# Measure the whole system before claiming an efficiency breakthrough

Proposed benchmark / no claimed speedup. XDPU’s displayed 13.37× performance and 37% savings belong to its fictional demonstration, not the Necromancer budget. No speed, energy, cost, or return figure transfers from that interface into an engineering claim. A biological processor must be compared with a stated alternative on a stated task at a matched level of output quality.

Canonical page: https://givegigs.com/wiki/necro-chapter-xdpu-energy-accounting
Last updated: 2026-11-09T18:58:12.584Z
Parent topic: [necromancer-neuroscience](https://givegigs.com/wiki/necromancer-neuroscience)

## Contents
- [Overview](#wiki-section-overview-1)
- [Chapter notes](#wiki-section-chapter-notes-2)
- [Dossier section](#wiki-section-dossier-link-3)

<a id="wiki-section-overview-1"></a>
## Overview

Proposed benchmark / no claimed speedup. This page expands the “Measure the whole system before claiming an efficiency breakthrough” chapter in the neuroscience collection.

<a id="wiki-section-chapter-notes-2"></a>
## Chapter notes

XDPU’s displayed 13.37× performance and 37% savings belong to its fictional demonstration, not the Necromancer budget. No speed, energy, cost, or return figure transfers from that interface into an engineering claim. A biological processor must be compared with a stated alternative on a stated task at a matched level of output quality.

Proposed accounting defines E_task = ∫ P_system(t) dt / N_accepted, with joules per accepted task and an explicit measurement boundary. Include allocated maintenance, incubators, pumps, acquisition, host compute, cooling, preparation overhead, idle periods, and failed runs. Report both marginal energy and allocated total energy, the allocation rule, and throughput; counting only neuronal metabolism can conceal most operational demand.

Proposed gate: repeatable workload-specific benefit after uncertainty and life-support overhead are included. Financial modeling separately includes consumables, staff, device replacement, quality control, and downtime. Model assumptions must remain editable and visible. Even if no efficiency advantage emerges, a superior biological assay may have research value; that is a distinct customer proposition requiring validation.

<a id="wiki-section-dossier-link-3"></a>
## Dossier section

[Open this exact section in the Necromancer dossier](/necromancer#xdpu-energy-accounting). The original chapter status is “Proposed benchmark / no claimed speedup.”

## Sources

- [XDPU Proving Ground](https://mattyjacks.com/xdpu)

---
This article is collaboratively maintained. Check the canonical page and cited sources for the latest context.