Pith. sign in

REVIEW 3 major objections 5 minor 65 references

VersaQ-3D: Architecture Support for Visual Geometry Grounded Transformers via Versatile Quantization

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A calibration-free quantization pipeline and a reconfigurable accelerator claim to bring billion-parameter feed-forward 3D reconstruction models down to 4-bit weights without retraining, while preserving 98-99% accuracy at W4A8.

desk verdict VersaQ-3D is a credible calibration-free PTQ plus real accelerator for VGGT, but the abstract overclaims and the transform normalization constants are left unspecified, so the accuracy claims need that gap closed before they are fully verifiable. read the letter →

arxiv 2601.20317 v2 pith:6MIR5JPU submitted 2026-01-28 cs.AR

Yipu Zhang , Jintao Cheng , Xingyu Liu , Zeyu Li , Carol Jingyi Li , Jin Wu , Lin Jiang , Ceyu Xu
show 3 more authors
Yuan Xie Jiang Xu Wei Zhang
This is my paper · ORCID
classification cs.AR
keywords 3DreconstructionVisualGeometryGroundedTransformerpost-trainingquantizationWalsh-Hadamardtransformdiscretecosinereconfigurableacceleratoralgorithm-architectureco-design4-bitinference
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

VersaQ-3D claims that the Visual Geometry Grounded Transformer (VGGT), a billion-parameter feed-forward model that predicts camera poses, depth maps, and point maps directly from images, can be quantized down to 4-bit weights and activations without any calibration data. The paper argues that LLM-style quantization fails on VGGT because its activations have saturated channels rather than isolated outliers, and because 3D scene semantics are too diverse for a small calibration set to represent. Its proposed fix is a transform-based quantization pipeline that folds a Walsh-Hadamard transform and a discrete cosine transform into the weights offline, making activations more uniform while preserving structural weight features. On the hardware side, a reconfigurable multi-precision accelerator executes BF16, INT8, and INT4 on a shared systolic datapath and uses two-stage recomputation-based tiling for long-sequence attention. If the paper is right, instant, scene-agnostic 3D reconstruction can run on a compact edge chip rather than a server GPU.

What carries the argument

The central object is the fused orthogonal-transform weight matrix: each projection weight $W$ is replaced by $H^\top \gamma W D$, where $H$ is the Walsh-Hadamard transform (a square matrix of $\pm 1$ entries that decorrelates activation outliers) and $D$ is an integer discrete cosine transform that preserves structural weight features. The identity $(XH)(H^\top W)=XW$ carries the algorithm's correctness: the transforms cancel mathematically, so the model's function is unchanged before quantization, and only the online inverse DCT costs chip cycles. The hardware argument is carried by three mechanisms: bit-fusion INT4 processing elements that compose into INT8 units and then into BF16-capable brain-float units; a two-stage recomputation-based tiling scheme that recomputes attention scores instead of caching them; and a quantization unit that shifts data between BF16, INT8, and INT4 modes.

What would settle it

Run the fused transform pair (WHT, DCT, inverse DCT) in full precision on random tensors using the exact constants from the implementation and measure the round-trip relative error; if it is not at machine-precision level, the computational-invariance claim is false. Separately, evaluate the W4A8 model on a held-out scene type not represented in Co3Dv2 or 7-Scenes; if pose-estimation AUC falls well below 98-99% of full precision, the calibration-free robustness claim fails.

Watch

Extended reading notes

Core claim

VGGT's activation distribution is the paper's central difficulty: unlike LLMs, which show isolated spiking outliers, VGGT has saturated channels where many values stay high across most of the percentile range, and its 3D semantics are too scene-dependent for a calibration set to be reliable. The claim is that a calibration-free, input-agnostic quantization can handle both: a Walsh-Hadamard transform (WHT) is applied to activations via the identity $(XH)(H^\top W)=XW$ to decorrelate saturated channels, while an integer discrete cosine transform (DCT) is applied to the weights to preserve their structural features under low-bit rounding. Both transforms are fused offline into the weight matrices as $W_{K,\text{final}} \leftarrow H^\top \gamma W_K D$, so the only on-chip transform cost is an inverse DCT. With this, the paper reports 98-99% of full-precision accuracy at W4A8, acceptable accuracy at W4A4 that beats round-to-nearest and QuaRot by 1.61x-2.39x across diverse scenes, and an accelerator that delivers 5.4-22.0x speedup over edge GPUs under iso-PE-area comparison while consuming 2.18 W.

Load-bearing premise

The load-bearing premise is that the integer discrete cosine transform computed on chip exactly cancels the transform folded into the weights offline, so the network's function is unchanged before any quantization is applied; if the scaling constants are not folded in correctly, the method inherits a systematic error that the reported accuracies do not measure.

