Dense PFT / HKD Inf for an 8K Graphics Engine: Measured Speed, Numerical Accuracy, and Disclosure Boundaries

We examine two recorded experiments relevant to the proposed dense PFT/HKD Inf 8K graphics engine: a structured dense-training contraction and a matched 8K image-warp benchmark. On a 576-parameter polynomial network, five-node weighted evaluation reduced the median time for twenty dense Adam updates by factors of 1.288, 5.719, and 30.905 relative to full sample evaluation. Against a polynomial-moment implementation the factors were 1.385–1.411. Maximum final-parameter discrepancies were below 2.3 × 10−16. Separately, the 8K PFT JPEG path achieved 1.186 times the speed of direct dense rendering but 0.945 times that of optimized separable dense rendering. These are distinct workloads: the training result is not an end-to-end 8K Renderdon speedup. A self-contained Python appendix reproduces every table from the recorded observations.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-13
DOI
https://doi.org/10.5281/zenodo.24197307
Primary Topic
Parallel Computing and Optimization Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Dense PFT / HKD Inf for an 8K Graphics Engine: Measured Speed, Numerical Accuracy, and Disclosure Boundaries

Michael Yang
3 links
Zenodo (CERN European Organization for Nuclear Research)
Parallel Computing and Optimization Techniques
article

Dense PFT / HKD Inf for an 8K Graphics Engine: Measured Speed, Numerical Accuracy, and Disclosure Boundaries

Michael Yang
article en

Abstract

We examine two recorded experiments relevant to the proposed dense PFT/HKD Inf 8K graphics engine: a structured dense-training contraction and a matched 8K image-warp benchmark. On a 576-parameter polynomial network, five-node weighted evaluation reduced the median time for twenty dense Adam updates by factors of 1.288, 5.719, and 30.905 relative to full sample evaluation. Against a polynomial-moment implementation the factors were 1.385–1.411. Maximum final-parameter discrepancies were below 2.3 × 10−16. Separately, the 8K PFT JPEG path achieved 1.186 times the speed of direct dense rendering but 0.945 times that of optimized separable dense rendering. These are distinct workloads: the training result is not an end-to-end 8K Renderdon speedup. A self-contained Python appendix reproduces every table from the recorded observations.

Zenodo (CERN European Organization for Nuclear Research)
Parallel Computing and Optimization Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon