X-40

X-40 ResearchCore — Accelerate Discovery with Quantum Attention

Run physics-grade Φ-Anchors and entropy analytics on your own datasets — in milliseconds and with NVML-audited energy use near zero. Powered by QEIv15, the quantum-entropy core of X-40.

Testbed & API Version

  • GPU: NVIDIA H100 80GB HBM3 (Runpod)
  • CUDA: 12.1 · Torch: bf16 where applicable
  • API: X-40 ResearchCore v0.6 (Anchors, Audit, ZIP Batch)
  • Audit: NVML-based (avg_W, energy_Wh) measured per batch

How it works

You send one or more timeseries (e.g., an EEG channel, a climate station, a returns series). The API returns:

  • indices — anchor points (events / outliers / turns)
  • values — values at those indices
  • phi — compact entropy / structure score for the series
  • latency_ms — per-series processing time
  • _audit{ elapsed_ms, avg_W, energy_Wh } (always on here)

JSON Batch (audit=1)

Paste your items array and run. The response includes anchors, Φ, per-series latency, batch elapsed, and NVML-audited energy.

ZIP Batch Upload (CSV/JSON, audit=1)

Put many files into a .zip. CSV should have a single numeric column named value. JSON can be an array or { series: [...] }.

Case studies

Real runs on NVIDIA H100 80GB (Runpod), CUDA 12.1, Torch bf16. Each card shows ms (top), then gains and energy.

~6.08 ms
Finance · EOD (SPY/QQQ/DIA) — Batch (4 series)
427.6× faster vs CPU13.2× faster vs GPU98.8% less energy vs CPU88.4% less energy vs GPU
Energy ≈ 1.16×10⁻⁴ Wh
  • Method: mad(k_sigma=3.5, top_k=128)
  • Data: daily closes → log returns
  • Audit: avg_W≈69 W via NVML
~96 ms
Neuroscience · UCI EEG Eye State — Single series (demo)
6.8× faster vs CPU0.2× faster vs GPU81.7% less energy vs CPU less energy vs GPU
Energy ≈ 1.83×10⁻³ Wh
  • Method: zscore(window=256, z=3.0)
  • Data: ARFF → first channel (or median of few)
  • Audit: avg_W≈69 W via NVML
~13 ms
Climate · NOAA Mauna Loa CO₂ (weekly) — Single series
50.0× faster vs CPU1.5× faster vs GPU97.5% less energy vs CPU75.2% less energy vs GPU
Energy ≈ 2.48×10⁻⁴ Wh
  • Method: peaks(min_prom=0.6, neighborhood=7)
  • Data: NOAA GML weekly trend
  • Audit: avg_W≈69 W via NVML