Editorial extensions

If this is right

  • Feed-forward 3D reconstruction can move to edge devices: the W4A4 mode targets instant reconstruction and the W4A8 mode targets high fidelity, both running on the same reconfigurable chip.
  • Deployment no longer requires collecting calibration data, so a model can be quantized once and applied to arbitrary unseen scenes without per-dataset tuning.
  • Because the Hadamard transform is a matrix of $\pm 1$ entries, the accelerator computes it with adders rather than multipliers, keeping the quantization overhead small and unifying linear and nonlinear operators on one datapath.
  • The two-stage recomputation tiling reduces the on-chip memory footprint of global attention, so the speedup over the baseline grows as the number of input frames increases.
  • At 4-bit weights, weight memory drops to one quarter of the BF16 baseline, directly attacking the memory-capacity and bandwidth bottleneck that dominates VGGT latency on edge devices.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Inference: the WHT/DCT recipe is not 3D-specific in its mechanics, so it is a plausible calibration-free quantization recipe for other vision transformers whose activations saturate rather than spike; the paper does not claim this.
  • Inference: the bitwidth sensitivity results show weights staying stable down to 3 bits, which suggests an asymmetric W3A4 configuration could stretch the instant-reconstruction regime further than the paper tests.
  • Inference: the two-stage recomputation tiling is a general memory-versus-compute trade for any long-sequence attention whose score matrix exceeds on-chip capacity, not just multi-view 3D reconstruction.
  • Inference: a decisive test the paper leaves implicit is to run W4A4 on a scene category whose geometry is unlike anything in Co3Dv2 or 7-Scenes; the calibration-free claim predicts accuracy should not collapse, while a calibration-dependent method would degrade.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes VersaQ-3D, an algorithm-architecture co-design framework for quantized, on-device inference of the Visual Geometry Grounded Transformer (VGGT). On the algorithm side, it presents a calibration-free, input-agnostic post-training quantization scheme that applies offline-fused Walsh-Hadamard Transform (WHT) and Discrete Cosine Transform (DCT) to suppress activation saturation and preserve weight structure, targeting W4A8 and W4A4 precision. On the architecture side, it designs a reconfigurable multi-precision systolic accelerator supporting BF16/INT8/INT4 modes, with a two-stage recomputation-based tiling scheme for long-sequence global attention. Evaluation on Co3Dv2 and 7-Scenes reports accuracy close to full precision at W4A8, better W4A4 accuracy than RTN and QuaRot baselines, and speedups over Jetson edge GPUs.

Significance. If the transform-invariance and performance claims hold, this is a genuinely useful contribution: it identifies a real deployment bottleneck for VGGT, motivates a calibration-free quantization scheme with a plausible mechanism (WHT for activation smoothing, DCT for weight structure), validates it on two benchmarks with an ablation, and backs the hardware proposal with RTL synthesis and a cycle-level simulator. The W4A8 accuracy results in Tables I and II are mutually consistent and broadly support the 98–99%-of-full-precision claim when the appropriate higher-is-better metrics are used. The absence of fitted calibration constants is a strength of the algorithm. However, the paper currently lacks the normalization details needed to verify the claimed exact computational invariance of the fused transforms, and the abstract contains several speedup claims that are not supported by the body. These issues are load-bearing for the central claims and must be fixed before the paper can be accepted.

major comments (3)
  1. [§III-C, Eqs. (4)–(7); §IV-B and Fig. 7(b)] The computational-invariance step is the foundation of the calibration-free claim, but the scaling constants that make the transforms exact inverses are never specified. Equation (4) requires an orthogonal H with H H^T = I, while §IV-B states that the WHT is implemented by multiplying by ±1 with no multipliers; the unnormalized Hadamard matrix satisfies H H^T = n I, and the paper does not say whether the compensating 1/n factor is folded into W, into the activation scaling, or into the quantization/dequantization scales. Likewise, the 'integer DCT matrix in HEVC/H.265' is not an orthogonal matrix and is only specified for sizes 4/8/16/32, while VGGT uses 1024-dimensional features; the paper neither gives D for this size nor specifies the per-coefficient normalization that makes the on-chip IDCT the exact inverse. Until these constants are specified and verified (for example, by reporting the pre-quantization reconstruction error of the fused transform), the claim that the transformed network is functionally equivalent to the original before quantization is unverifiable, and the W4A8/W4A4 accuracy numbers could include a systematic transform error that Fig. 11 does not separate from quantization error.
  2. [Abstract vs. §V-C and §VI] The speedup claims in the abstract are not supported by the body. The abstract states 5.4×–22.0× speedup over edge GPUs and 2.2×–3.0× over prior quantization accelerators under iso-PE-area comparison, and a 77% end-to-end latency reduction, but §V-C reports only 2.7×–10.8× over the two Jetson baselines and never compares with any prior quantization accelerator. The full-text abstract says 5.2×–10.8×, which is yet another range, and the body's latency claim is 60% plus an additional 7% (Fig. 13), which does not add to 77%. The 'iso-PE-area' methodology is not defined, and no such baseline appears in any table or figure. Please correct the abstract or supply the missing comparison and methodology.
  3. [§V-B, Tables I–II] The '98–99% of full-precision accuracy' claim needs a precise definition. Table II contains lower-is-better metrics (Acc, Comp) and one higher-is-better metric (N.C.), and the paper does not state which metric or aggregation is used to compute '98%' and '99%'. For example, Ours W4A8 has Acc mean 0.0518 versus full-precision 0.0442 (117% of full-precision) and Comp mean 0.0618 versus 0.0567 (109%), while N.C. is 0.7217 versus 0.7321 (98.6%). Please specify the aggregation rule and report all metrics consistently. Similarly, the 'up to 2.39× higher accuracy' statement in §V-B is only for AUC@30 on Co3Dv2; the text should clarify that it does not apply to the 7-Scenes metrics in Table II.
minor comments (5)
  1. [Fig. 11 and §V-B] The figure labels (34.7% and 28.7%) do not match the caption and text values (29% and 35%). Please reconcile these numbers and clarify whether they are per-step relative improvements or average AUC gains.
  2. [Eqs. (8)–(10)] The notation M_i, Σ_i and M_i', Σ_i' is used without a clear final definition; please define the finalized running maximum and normalization sum before Eq. (10).
  3. [Algorithm 1] In Stage 1, the precision of the dequantized score S_{i,j} is not stated, while Stage 2 explicitly quantizes S; please specify the data type used to accumulate the softmax statistics.
  4. [Sec. V-A] The word 'offical' in the dataset description should be 'official'.
  5. [Table III] The 'BFU Units' row mixes the BFU count and the BFU buffer area; please clarify whether the 2.77 mm² entry includes the buffers and how the per-BFU area relates to the total area.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: VersaQ-3D's accuracy claims are empirical and its transform fusion is a fixed algebraic preprocessing; the main caveat is unstated transform scaling constants, which is a verifiability issue, not a circularity.

