Menu
Hardware
Homelab & NAS
Tools & Builders
Prices & Data
Troubleshoot

Attributed GPU Tuning Recipe Cards · 5-Rung Evidence Ladder · CC BY 4.0 Open Dataset

The Undervolt Cookbook: Attributed GPU Tuning Recipe Cards

Browse and print attributed GPU undervolt and power-limit recipe cards for RTX 5080, RTX 5070 Ti, RX 9070 XT, RX 9070, and RTX 4090 with stability caveats.

Silicon lottery & provenance rule: Every card below is a record of what one named tester measured on one specific sample—never a guarantee your card will hold the same voltage curve. TechFuelHQ's own bench row (ASUS ROG Astral RTX 5080 OC stock baseline) is fenced separately from reviewer tests and owner reports, and group medians ("Chef's Notes") stay suppressed until n ≥ 5 protocol-matched runs exist for the same GPU.

Loading attributed recipes from /data/undervolt_results.json…

Paired Stock-vs-Tuned Test Kitchen Checklist (Saved Locally)

Use this six-step protocol when testing your own card so stock and tuned numbers come from the same thermal state, and report failed bakes or 30-day held follow-ups openly.

    Evidence cards, not blind copy-paste offsets

    Every card in The Undervolt Cookbook renders directly from our open CC BY 4.0 GPU Undervolt Settings Database (/data/undervolt_results.json).

    How to read the 5-rung Evidence Ladder

    1. 1·Cited Source: Verbatim passage, author, publication/forum, and capture date are on file.
    2. 2·Matched Pair: Stock and tuned figures were recorded in the same named workload on the same card.
    3. 3·Stability Hours: The tester stated how many hours the profile held without crashing or resetting (unlit when omitted).
    4. 4·2nd Source: A second independent publication or report documented the same test or setting.
    5. 5·30-Day Held: A follow-up report confirmed the curve remained crash-free across 30 days of mixed games and idle transitions.

    Frequently asked questions

    Why does every recipe card put the 5-rung Evidence Ladder above the voltage numbers?
    An aggressive curve offset that passed one 60-second benchmark run often crashes two weeks later in a light desktop transition or menu screen. Lighting only the completed rungs—cited source, matched stock-vs-tuned pair, stated stability duration, corroborating source, and 30-day held follow-up—makes missing durability evidence visible before you copy an offset.
    Why do some RX 9070 XT and RX 9070 undervolt cards draw more watts than stock?
    On modern AMD Adrenalin builds, pairing a negative voltage offset (such as -125 mV or -170 mV) with a +10% power target raises boost clocks rather than lowering power draw. The cookbook tags those explicitly as Performance Tunes (+Power / +Clock) so they are never confused with watt-saving recipes.
    Why are Chef’s Note medians withheld on current cards?
    Computing a median across unlike games, unlike board power limits, or sample sizes of one or two misleads readers. Each card prints the exact row count (n) for that GPU and withholds group medians until at least five protocol-matched paired runs exist.

    Updates and corrections

    Last updated
    Methodology
    See our methodology for research and review standards.
    Update log
    • 2026-11-21 — Initial public release: printable attributed GPU tuning recipe cards over data/undervolt_results.json, headed by a 5-rung evidence ladder, explicit stability-caveat tags, and n >= 5 group median suppression.
    Corrections
    Spotted an error? Email a correction with the affected line and what you found. The draft includes this page’s address. Verified changes are recorded here. Contact details.