3d gaussian splatting,” in2025 62nd ACM/IEEE Design Automation Conference (DAC). IEEE, 2025, pp. 1–7

    2025

  41. [41]

    Tanks and temples: Benchmarking large-scale scene reconstruction,

    A. Knapitsch, J. Park, Q.-Y . Zhou, and V . Koltun, “Tanks and temples: Benchmarking large-scale scene reconstruction,”ACM Transactions on Graphics (ToG), vol. 36, no. 4, pp. 1–13, 2017

    2017

  42. [42]

    Deep blending for free-viewpoint image-based rendering,

    P. Hedman, J. Philip, T. Price, J.-M. Frahm, G. Drettakis, and G. Bros- tow, “Deep blending for free-viewpoint image-based rendering,”ACM Transactions on Graphics (ToG), vol. 37, no. 6, pp. 1–15, 2018

    2018

  43. [43]

    Mip-nerf 360: Unbounded anti-aliased neural radiance fields,

    J. T. Barron, B. Mildenhall, D. Verbin, P. P. Srinivasan, and P. Hedman, “Mip-nerf 360: Unbounded anti-aliased neural radiance fields,” inPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 5470–5479

    2022

  44. [44]

    GeForce RTX 3090 Family,

    NVIDIA Corporation, “GeForce RTX 3090 Family,” https://www.nvidia.com/en-us/geforce/graphics-cards/30-series/rtx- 3090/, 2020

    2020

  45. [45]

    LPDDR5 SDRAM,

    Micron Technology, Inc., “LPDDR5 SDRAM,” Micron Technology, Inc., Datasheet Rev. A, May 2021, accessed: 2026-03-29. [Online]. Available: https://assets.sourcengine.com/datasheets/8d27f742- 173e-4aa6-85b2-d1eef2c66360.pdf

    2021

  46. [46]

    NVIDIA Jetson AGX Orin Series Technical Brief,

    NVIDIA Corporation, “NVIDIA Jetson AGX Orin Series Technical Brief,” NVIDIA Corporation, Technical Brief, 2022, accessed: 2026- 03-29. [Online]. Available: https://www.nvidia.com/content/dam/en- zz/Solutions/gtcf21/jetson-orin/nvidia-jetson-agx-orin-technical- brief.pdf

    2022

  47. [47]

    Cacti- p: Architecture-level modeling for sram-based structures with advanced leakage reduction techniques,

    S. Li, K. Chen, J. H. Ahn, J. B. Brockman, and N. P. Jouppi, “Cacti- p: Architecture-level modeling for sram-based structures with advanced leakage reduction techniques,” inICCAD: International Conference on Computer-Aided Design, 2011, pp. 694–701

    2011

  48. [48]

    Vr-pipe: Streamlining hardware graphics pipeline for volume rendering,

    J. Lee, J. Kim, J. Park, and J. Sim, “Vr-pipe: Streamlining hardware graphics pipeline for volume rendering,”arXiv preprint arXiv:2502.17078, 2025

    Pith/arXiv arXiv 2025

  49. [49]

    Lumina: Real-time neural rendering by exploiting compu- tational redundancy,

    Y . Feng, W. Lin, Y . Cheng, Z. Liu, J. Leng, M. Guo, C. Chen, S. Sun, and Y . Zhu, “Lumina: Real-time neural rendering by exploiting compu- tational redundancy,” inProceedings of the 52nd Annual International Symposium on Computer Architecture, 2025, pp. 1925–1939

    2025

  50. [50]

    Hypergs: Efficient real-time 3d gaussian rendering processor through hierarchical sorting,

    C. Nian, X. Mo, J. Peng, W. Zhang, F. U. D. Farrukh, F. Chen, and C. Zhang, “Hypergs: Efficient real-time 3d gaussian rendering processor through hierarchical sorting,” in2025 IEEE International Symposium on Circuits and Systems (ISCAS). IEEE, 2025, pp. 1–5

    2025

  51. [51]

    Gauspu: 3d gaussian splatting processor for real-time slam systems,

    L. Wu, H. Zhu, S. He, J. Zheng, C. Chen, and X. Zeng, “Gauspu: 3d gaussian splatting processor for real-time slam systems,” in2024 57th IEEE/ACM International Symposium on Microarchitecture (MICRO). IEEE, 2024, pp. 1562–1573