full rationale

VersaQ-3D's central claims are empirical measurements on Co3Dv2 and 7-Scenes, not outputs of a derivation that consumes the evaluation targets. The WHT/DCT weight fusion in Eq. (7) is a fixed, calibration-free algebraic preprocessing; no parameter is fitted to the reported accuracy numbers, and the ablation in Fig. 11 independently measures the contribution of each transform step. Eq. (4) is a standard orthogonality identity, not a prediction that reduces to its own inputs. The references to prior calibration-free quantization [58] and incoherence processing [54] are motivational, and even if [58] is a self-citation, it is not load-bearing: the accuracy and speedup results stand on the presented measurements and RTL/simulator evaluation. The one notable gap is mathematical verifiability rather than circularity: the HEVC integer DCT is not orthogonal without per-coefficient normalization, and the ±1 Hadamard implementation in Sec. IV-B omits the 1/sqrt(n) scaling of Eq. (1), so the exact-invertibility claim in Eqs. (4)-(7) is missing stated constants. If these constants are not folded into the fused weights or dequantization, the pre-quantization network may not be functionally equivalent; that is a correctness/omission concern, not a circular reduction of the prediction to its inputs.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The ledger shows that the central accuracy claim depends mainly on empirical properties of VGGT (saturated channels) and on an unstated assumption about the integer DCT's orthogonality. There are no invented physical entities. The free parameters are design choices (tile sizes, precision modes) plus the missing DCT scaling.

free parameters (3)
  • HEVC integer DCT scaling factors
    The paper states it uses the integer DCT matrix from HEVC/H.265 (Sec. III-A) to guarantee INT computation, but never gives the scaling constants required for D·D^T = I. The fused weight formula (Eq. 7) depends on D being exactly orthogonal; the missing scaling is an unstated parameter.
  • Attention tiling sizes T_Q=64, T_K=64, T_V=2048 = 64/64/2048
    Algorithm 1 hard-codes these tile sizes; they are hand-chosen to fit on-chip buffers and directly influence the reported 7% runtime reduction and the memory-traffic claims.
  • Precision mode policy (W4A8/W4A4) = W4A8 and W4A4
    The two bit-width settings are chosen to represent 'instant' and 'high-fidelity' modes; they are not fitted but define the scope of the accuracy claims.
assumptions (4)
  • standard math WHT and DCT orthogonal transform pairs preserve dot products exactly (HH^T = I, DD^T = I)
    Used in Eqs. (1)-(7) to justify computational invariance; standard linear algebra, but for the integer DCT this is only approximately true.
  • domain assumption VGGT has saturated activation channels and no small calibration set can represent 3D scene diversity
    Sec. II-D, Figs. 1 and 4; this motivates the entire calibration-free design and is based on the authors' profiling of the model.
  • domain assumption Quantization to INT4/INT8 after the transform preserves accuracy without calibration data
    The method assumes dynamic or fixed per-tensor ranges are sufficient; the paper never specifies how activation scales are derived, so the accuracy results rest on this unstated assumption.
  • ad hoc to paper HEVC integer DCT behaves as an exactly orthogonal transform in the fused computation
    Sec. III-C Eq. (7) treats D as orthogonal; the integer HEVC DCT is a scaled approximation, so this is an ad hoc idealization that the paper does not justify.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VersaQ-3D: Architecture Support for Visual Geometry Grounded Transformers via Versatile Quantization." pith.science (2026). https://pith.science/paper/6MIR5JPU

