> ## Documentation Index
> Fetch the complete documentation index at: https://docs.inkwell.finance/llms.txt
> Use this file to discover all available pages before exploring further.

# Performance

> Measured match-compute wall-clock across CPU, consumer GPU, and H100 — with correctness verified on every tagged number. Headline: 7.77 s for an N=32 match cycle on H100, every pair decrypted and asserted.

> **CONFIDENTIAL & PROPRIETARY © 2026 Inkwell Finance, Inc. All Rights Reserved.**
>
> This document is for informational purposes only and does not constitute legal, tax, or investment advice, nor an offer to sell or a solicitation to buy any security or other financial instrument. Any examples, structures, or flows described here are design intent only and may change.

Every number in this section is status-tagged — `measured`,
`measured-with-caveat`, `projected`, or `extrapolated` — and backed
by a committed run log.

<CardGroup cols={2}>
  <Card title="N=32 match cycle · H100" icon="stopwatch">
    **7.77 s** on an NVIDIA H100 SXM, every output decrypted and
    asserted against the plaintext expected value. The
    correctness-preserving floor on the current FHE backend.
  </Card>

  <Card title="Atomic min · H100" icon="bolt">
    **\~15 ms** mean over 20 iterations, **20/20** outputs decrypted
    and asserted. The one-comparison floor — every higher-level op
    composes from it.
  </Card>

  <Card title="Verification discipline" icon="list-check">
    Four status tags (`measured` / `measured-with-caveat` /
    `projected` / `extrapolated`) govern every number on these
    pages. See [Methodology](/dagon/performance/methodology) for the
    citation rules.
  </Card>
</CardGroup>

The full picture: single-batch compute is measured on hardware from
consumer CPU through H100; correctness is verified on every cited
measurement; and the cells that can't yet clear correctness (CKKS
past the bootstrap noise floor) are tagged so no one mis-cites them.

## What's in this section

<CardGroup cols={2}>
  <Card title="Benchmarks" icon="gauge" href="/dagon/performance/benchmarks">
    The headline scheme × hardware figures with the verification
    discipline that backs every number.
  </Card>

  <Card title="Methodology" icon="list-check" href="/dagon/performance/methodology">
    The four status tags and the rules for citing each. Why the
    discipline matters for an FHE product.
  </Card>

  <Card title="Why CKKS fails" icon="triangle-exclamation" href="/dagon/performance/why-ckks-fails">
    Why CKKS at production matching depth does not clear
    correctness, and why programmable-bootstrap is the
    correctness-preserving path.
  </Card>

  <Card title="SIMD amortization" icon="layer-group" href="/dagon/performance/simd-amortization">
    How packing delivers near-free per-batch amortization on a
    saturated GPU.
  </Card>

  <Card title="Scaling dashboard ↗" icon="chart-line" href="https://dagon.so/scaling">
    Interactive model anchored to the measured headline numbers.
  </Card>
</CardGroup>

## The short version

* **Correctness-preserving path is programmable-bootstrap FHE.**
  Each homomorphic primitive folds one bootstrap, so there is no
  chain-depth limit on the bisection.
* **CKKS at production bootstrap chain depth does not clear
  correctness** on Dagon's matching circuit. We measure it for
  throughput parity on the analysis pages, but we do not ship it.
* **H100 is the production target.** Consumer GPU (4070 Ti) stays
  available for local development; measurements show a 5–6× consumer
  → datacenter speedup on the same workload.
* **The next-generation backend is a target, not a measurement.**
  Every projected figure for it is derived from published primitive
  cost and tagged accordingly.