Quick answer

This open dataset collects real GPU undervolt settings and their measured results — voltage or offset, power saved, temperature drop, and FPS change — for cards from the RTX 4090 to the RX 9070 XT. Most modern GPUs undervolt to save 50–100W and run cooler while holding performance; power-limited cards can gain FPS.

By LK Wood IV · Published 2026-07-24 · ~6 min read · St. Louis County, MO

Undervolting is one of the highest-value ten-minute tweaks in a PC — but the single question everyone asks is “what settings should I actually use for my card?” Published guides give a method; this page gives data: an open, crowdsourced table of real undervolt settings and the results people measured, one row per tested configuration.

It is the data companion to our complete guide to undervolting your GPU, which walks through the method itself on both NVIDIA and AMD. Use that to learn how; use the table below to see where other people’s cards landed.

The results

Each row is a real, sourced result. NVIDIA rows list the MSI Afterburner voltage-frequency point (voltage @ clock); AMD rows list the Adrenalin offset (or, on RDNA 3, the absolute max voltage) plus any power-limit change. Where a number was not measured, the cell is blank rather than guessed.

8 verified results CC BY 4.0 · updated 2026-07-24 · open dataset
Community GPU undervolt results — measured settings and outcomes, one row per tested configuration. Sourced; contribute your card by pull request.
GPUMethod & settingPower savedTemp dropFPS changeStability testSource
RTX 5080ASUS ROG Astral OC950 mV @ 2750 MHz, power limit 90%MSI Afterburner V/F curve~50 W—±0% (parity)Settings follow the worked example in our RTX 5080 undervolt guide; the ~50W saving sits inside the 30-70W range that guide cites from community testing at hardware review sites. Not a logged before/after measurement on our own card.TechFuel HQ
RX 9070 XTSapphire NITRO+-100 mV offset (power limit left at stock)AMD Adrenalin voltage offset——+10%Stress-tested in a demanding game after stepping the offset down from -40 mV; validated in real play, not just a short benchmarkwww.xda-developers.com · 2
RTX 4090940 mV @ 2700 MHz (+1200 memory)MSI Afterburner V/F curve~100 W—±0% (parity)Loop a demanding title / stress test (instability shows as crashes or black screens). FE silicon often stable ~925 mV; AIB cards ~950-975 mV.forums.overclockers.co.uk · 2
RTX 4080 Super950 mV @ 2760 MHzMSI Afterburner V/F curve~50 W~8°C±0% (parity)Extended in-game / ray-tracing stress test (not just synthetics); some cards need +12-25 mV in heavy RT titles.evezone.evetech.co.za · 2
RTX 5090900 mV @ 2800 MHz (flatten curve above the point)MSI Afterburner V/F curve~100 W~10°C-2%Several hours in a memory-heavy title (Cyberpunk 2077 / Monster Hunter). Verify your Afterburner build supports 50-series curve editing.www.overclock.net · 2
RTX 5070 Ti935 mV @ 2955 MHz (via +450 MHz core offset), +1500 memoryMSI Afterburner V/F curve~50 W—+4%Flatten the V/F curve at the target point; achievable clock at 935 mV varies by card (~2932-2945 MHz measured on the source unit).www.esportstales.com · 2
RX 7900 XTX1100 mV max voltage (~-50 mV vs ~1150 mV stock), power limit +15%AMD Adrenalin max voltage (RDNA3 absolute)——±0% (parity)Lower max voltage in 10-15 mV steps; if it crashes/black-screens, raise 10-15 mV and retest (an invalid UV just resets on reboot). Validate with 15-20 min stress + 20-30 min gameplay.tier1settings.com · 2
RX 9070-100 mV offset, -30% power limit (efficiency-biased)AMD Adrenalin voltage offset——-7%-100 mV reported daily-stable; community reports crashes past ~-170 mV. This is an EFFICIENCY tune: the -30% power limit trades ~7% FPS for large power/thermal savings. A milder -80 to -100 mV at 0% power limit is the performance-neutral option.www.notebookcheck.net · 2

Have results for your card? This is an open dataset — add a row by pull request against data/undervolt_results.json. Submit a pull request against https://github.com/iBlessi/gpu-undervolt-db adding one object to the "results" array in data/undervolt_results.json. The repository's .github/PULL_REQUEST_TEMPLATE.md fills in automatically with the required fields and methodology, and an automated check validates schema, plausibility, and sourcing on every PR. Every numeric value must cite a resolvable source.

How to read it

  • Method & setting — the exact tune. NVIDIA cards undervolt through the MSI Afterburner voltage-frequency curve (pin a low-voltage point to your stock clock); AMD cards use a negative voltage offset in Adrenalin.
  • Power saved / temp drop — the efficiency win at the same performance. This is where the value is on most cards.
  • FPS change — usually ±0% (same frames, cooler and quieter). A positive number means the card was power-limited, so freeing voltage let it boost higher — the reason the RX 9070 XT can gain about 10% FPS.
  • Silicon lottery — treat every row as a starting point, not a guarantee. Two identical cards can need 25–50mV apart. Start conservative, test, and step down.

None of the seed rows is a logged first-party before/after run yet — they are community-consensus settings, community-verified results, and review-measured values, each with its source linked in the table. Our own RTX 5080 and RX 9070 XT guides are where those settings are worked through. First-party measured rows are exactly what this database is built to accumulate, starting with submissions.