@misc{pith2026260120317,
  author       = {Pith},
  title        = {Pith review of: VersaQ-3D: Architecture Support for Visual Geometry Grounded Transformers via Versatile Quantization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6MIR5JPU}},
  note         = {Machine review of arXiv:2601.20317}
}
abstract

3D reconstruction and view synthesis are fundamental to AR/VR, robotics, and digital twins. The Visual Geometry Grounded Transformer (VGGT) enables strong feed-forward 3D reconstruction while its billion-parameter scale limits on-device deployment. LLM-oriented quantization methods fail on VGGT due to saturated activation channels that resist low-bit quantization and diverse 3D semantics that impede calibration. VGGT further poses hardware challenges from multi-precision architecture support and long-sequence global attention with excessive memory demands. We propose VersaQ-3D, an algorithm-architecture co-design framework for efficient VGGT inference. At the algorithm level, we present the first calibration-free, input-agnostic quantization method for VGGT, leveraging transform coding to suppress outliers and preserve structural weight features, enabling robust low-bit inference down to 4 bits. At the architecture level, we design a reconfigurable accelerator with a hierarchical multi-precision compute unit (BF16/INT8/INT4) that executes both linear and non-linear operators within a shared systolic datapath, reducing end-to-end latency by 77%. A two-stage recomputation-based tiling strategy further cuts runtime by 7% by alleviating on-chip memory pressure for long-sequence attention. Evaluations across various datasets show that VersaQ-3D incurs negligible accuracy loss at W4A8 and consistently achieves leading accuracy at W4A4 over prior quantization methods across diverse scenes. The co-designed accelerator delivers 5.4$\times$-22.0$\times$ speedup over edge GPUs and 2.2$\times$-3.0$\times$ over prior quantization-based accelerators under iso-PE-area comparison, enabling instant and energy-efficient feed-forward 3D reconstruction on edge devices.

Figures

Figures reproduced from arXiv: 2601.20317 by the authors.

Figure 1
Figure 1. Llama and VGGT activation value distributions and corresponding [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. VGGT model structure. A DINO-based feature extractor feeds an [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Inference runtime breakdown on (a) different GPUs with S=3 [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Salient distribution in VGGT. Here we take channel variance to [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: VersaQ-3D quantization framework. The attention module (with MLP blocks treated analogously) follows a four-stage pipeline that combines offline [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: VersaQ-3D offline weight preparation. Each weight matrix is trans [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: VersaQ-3D hardware architecture. (a) Overall hardware architecture. The accelerator works as output stationary systolic array during INT8 and INT4 [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Proposed tiling method to alleviate the memory bottleneck caused [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Point map reconstruction result on Fire/7-Scenes. [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Bitwidth sensitivity study on 7-Scenes, where lower accuracy means [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: Ablation of VersaQ-3D on Co3Dv2 (W4A4). WHT alone provides [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]
Figure 12
Figure 12. Figure 12: The normalized speedup and energy efficiency achieved by the proposed VersaQ-3D over baseline devices on Co3Dv2 and 7-Scenes dataset. [PITH_FULL_IMAGE:figures/full_fig_p011_12.png]
Figure 13
Figure 13. Figure 13: Runtime breakdown on Co3Dv2 (W4A4). Our quantized VGGT [PITH_FULL_IMAGE:figures/full_fig_p011_13.png]
Figure 14
Figure 14. Figure 14: Normalized speedup vs. frame count S on Co3Dv2. VersaQ-3D consistently outperforms the baseline across all S, with the largest speedup at S=1 where the memory bottleneck dominates and reduced off-chip access yields greater gains. Ablation Study and Robustness Anaylsis…

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 33 canonical work pages

  1. [1]

    Ahmed and K

    N. Ahmed and K. R. Rao,Orthogonal transforms for digital signal processing. Springer Science & Business Media, 2012

    work page 2012

  2. [2]

    Quarot: Outlier-free 4-bit inference in rotated llms,

    S. Ashkboos, A. Mohtashami, M. L. Croci, B. Li, P. Cameron, M. Jaggi, D. Alistarh, T. Hoefler, and J. Hensman, “Quarot: Outlier-free 4-bit inference in rotated llms,”Advances in Neural Information Processing Systems, vol. 37, pp. 100 213–100 240, 2024

    2024

  3. [3]

    Human-robot perception in industrial environments: A survey,

    A. Bonci, P. D. Cen Cheng, M. Indri, G. Nabissi, and F. Sibona, “Human-robot perception in industrial environments: A survey,”Sensors, vol. 21, no. 5, p. 1571, 2021

    2021

  4. [4]

    Diannao: a small-footprint high-throughput accelerator for ubiquitous machine-learning,

