Skip to content

Releases: OriginNeuralAI/DSC3-DWave-Comparison-2026

Release list

v0.15.2-paper — Zenodo DOI minted

v0.15.2-paper — Zenodo DOI minted Latest
Latest

Choose a tag to compare

View all tags
@OriginNeuralAI OriginNeuralAI released this 15 May 00:54
v0.15.2-paper
2b3dcb1

Zenodo-anchored release

Zenodo DOI: 10.5281/zenodo.20192275

Same paper content as v0.15.1-paper, now with permanent identifier and full citation infrastructure.

What changed since v0.15.1-paper

  • Zenodo deposit published with DOI 10.5281/zenodo.20192275
  • Paper title page now shows the DOI alongside the GitHub URL and live-demo link
  • README header gained a DOI badge
  • CITATION.cff includes doi: and identifiers: fields
  • BibTeX in README updated to @misc{...} with doi and publisher = {Zenodo} fields
  • CHANGELOG documents the DOI mint

How to cite

@misc{daugherty_ward_ryan_dsc3_dwave_2026,
  author    = {Bryan W. Daugherty and Gregory Ward and Shawn Ryan},
  title     = {A Reproducible Classical Reference for D-Wave Advantage2's
               2024--2026 Industrial Benchmarks: Ground-state 3D \$\pm J\$
               Ising at \$N=10^{6}\$ Spins on a \\$1.57/Hour GPU Droplet},
  publisher = {Zenodo},
  year      = {2026},
  month     = may,
  doi       = {10.5281/zenodo.20192275},
  url       = {https://doi.org/10.5281/zenodo.20192275}
}

Links

Authors

Bryan W. Daugherty, Gregory Ward, Shawn Ryan — Origin Neural — https://originneural.ai

Assets 2
Loading

v0.15.1-paper — Initial release

v0.15.1-paper — Initial release

Choose a tag to compare

View all tags
@OriginNeuralAI OriginNeuralAI released this 15 May 00:38
v0.15.1-paper
c80a8a9

DSC-3 vs D-Wave Advantage2 — initial public release

A 40-page paper + reproducible artefacts reproducing D-Wave Advantage2's 2024–2026 published industrial benchmarks on a classical 16-solver ensemble running on a single $1.57/hour cloud GPU droplet.

Headline findings

Axis D-Wave Advantage2 DSC-3 (this work) Ratio
Max embeddable problem 4,400 qubits 1,000,000 (droplet, n=4) ~227×
Hardware capex / hourly $10–15M list $1.57/hour droplet 10⁴–10⁵×
$/solve at N=1,728 $0.05–$1.30 (Leap floor) $0.024 10²–10⁵×
Energy / solve derived ~200 Wh 4.6 Wh 42×
3D EA quality vs Hartmann sampling, not GS ±1% at L≤40 production preset matched
MaxCut Δ vs SA-only at N=10,000 not embeddable +0.13–0.20% (σ ≤ 0.02%, n=3) DSC-3 only
Cryptanalysis verticals (SHA/AES/RSA/GNFS) no publication production encoders DSC-3 only

Reproducibility

  • 12 load-bearing JSON artefacts with SHA-256 manifest (Appendix E of main.pdf)
  • aggregate_results.py + make_plots.py regenerate tables and figures from the JSON
  • run_*.sh / run_*.bat scripts contain the exact commands used on the droplet + workstation
  • Compute-intensity-matched SA-only baseline definition in §2 of main.pdf

Files

  • main.pdf — the paper, 40 pages, 9 figures, 22 tables
  • main.tex — full LaTeX source
  • results/*.json — every measured datapoint
  • figures/*.pdf + figures/png/*.png — plotted figures + README previews
  • tables/*.tex — generated LaTeX tables
  • CHANGELOG.md, CITATION.cff, LICENSE (CC-BY-4.0)

Authors

Bryan W. Daugherty, Gregory Ward, Shawn Ryan — Origin Neural — https://originneural.ai

What this paper does not claim

  • That classical methods invalidate D-Wave's sampling-class quantum supremacy demonstration (we solve a different problem)
  • That DSC-3 outperforms every classical algorithm at every scale (Concorde / LKH3 / Goemans–Williamson dominate their specialised classes)
  • That this paper exhausts the relevant benchmarks (four of five SCM verticals, reverse annealing, and the LLM-training drug-discovery pipeline remain future work)
Loading