Contribute your card

This dataset grows by submission, and the submissions repository is now open: github.com/iBlessi/gpu-undervolt-db. There is no login to create, no account, and no backend — git is the moderation queue. To add your result:

  1. Undervolt your card (our guide covers the method) and record what you changed and what you measured.
  2. Fork the repo and add one object to the results array in data/undervolt_results.json, then open a pull request. The submission template fills in automatically with the field guide and checklist — the card, the offset or voltage curve you set, and the numbers you measured.
  3. An automated check runs on your PR. It blocks on hard problems — malformed schema, impossible values, a missing source — and warns on statistical outliers, which can still merge but are held out of the table above until a maintainer confirms them. You can run the same check locally with python scripts/undervolt_outlier_flag.py --check.

Prefer not to use GitHub? Email hello@techfuelhq.com with the same information and we will open the pull request for you.

The only hard rule is honesty: every number should trace to something real — a review, a forum thread, or your own screenshot. Unmeasured fields stay null. Implausible submissions are flagged and held out of the table until reviewed.

What the data shows so far

  • Undervolting is almost free performance. Across every card here, the tuned result is the same frames (or a few more) at meaningfully lower power and temperature.
  • The bigger the card, the bigger the win. A 575W RTX 5090 or 450W RTX 4090 reclaims ~100W; a mid-range card saves ~50W. The power & cost calculator turns those watts into dollars per year.
  • Power-limited cards gain FPS. The RX 9070 XT is the clearest example — undervolting lets it hold higher clocks within its power budget.
  • The consensus voltages are tight. Modern NVIDIA cards cluster at 900–950mV; RDNA 4 AMD cards at -80 to -100mV. That convergence is exactly what a crowdsourced dataset is good at confirming.

Sources and methodology

Every row cites its source in the table. Community rows link to the review or forum thread the numbers came from; rows drawn from our own guides link to the guide that works the settings through. The dataset is published under CC BY 4.0, and the field definitions and contribution rules travel with it in the file’s own meta block. If you find an error, open a pull request against the repository or email hello@techfuelhq.com with the card and the correction — corrections are treated the same as new submissions and are checked against a source before the table changes.

Frequently asked questions

What voltage should I undervolt my GPU to?
It depends on the card and its silicon, but the community-consensus starting points cluster tightly. On modern NVIDIA cards, a voltage-frequency point around 900–950mV at your stock sustained clock is the usual daily-stable target (a 5090 leans nearer 900mV, a 4080 Super nearer 950mV). On AMD, you apply a negative offset — around -80 to -100mV on RDNA 4 (RX 9070 series). Treat any published number as a starting point to test on your own card, since silicon quality varies from chip to chip.
How much power does undervolting a GPU save?
In this dataset the measured savings run from about 50W on a mid-range card to 100W on a high-end RTX 4090 or 5090, at the same performance. The bigger and hotter the card, the more watts an undervolt reclaims, because you are cutting voltage on a chip that was drawing 300–575W at stock. The power & cost calculator turns the watts you save into an annual dollar figure.
Does undervolting reduce FPS?
Usually not. A tuned undervolt holds the same clocks at lower voltage, so performance stays flat (0% change) while temperature and power drop. On a power-limited card — like the RX 9070 XT — lowering voltage frees power budget for higher sustained clocks, and FPS can actually rise a few percent. You only lose performance if you pair a deep voltage cut with an aggressive power-limit reduction, which trades frames for maximum efficiency.
What are the best undervolt settings for an RTX 4090?
The most-repeated daily-stable point across community threads is roughly 940mV at 2700MHz in the MSI Afterburner voltage-frequency curve, often paired with a small memory overclock. Founders Edition cards frequently hold a notch lower (~925mV); AIB models tend to want 950–975mV. The row here records about 100W saved — roughly 450W dropping toward 350W at essentially stock performance (some owners push undervolts further, toward 300–320W). Validate any point with a long loop of a demanding game, since instability shows up as a crash or black screen.
How do I add my own undervolt results?
Open a pull request against github.com/iBlessi/gpu-undervolt-db, adding one object to the results array in data/undervolt_results.json. The submission template fills in automatically, and an automated check validates schema, plausibility, and sourcing on every PR — git is the moderation queue, so there is no login or account to create. If you would rather not use GitHub, email hello@techfuelhq.com and we will open the PR for you. Every numeric value should trace to a source or your own screenshot.
Is undervolting a GPU safe?
Yes. Undervolting lowers the voltage the card runs at, which reduces electrical and thermal stress rather than adding it — the opposite of overvolting. The only realistic failure is instability: if you go too low, a game or benchmark crashes, which you fix by raising the voltage slightly. Every setting lives in software and resets on reboot or driver reinstall, so the card returns fully to stock at any time.

Evidence ledger

Last updated
Methodology
This guide was written and edited by Lowell K. Wood IV in St. Louis County, MO. Specs and prices verified against vendor and project documentation current on the date above. Full editorial standard: methodology.
Update log
  • 2026-07-25 — Last reviewed and updated.
Corrections
Spotted an error or a stale number? Email hello@techfuelhq.com. Confirmed corrections are added to the update log above.

About the author

Written by Lowell K. Wood IV, who builds and runs TechFuelHQ from St. Louis, Missouri.