    T. Chen, Z. Du, N. Sun, J. Wang, C. Wu, Y . Chen, and O. Temam, “Diannao: a small-footprint high-throughput accelerator for ubiquitous machine-learning,” inProceedings of the 19th international conference on Architectural support for programming languages and operating systems, 2014, pp. 269–284

    work page 2014

  5. [5]

    Eyeriss: A spatial architecture for energy-efficient dataflow for convolutional neural networks,

    Y .-H. Chen, J. Emer, and V . Sze, “Eyeriss: A spatial architecture for energy-efficient dataflow for convolutional neural networks,” in 2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA). IEEE, 2016, pp. 367–379

    work page 2016

  6. [6]

    Bitmod: Bit-serial mixture-of- datatype llm acceleration,

    Y . Chen, A. F. AbouElhamayed, X. Dai, Y . Wang, M. Andronic, G. A. Constantinides, and M. S. Abdelfattah, “Bitmod: Bit-serial mixture-of- datatype llm acceleration,” in2025 IEEE International Symposium on High Performance Computer Architecture (HPCA). IEEE, 2025, pp. 1082–1097

    work page 2025

  7. [7]

    Nvidia a100 gpu: Performance & innova- tion for gpu computing,

    J. Choquette and W. Gandhi, “Nvidia a100 gpu: Performance & innova- tion for gpu computing,” in2020 IEEE Hot Chips 32 Symposium (HCS). IEEE Computer Society, 2020, pp. 1–43

    work page 2020

  8. [8]

    Flashattention: Fast and memory-efficient exact attention with io-awareness,

    T. Dao, D. Fu, S. Ermon, A. Rudra, and C. R ´e, “Flashattention: Fast and memory-efficient exact attention with io-awareness,”Advances in neural information processing systems, vol. 35, pp. 16 344–16 359, 2022

    2022

Show all 65 references
  1. [9]

    A systematic review of a digital twin city: A new pattern of urban governance toward smart cities,

    T. Deng, K. Zhang, and Z.-J. M. Shen, “A systematic review of a digital twin city: A new pattern of urban governance toward smart cities,” Journal of management science and engineering, vol. 6, no. 2, pp. 125– 134, 2021

    2021

  2. [10]

    Gpt3. int8 (): 8-bit matrix multiplication for transformers at scale,

    T. Dettmers, M. Lewis, Y . Belkada, and L. Zettlemoyer, “Gpt3. int8 (): 8-bit matrix multiplication for transformers at scale,”Advances in neural information processing systems, vol. 35, pp. 30 318–30 332, 2022

    2022

  3. [11]

    Nvidia orin system-on-chip,

    M. Ditty, “Nvidia orin system-on-chip,” in2022 IEEE Hot Chips 34 Symposium (HCS). IEEE Computer Society, 2022, pp. 1–17

    2022

  4. [12]

    Nvidia’s xavier soc,

    M. Ditty, A. Karandikar, and D. Reed, “Nvidia’s xavier soc,” inHot chips: a symposium on high performance chips, 2018

    2018

  5. [13]

    Quantized visual geometry grounded transformer,

    W. Feng, H. Qin, M. Wu, C. Yang, Y . Li, X. Li, Z. An, L. Huang, Y . Zhang, M. Magnoet al., “Quantized visual geometry grounded transformer,”arXiv preprint arXiv:2509.21302, 2025

    2025

  6. [14]

    Lumina: Real-time mobile neural rendering by exploiting computational redundancy,

    Y . Feng, W. Lin, Y . Cheng, Z. Liu, J. Leng, M. Guo, C. Chen, S. Sun, and Y . Zhu, “Lumina: Real-time mobile neural rendering by exploiting computational redundancy,”arXiv preprint arXiv:2506.05682, 2025

    2025 arXiv

  7. [15]

    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,” in2024 ACM/IEEE 51st Annual International Symposium on Computer Architecture (ISCA). IEEE, 2024, pp....

    2024

  8. [16]

    Olive: Accelerating large language models via hardware- friendly outlier-victim pair quantization,

    C. Guo, J. Tang, W. Hu, J. Leng, C. Zhang, F. Yang, Y . Liu, M. Guo, and Y . Zhu, “Olive: Accelerating large language models via hardware- friendly outlier-victim pair quantization,” inProceedings of the 50th Annual International Symposium on Computer Architecture, 2023, pp. 1–15

    2023

  9. [17]

    Eie: Efficient inference engine on compressed deep neural network,

    S. Han, X. Liu, H. Mao, J. Pu, A. Pedram, M. A. Horowitz, and W. J. Dally, “Eie: Efficient inference engine on compressed deep neural network,” in2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA). IEEE Computer Society, 2016, pp. 243–254

    2016

  10. [18]

    Hartley and A

    R. Hartley and A. Zisserman,Multiple view geometry in computer vision. Cambridge university press, 2003

    2003

  11. [19]

