Performance Benchmarks

Memory & energy efficiency
at near-parity accuracy

PNN replaces dense weight matrices with a patented connectivity configuration — cutting parameters 1.7–4.5× while holding accuracy within ~0.4 pp on vision and ~1.1 pp on tabular. On a Waymo-class BEV occupancy task the savings grow with width — up to 6.7× at parity accuracy. Because the pattern is index-free, INT8 models are up to 18× smaller than dense FP32 and stay lossless on vision.

1.7–6.7×

fewer parameters than dense

18×

smaller INT8 model (vision, lossless)

≤0.4 pp

accuracy gap on vision tasks

~12 nJ

per inference on the PNN chip (est)

Results

Summary — 5 seeds, iso-architecture

Task	Metric	Dense	PNN	PNN int8	Savings
MNIST	acc % ↑	97.45	97.02	97.00	4.5× fewer · 18× smaller int8
Fashion	acc % ↑	88.57	88.17	88.14	4.5× · 18×
Tabular*	acc % ↑	88.71	87.65	87.38	2.4× · 9.6×
NLP-GPT	val loss ↓	2.268	2.387	—	1.7× · 6.9×
Waymo BEV occ.†	acc % ↑	93.35	93.71	—	6.7× fewer (width 4096)
Waymo real LiDAR‡	acc % ↑	92.92	92.92	—	6.67× (width 4096)
microGPT	op-bound speed	1.0×	1.59× faster	—	scalar / no-BLAS

↑ higher = better, ↓ lower = better. NLP metric is validation cross-entropy loss. * Tabular = recommended fc2-only config. † Waymo-class BEV occupancy (synthetic), 3-seed, width 4096 — savings grow with width (below). ‡ Real Waymo Open Dataset v2 LiDAR, occupancy forecast +0.5 s: dense and PNN identical to every printed digit — exact parity. INT8 lossless on vision.

Scaling — perception

Savings grow with width

The connectivity gets structurally sparser as the layers widen, so on a Waymo-class BEV occupancy task the parameter savings rise from 4.0× to 6.7× as width goes 512 → 4096 — while accuracy stays at parity or better (negative gap = PNN ahead). This is the full-scale story: the wider the model, the more PNN saves.

Validated on real Waymo LiDAR: on real Open Dataset v2 point clouds (occupancy forecast, +0.5 s, ego-motion compensated) dense and PNN score identically to every printed digit — 92.92% acc, 0.587 IoU — at both 4.93× (width 1024) and 6.67× fewer parameters (width 4096). The dense model's extra 43 M parameters buy nothing.

Layer width	Connectivity density	Params saved	Accuracy gap
512	22.9%	4.05×	+0.9 pp
1024	19.3%	4.93×	−0.1 pp
2048	16.6%	5.85×	−0.12 pp
4096	14.7%	6.67×	−0.36 pp

Waymo-class BEV occupancy (perception-shaped), 3-seed mean, iso-architecture. Accuracy gap = dense − PNN. At width 4096 the PNN is smaller, slightly more accurate, and faster per epoch than the dense baseline.

Parameters

Fewer parameters per model

Parameter-reduction factor: dense ÷ PNN active parameters, iso-architecture (higher = better). Waymo LiDAR = real Open Dataset v2, width 4096 (50.5 M dense vs 7.6 M PNN).

Accuracy

Near-parity, shown truthfully

Dense baseline PNN

Waymo LiDAR (real data, occupancy forecast +0.5 s): dense and PNN are identical to every printed digit. Test accuracy %, dense vs PNN side by side, y-axis from 0 so the small gap is shown honestly. Mean of 3–5 seeds.

Memory

Model footprint — dense FP32 vs PNN INT8

Dense baseline PNN

Megabytes, log scale. Index-free connectivity means columns are computed, not stored.

Speed & energy

Where the speed actually shows up

~12 nJ

PNN chip (est)

~1.3 µs / inference, ~10–25× better energy than a matched NVIDIA part on a deployed fixed model.

Scaling — language

GPT-3 on one die

GPT-3's widest layers are PNN's best regime: connectivity density falls to 10–12%, cutting 175 B parameters to 19.3 B active (9.08×). That is exactly the factor that makes the impossible chip possible — all of GPT-3, weights resident on a single 2 nm reticle-limit die, no HBM. The dense model needs ~7 reticles at the same precision.

9.08×

175 B → 19.3 B active parameters

1 die

826 mm² @ 2 nm — dense needs ~7 reticles

~6,500

tokens/s single-stream at ~15 W (est)

2.4 mJ

per token — ~600× less than GPU serving (est)

Analytical estimate from a reproducible calculator (INT4 weights in mask ROM, KV on-die, ctx 512, 16.7 TOPS/W sustained); silicon figures are engineering estimates pending back-annotated synthesis. Accuracy at GPT-3 width is the evaluation-phase measurement — parity is demonstrated to width 4096.

Honest regime map

Where PNN wins — and where it doesn't

Wins

• Memory: always — 4.5× fewer params (up to 6.7× on wide perception), 18× smaller INT8.
• Accuracy: near-parity on vision, graceful on tabular (≤1.1 pp).
• Op-bound speed: 1.6× on scalar / no-BLAS edge.
• Custom silicon: ~1.3 µs, ~12 nJ; ~10–25× better energy than NVIDIA (est).

Limitations

• No CPU/GPU speed win vs tuned dense GEMM — the prime gather is SIMD-hostile.
• INT8 not universally free — lossless on vision, harmful if input layers are over-sparsified.
• NLP gain modest (1.7×, FFN-only); attention/embeddings stay dense.
• Hardware numbers are engineering estimates; measured CPU reality is parity.

Memory & energy efficiency at near-parity accuracy