    2024

  52. [52]

    Splatonic: Architecture support for 3d gaussian splatting slam via sparse processing,

    X. Huang, H. Zhu, T. Ma, Y . Xiong, F. Liu, Z. He, Y . Gan, Z. Liu, J. Leng, Y . Fenget al., “Splatonic: Architecture support for 3d gaussian splatting slam via sparse processing,”arXiv preprint arXiv:2511.18755, 2025

    arXiv 2025

  53. [53]

    Arc: Warp-level adaptive atomic reduction in gpus to accelerate differentiable rendering,

    S. Durvasula, A. Zhao, F. Chen, R. Liang, P. K. Sanjaya, Y . Guan, C. Giannoula, and N. Vijaykumar, “Arc: Warp-level adaptive atomic reduction in gpus to accelerate differentiable rendering,” inASPLOS, 2025

    2025

  54. [54]

    Gsarch: Breaking memory barriers in 3d gaussian splatting training via architectural support,

    H. He, G. Li, F. Liu, L. Jiang, X. Liang, and Z. Song, “Gsarch: Breaking memory barriers in 3d gaussian splatting training via architectural support,” in2025 IEEE International Symposium on High Performance Computer Architecture (HPCA), 2025, pp. 366–379

    2025

  55. [55]

    Accelerating 3d gaussian splatting with neural sort- ing and axis-oriented rasterization,

    Z. Wang, G. He, D. Liu, L. Gao, S. X. Hu, C. Zhang, Z. Song, N. Lane, W. Luk, and H. Fan, “Accelerating 3d gaussian splatting with neural sort- ing and axis-oriented rasterization,”arXiv preprint arXiv:2506.07069, 2025

    Pith/arXiv arXiv 2025

  56. [56]

    Rtgs: Real-time 3d gaussian splatting slam via multi-level redundancy reduction,

    L. Li, J. Qin, J. Peng, Z. Wan, H. Qu, Y . Han, P. Zheng, H. Zhang, Y . Cao, T. Chenet al., “Rtgs: Real-time 3d gaussian splatting slam via multi-level redundancy reduction,” inProceedings of the 58th IEEE/ACM International Symposium on Microarchitecture, 2025, pp. 1838–1851

    2025

  57. [57]

    Area-efficient automated logic design with Monte-Carlo tree search,

    S. Cheng, X. Guan, and Z. Du, “Area-efficient automated logic design with Monte-Carlo tree search,” inAdvanced Parallel Processing Tech- nologies: 16th International Symposium, APPT 2025, Athens, Greece, July 13–16, 2025, Proceedings, ser. Lecture Notes in Computer Science, vol. 16062. Springer Nature Singapore, 2025, pp. 47–58

    2025

  58. [58]

    VecTEE: Compact TEE metadata caching for efficient secure vector computing,

    H. Han, D. Huang, T. Ma, X. Zheng, J. Mu, Z. Du, X. Hu, and Q. Guo, “VecTEE: Compact TEE metadata caching for efficient secure vector computing,” Presented at the 16th International Symposium on Advanced Parallel Processing Technologies (APPT), Athens, Greece, Jul. 2025

    2025

  59. [59]

    InputSnatch: Stealing input in LLM services via cache-sharing timing side-channel attacks,

    X. Zheng, H. Han, S. Shi, Q. Fang, Z. Du, X. Hu, and Q. Guo, “InputSnatch: Stealing input in LLM services via cache-sharing timing side-channel attacks,” Presented at the 16th International Symposium on Advanced Parallel Processing Technologies (APPT), Athens, Greece, Jul. 2025

    2025

  60. [60]

    MV-BSD: Multi-variable speculation diagrams for compact automated logic design,

    X. Guan, S. Cheng, R. Zhang, and Z. Du, “MV-BSD: Multi-variable speculation diagrams for compact automated logic design,” Presented at the 16th International Symposium on Advanced Parallel Processing Technologies (APPT), Athens, Greece, Jul. 2025

    2025

  61. [61]