    ITU-T Recommendation H.265: High Efficiency Video Coding,

    International Telecommunication Union, “ITU-T Recommendation H.265: High Efficiency Video Coding,” ITU-T, Tech. Rep., 2023, [Online; accessed 2025-11-16]. [Online]. Available: https://www.itu.int/ rec/T-REC-H.265

    2023

  12. [20]

    JESD209-4: Low Power Double Data Rate 4 (LPDDR4),

    JEDEC Solid State Technology Association, “JESD209-4: Low Power Double Data Rate 4 (LPDDR4),” JEDEC Standard, 2014, jEDEC Publication, Arlington, V A, USA

    2014

  13. [21]

    JESD235D: High Bandwidth Memory (HBM) DRAM,

    ——, “JESD235D: High Bandwidth Memory (HBM) DRAM,” JEDEC Standard, 2021, jEDEC Publication, Arlington, V A, USA

    2021

  14. [22]

    JESD209-5C: Low Power Double Data Rate (LPDDR5/5X),

    ——, “JESD209-5C: Low Power Double Data Rate (LPDDR5/5X),” JEDEC Standard, 2023, jEDEC Publication, Arlington, V A, USA

    2023

  15. [23]

    JESD238B.01: High Bandwidth Memory (HBM3) DRAM,

    ——, “JESD238B.01: High Bandwidth Memory (HBM3) DRAM,” JEDEC Standard, 2025, jEDEC Publication, Arlington, V A, USA

    2025

  16. [24]

    Large scale multi-view stereopsis evaluation,

    R. Jensen, A. Dahl, G. V ogiatzis, E. Tola, and H. Aanæs, “Large scale multi-view stereopsis evaluation,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2014, pp. 406–413

    2014

  17. [25]

    Stripes: Bit-serial deep neural network computing,

    P. Judd, J. Albericio, T. Hetherington, T. M. Aamodt, and A. Moshovos, “Stripes: Bit-serial deep neural network computing,” in2016 49th Annual IEEE/ACM International Symposium on Microarchitecture (MI- CRO). IEEE, 2016, pp. 1–12. 12

    2016

  18. [26]

    3d gaussian splatting for real-time radiance field rendering

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

    2023

  19. [27]

    The discrete cosine transform (dct): theory and appli- cation,

    S. A. Khayam, “The discrete cosine transform (dct): theory and appli- cation,”Michigan State University, vol. 114, no. 1, p. 31, 2003

    2003

  20. [28]

    Tender: Accelerating large language models via tensor decomposition and runtime requantization,

    J. Lee, W. Lee, and J. Sim, “Tender: Accelerating large language models via tensor decomposition and runtime requantization,” in2024 ACM/IEEE 51st Annual International Symposium on Computer Archi- tecture (ISCA). IEEE, 2024, pp. 1048–1062

    2024

  21. [29]

    Neurex: A case for neural rendering acceleration,

    J. Lee, K. Choi, J. Lee, S. Lee, J. Whangbo, and J. Sim, “Neurex: A case for neural rendering acceleration,” inProceedings of the 50th Annual International Symposium on Computer Architecture, 2023, pp. 1–13

    2023

  22. [30]

    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, Volu...

    2024

  23. [31]

    Rt-nerf: Real-time on-device neural radiance fields towards immersive ar/vr rendering,

    C. Li, S. Li, Y . Zhao, W. Zhu, and Y . Lin, “Rt-nerf: Real-time on-device neural radiance fields towards immersive ar/vr rendering,” inProceed- ings of the 41st IEEE/ACM International Conference on Computer-Aided Design, 2022, pp. 1–9

    2022

  24. [32]

    Mbq: Modality-balanced quantization for large vision- language models,

    S. Li, Y . Hu, X. Ning, X. Liu, K. Hong, X. Jia, X. Li, Y . Yan, P. Ran, G. Daiet al., “Mbq: Modality-balanced quantization for large vision- language models,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 4167–4177

    2025

  25. [33]

    Instant-3d: Instant neural radiance field training towards on-device ar/vr 3d reconstruction,

    S. Li, C. Li, W. Zhu, B. Yu, Y . Zhao, C. Wan, H. You, H. Shi, and Y . Lin, “Instant-3d: Instant neural radiance field training towards on-device ar/vr 3d reconstruction,” inProceedings of the 50th Annual International Symposium on Computer Architecture, 2023, pp. 1–13

    2023

  26. [34]

    Awq: Activation-aware weight quanti- zation for on-device llm compression and acceleration,

    J. Lin, J. Tang, H. Tang, S. Yang, W.-M. Chen, W.-C. Wang, G. Xiao, X. Dang, C. Gan, and S. Han, “Awq: Activation-aware weight quanti- zation for on-device llm compression and acceleration,”Proceedings of machine learning and systems, vol. 6, pp. 87–100, 2024

    2024

  27. [35]

    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

  28. [36]

    Flightvgm: Efficient video generation model inference with online sparsification and hybrid precision on fpgas,

    J. Liu, S. Zeng, L. Ding, W. Soedarmadji, H. Zhou, Z. Wang, J. Li, J. Li, Y . Dai, K. Wenet al., “Flightvgm: Efficient video generation model inference with online sparsification and hybrid precision on fpgas,” in Proceedings of the 2025 ACM/SIGDA International Symposium on Fi...

    2025

  29. [37]

    Hq-dit: Efficient diffusion transformer with fp4 hybrid quantization,

    W. Liu and S. Q. Zhang, “Hq-dit: Efficient diffusion transformer with fp4 hybrid quantization,”arXiv preprint arXiv:2405.19751, 2024

    2024 arXiv

  30. [38]

    Spinquant: Llm quantization with learned rotations,

    Z. Liu, C. Zhao, I. Fedorov, B. Soran, D. Choudhary, R. Krishnamoorthi, V . Chandra, Y . Tian, and T. Blankevoort, “Spinquant: Llm quantization with learned rotations,”arXiv preprint arXiv:2405.16406, 2024

    2024 arXiv

  31. [39]

    Fast inverse square root,

    C. Lomont, “Fast inverse square root,”Tech-315 nical Report, vol. 32, pp. 44–46, 2003

    2003

  32. [40]

    Ramulator 2.0: A modern, modular, and extensible dram simulator,

    H. Luo, Y . C. Tu ˘grul, F. N. Bostancı, A. Olgun, A. G. Ya ˘glıkc ¸ı, and O. Mutlu, “Ramulator 2.0: A modern, modular, and extensible dram simulator,”IEEE Computer Architecture Letters, vol. 23, no. 1, pp. 112– 116, 2023

    2023

  33. [41]

    Digital twins: A survey on enabling technologies, challenges, trends and future prospects,

    S. Mihai, M. Yaqoob, D. V . Hung, W. Davis, P. Towakel, M. Raza, M. Karamanoglu, B. Barn, D. Shetve, R. V . Prasadet al., “Digital twins: A survey on enabling technologies, challenges, trends and future prospects,”IEEE Communications Surveys & Tutorials, vol. 24, no. 4, pp. 22...

    2022

  34. [42]

    Nerf: Representing scenes as neural radiance fields for view synthesis,

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,”Communications of the ACM, vol. 65, no. 1, pp. 99–106, 2021

    2021

  35. [43]

    Response time in man-computer conversational transac- tions,

    R. B. Miller, “Response time in man-computer conversational transac- tions,” inProceedings of the December 9-11, 1968, fall joint computer conference, part I, 1968, pp. 267–277

    1968

  36. [44]

    Instant neural graphics primitives with a multiresolution hash encoding,

    T. M ¨uller, A. Evans, C. Schied, and A. Keller, “Instant neural graphics primitives with a multiresolution hash encoding,”ACM transactions on graphics (TOG), vol. 41, no. 4, pp. 1–15, 2022

    2022

  37. [45]

    A study on tolerable waiting time: how long are web users willing to wait?

    F. F.-H. Nah, “A study on tolerable waiting time: how long are web users willing to wait?”Behaviour & Information Technology, vol. 23, no. 3, pp. 153–163, 2004

    2004

  38. [46]

    Dinov2: Learning robust visual features without supervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V . V o, M. Szafraniec, V . Khali- dov, P. Fernandez, D. Haziza, F. Massa, A. El-Noubyet al., “Dinov2: Learning robust visual features without supervision,”Trans. Mach. Learn. Res., 2024

    2024

  39. [47]

    Pytorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antigaet al., “Pytorch: An imperative style, high-performance deep learning library,”Advances in neural information processing systems, vol. 32, 2019

    2019

  40. [48]

    Hadamard transform image coding,

    W. K. Pratt, J. Kane, and H. C. Andrews, “Hadamard transform image coding,”Proceedings of the IEEE, vol. 57, no. 1, pp. 58–68, 1969

    1969

  41. [49]

    Common objects in 3d: Large-scale learning and evaluation of real-life 3d category reconstruction,

    J. Reizenstein, R. Shapovalov, P. Henzler, L. Sbordone, P. Labatut, and D. Novotny, “Common objects in 3d: Large-scale learning and evaluation of real-life 3d category reconstruction,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 10 901–10 911

    2021

  42. [50]

    Structure-from-motion revisited,

    J. L. Schonberger and J.-M. Frahm, “Structure-from-motion revisited,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 4104–4113

    2016

  43. [51]

    Bit fusion: Bit-level dynamically composable architecture for accelerating deep neural network,

    H. Sharma, J. Park, N. Suda, L. Lai, B. Chau, J. K. Kim, V . Chandra, and H. Esmaeilzadeh, “Bit fusion: Bit-level dynamically composable architecture for accelerating deep neural network,” in2018 ACM/IEEE 45th Annual International Symposium on Computer Architecture (ISCA). IEE...