    COMET: An FP32 matrix multiplication accelerator by extending INT8-based arrays,

    R. Xia, Y . Hao, Y . Zhao, Y . Liu, Y . Zhao, Z. Du, X. Hu, W. Li, and Q. Guo, “COMET: An FP32 matrix multiplication accelerator by extending INT8-based arrays,” Presented at the 16th International Symposium on Advanced Parallel Processing Technologies (APPT), Athens, Greece, Jul. 2025

    2025

  62. [62]

    Lightgaussian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps,

    Z. Fan, K. Wang, K. Wen, Z. Zhu, D. Xu, Z. Wanget al., “Lightgaussian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps,” inNeurIPS, 2025

    2025

  63. [63]

    Compgs: Smaller and faster gaussian splatting with vector quantiza- tion,

    K. Navaneet, K. P. Meibodi, S. A. Koohpayegani, and H. Pirsiavash, “Compgs: Smaller and faster gaussian splatting with vector quantiza- tion,” inECCV, 2024

    2024

  64. [64]

    Mini-splatting: Representing scenes with a constrained number of gaussians,

    G. Fang and B. Wang, “Mini-splatting: Representing scenes with a constrained number of gaussians,” inECCV, 2024

    2024

  65. [65]

    Eagles: Efficient accelerated 3d gaussians with lightweight encodings,

    S. Girish, K. Gupta, and A. Shrivastava, “Eagles: Efficient accelerated 3d gaussians with lightweight encodings,” inECCV, 2024

    2024

  66. [66]

    Reducing the memory footprint of 3d gaussian splatting,

    P. Papantonakis, G. Kopanas, B. Kerbl, A. Lanvin, and G. Drettakis, “Reducing the memory footprint of 3d gaussian splatting,” inPACM- CGIT, 2024

    2024

  67. [67]

    Potamoi: Accelerating neural rendering via a unified streaming architecture,

    Y . Feng, W. Lin, Z. Liu, J. Leng, M. Guo, H. Zhao, X. Hou, J. Zhao, and Y . Zhu, “Potamoi: Accelerating neural rendering via a unified streaming architecture,” inTACO, 2024

    2024

  68. [68]

    Cicero: Addressing algorithmic and architectural bottlenecks in neural rendering by radiance warping and memory optimizations,

    Y . Feng, Z. Liu, J. Leng, M. Guo, and Y . Zhu, “Cicero: Addressing algorithmic and architectural bottlenecks in neural rendering by radiance warping and memory optimizations,” inISCA, 2024

    2024

  69. [69]

    Déja view: Spatio-temporal compute reuse for energy-efficient 360 vr video streaming,

    S. Zhao, H. Zhang, S. Bhuyan, C. S. Mishra, Z. Ying, M. T. Kandemir, A. Sivasubramaniam, and C. R. Das, “Déja view: Spatio-temporal compute reuse for energy-efficient 360 vr video streaming,” inISCA, 2020

    2020

  70. [70]

    Euphrates: Algorithm-soc co-design for low-power mobile continuous vision,

    Y . Zhu, A. Samajdar, M. Mattina, and P. Whatmough, “Euphrates: Algorithm-soc co-design for low-power mobile continuous vision,” in ISCA, 2018

    2018

  71. [71]

    Cova: Exploiting compressed-domain analysis to accelerate video analytics,

    J. Hwang, M. Kim, D. Kim, S. Nam, Y . Kim, D. Kim, H. Sharma, and J. Park, “Cova: Exploiting compressed-domain analysis to accelerate video analytics,” inATC, 2022

    2022

  72. [72]

    Déjà vu: Efficient video-language query engine with learning-based inter-frame computation reuse,

    J. Hwang, D. Kim, S. Lee, Y . Kim, G. Heo, H. Kim, Y . Jeong, T. Meaza, E. Park, J. Ahn, and J. Park, “Déjà vu: Efficient video-language query engine with learning-based inter-frame computation reuse,” inPVLDB, 2025

    2025

  73. [73]

    Balanced 3dgs: Gaussian-wise parallelism rendering with fine-grained tiling,

    H. Gui, L. Hu, R. Chen, M. Huang, Y . Yin, J. Yang, Y . Wu, C. Liu, Z. Sun, X. Zhanget al., “Balanced 3dgs: Gaussian-wise parallelism rendering with fine-grained tiling,”arXiv preprint arXiv:2412.17378, 2024

    Pith/arXiv arXiv 2024