    2018

  44. [52]

    Scene coordinate regression forests for camera relocalization in rgb-d images,

    J. Shotton, B. Glocker, C. Zach, S. Izadi, A. Criminisi, and A. Fitzgib- bon, “Scene coordinate regression forests for camera relocalization in rgb-d images,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2013, pp. 2930–2937

    2013

  45. [53]

    Llama 2: Open foundation and fine-tuned chat models,

    H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y . Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosaleet al., “Llama 2: Open foundation and fine-tuned chat models,”arXiv preprint arXiv:2307.09288, 2023

    2023 arXiv

  46. [54]

    Quip#: Even better llm quantization with hadamard incoherence and lattice codebooks,

    A. Tseng, J. Chee, Q. Sun, V . Kuleshov, and C. De Sa, “Quip#: Even better llm quantization with hadamard incoherence and lattice codebooks,”arXiv preprint arXiv:2402.04396, 2024

    2024 arXiv

  47. [55]

    Vggt: Visual geometry grounded transformer,

    J. Wang, M. Chen, N. Karaev, A. Vedaldi, C. Rupprecht, and D. Novotny, “Vggt: Visual geometry grounded transformer,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 5294– 5306

    2025

  48. [56]

    Tataa: Programmable mixed-precision transformer acceleration with a trans- formable arithmetic architecture,

    J. Wu, M. Song, J. Zhao, Y . Gao, J. Li, and H. K.-H. So, “Tataa: Programmable mixed-precision transformer acceleration with a trans- formable arithmetic architecture,”ACM Transactions on Reconfigurable Technology and Systems, vol. 18, no. 1, pp. 1–31, 2025

    2025

  49. [57]

    Smoothquant: Accurate and efficient post-training quantization for large language models,

    G. Xiao, J. Lin, M. Seznec, H. Wu, J. Demouth, and S. Han, “Smoothquant: Accurate and efficient post-training quantization for large language models,” inInternational conference on machine learning. PMLR, 2023, pp. 38 087–38 099

    2023

  50. [58]

    Llm. 265: Video codecs are secretly tensor codecs,

    C. Xu, Y . Wu, X. Yang, B. Chen, M. Lentz, D. Zhuo, and L. W. Wills, “Llm. 265: Video codecs are secretly tensor codecs,” inProceedings of the 58th IEEE/ACM International Symposium on Microarchitecture, 2025, pp. 445–460

    2025

  51. [59]

    Advances in feed-forward 3d reconstruction and view synthesis: A survey,

    J. Zhang, Y . Li, A. Chen, M. Xu, K. Liu, J. Wang, X.-X. Long, H. Liang, Z. Xu, H. Suet al., “Advances in feed-forward 3d reconstruction and view synthesis: A survey,”arXiv preprint arXiv:2507.14501, 2025

    2025

  52. [60]

    D ´eja 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, “D ´eja view: Spatio-temporal compute reuse for ‘energy-efficient 360 vr video streaming,” in2020 ACM/IEEE 47th Annual International Symposium on Computer Archi- tecture (IS...

    2020

  53. [61]

    Vidit-q: Efficient and accurate quantization of diffusion transformers for image and video generation,

    T. Zhao, T. Fang, H. Huang, E. Liu, R. Wan, W. Soedarmadji, S. Li, Z. Lin, G. Dai, S. Yanet al., “Vidit-q: Efficient and accurate quantization of diffusion transformers for image and video generation,”arXiv preprint arXiv:2406.02540, 2024

    2024 arXiv

  54. [62]

    Mixdq: Memory-efficient few-step text-to-image diffu- sion models with metric-decoupled mixed precision quantization,

    T. Zhao, X. Ning, T. Fang, E. Liu, G. Huang, Z. Lin, S. Yan, G. Dai, and Y . Wang, “Mixdq: Memory-efficient few-step text-to-image diffu- sion models with metric-decoupled mixed precision quantization,” in European Conference on Computer Vision. Springer, 2024, pp. 285– 302

    2024

  55. [63]

    Unsupervised learning of depth and ego-motion from video,

    T. Zhou, M. Brown, N. Snavely, and D. G. Lowe, “Unsupervised learning of depth and ego-motion from video,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1851–1858

    2017

  56. [64]

    Megascale-infer: Efficient mixture- of-experts model serving with disaggregated expert parallelism,

    R. Zhu, Z. Jiang, C. Jin, P. Wu, C. A. Stuardo, D. Wang, X. Zhang, H. Zhou, H. Wei, Y . Chenget al., “Megascale-infer: Efficient mixture- of-experts model serving with disaggregated expert parallelism,” in 13 Proceedings of the ACM SIGCOMM 2025 Conference, 2025, pp. 592– 608

    2025

  57. [65]

    Nice-slam: Neural implicit scalable encoding for slam,

    Z. Zhu, S. Peng, V . Larsson, W. Xu, H. Bao, Z. Cui, M. R. Oswald, and M. Pollefeys, “Nice-slam: Neural implicit scalable encoding for slam,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 12 786–12 796. 14

    2022

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.