How To Undervolt GPU With Afterburner

Undervolting a GPU means reducing the voltage supplied at a given clock speed, which can lower temperatures and fan noise while keeping performance close to stock, since GPUs are typically shipped with more voltage headroom than most individual chips actually need for stability. MSI Afterburner is the standard free tool for this process. It requires patient, incremental testing rather than applying one universal setting, since the ideal values vary by individual GPU silicon and the specific card’s cooler.

Expect the full process, done properly with adequate testing at each step, to take a few separate sessions spread across a few days rather than a single sitting — rushing to find your final setting in one evening tends to either leave meaningful headroom on the table from being overly conservative, or land on a value that hasn’t actually been tested thoroughly enough to trust for daily use.

Quick answer

Open Afterburner’s voltage/frequency curve editor, lower the voltage at your GPU’s typical boost clock while keeping that clock speed’s frequency point, then test stability under sustained load and adjust incrementally. Expect this process to take multiple test sessions rather than a single sitting, since finding a stable, worthwhile undervolt is inherently iterative.

Why undervolting works

GPU manufacturers set default voltage curves conservatively enough to guarantee stability across the full range of individual chips produced, from the best-performing silicon to the weakest that still passes quality control. This means many individual GPUs have more voltage than they actually need at a given clock speed to run stably.

Reducing voltage while maintaining the same clock speed lowers power draw and heat output, since power consumption scales with voltage more steeply than with clock speed alone. Lower heat output means the GPU can run cooler at the same performance, or the same temperature with quieter fan behavior.

Because this variance in “how much voltage a specific chip actually needs” differs between individual units of the same GPU model, there’s no single undervolt setting guaranteed to work for every card — this is why testing your own specific GPU is necessary rather than copying a number from someone else’s card.

This chip-to-chip variance is sometimes called the “silicon lottery,” and it applies across the entire product stack, not just at the very top or bottom — even two cards bought from the same retailer on the same day can land at meaningfully different stable voltage points. Treat any specific number you read in a forum thread or comment section as a plausible starting point to try, not a value you should expect to reproduce exactly.

On the bench here, testing multiple samples of the same GPU model has shown noticeably different stable undervolt points between otherwise identical cards, which is a useful reminder that community-reported numbers are a starting reference point, not a guarantee for your specific unit.

The underlying relationship worth understanding in slightly more depth: power draw in a GPU scales roughly with the square of voltage and linearly with clock speed, which is why a relatively modest voltage reduction — often in the range of a few tenths of a volt off a typical boost voltage — can produce a proportionally larger reduction in power draw and heat than the voltage percentage alone would suggest. This is also why undervolting tends to be a more efficient lever for reducing heat than simply lowering clock speed by the same percentage would be.

Setting a baseline before changing anything

Before adjusting any settings, run a benchmark or a demanding game session at stock settings and record your GPU’s typical boost clock, temperature under sustained load, and a performance score or frame rate for later comparison. Our guide to GPU benchmark software covers tools useful for this baseline.

Note your GPU’s power draw and temperature specifically, since these are the two metrics an undervolt is meant to improve — without a documented baseline, it’s hard to confirm whether a given undervolt setting actually accomplished anything.

A screenshot of HWiNFO64’s summary panel at the end of your baseline run, saved somewhere you’ll actually find it again, is a more reliable record than jotting numbers by hand, since it captures every relevant sensor reading at once rather than the handful you remembered to write down — useful not just for this specific undervolting project but as a general reference point any time you want to check whether the system’s behavior has drifted from its original baseline months later.

This baseline session also serves as a sanity check that your system is stable before you start making changes, since starting from an already-unstable system makes it much harder to correctly attribute any crash you see later to the undervolt itself versus a pre-existing issue.

If your baseline session itself shows any instability before you’ve touched a single voltage setting, stop and resolve that first — chasing an undervolt on top of an already unstable system only compounds the confusion when you eventually try to figure out which of the two problems is causing a given crash.

Write down the specific numbers rather than relying on memory — GPU boost clock, average temperature over a 10-15 minute sustained load, peak power draw, and either a benchmark score or an average FPS figure from a repeatable in-game test scene. Having these four numbers on hand before you touch the curve editor turns “did this help” from a vague impression into a direct, checkable comparison once you’ve settled on a candidate undervolt setting.

Adjusting the voltage/frequency curve

Afterburner’s curve editor displays a graph of clock speed against voltage, showing the GPU’s stock behavior as a series of points. Undervolting typically means selecting a point near your GPU’s common boost clock and dragging it down in voltage while keeping the clock speed at or near its original value.

After dragging a point down, the curve to its right typically flattens, meaning the GPU will run at that lower voltage across a range of clock speeds rather than just the single point you adjusted — this is expected curve editor behavior, not an error.

Apply the change and test rather than adjusting several points across the curve simultaneously on a first attempt, since testing one adjusted region at a time makes it far easier to identify exactly which change caused an instability if one shows up.

Small voltage reductions, tested incrementally, are safer than attempting a large reduction immediately — work in small steps and confirm stability at each step before pushing further.

A reasonable starting increment for a first attempt: reducing voltage at your target clock point by a small amount (commonly a few tens of millivolts per step in community-documented approaches) and testing before taking the next step down is a more methodical path to your GPU’s actual stability limit than guessing a large reduction upfront and working backward from a crash, which tells you less about exactly where the safe boundary actually sits.

Testing for stability

Run a sustained, demanding workload — a stress test tool or an extended benchmark loop — after each change, since instability from an overly aggressive undervolt often appears only after several minutes of sustained load rather than immediately.

Watch for driver crashes, screen artifacts (unusual visual glitches or corruption), or a complete system crash during testing, all of which indicate the current voltage is too low for that clock speed and needs to be raised slightly.

Test across more than one type of workload if possible — a synthetic benchmark and an actual demanding game — since different workloads can stress the GPU differently, and a setting stable in one may not be stable in another.

It’s also worth including a workload that exercises the GPU’s video decode and encode engines specifically if you stream, record, or watch a lot of hardware-accelerated video, since some undervolt settings that are perfectly stable for pure 3D rendering can still be marginal for the separate media engine on the same die — a brief test playing back or recording a demanding video source alongside your usual gaming test catches this less commonly checked failure mode.

If a crash occurs, raise the voltage slightly at the point that caused the problem rather than abandoning the undervolt entirely — often a small adjustment back up resolves the instability while still retaining most of the temperature benefit.

A useful signal to watch for beyond outright crashes: some unstable undervolts show up first as a subtle drop in benchmark score or a slightly lower sustained boost clock rather than a dramatic crash, since the GPU’s own protection mechanisms can quietly correct for marginal instability by backing off clock speed rather than failing outright. Comparing your post-change benchmark score against your documented baseline, not just watching for crashes, catches this quieter failure mode that a crash-only testing approach would miss.

Confirming the actual benefit

After settling on a stable curve, run the same benchmark or game session used for your baseline and compare temperature, power draw, and performance score directly against the original numbers, since this is the only way to confirm the undervolt actually achieved something worthwhile rather than just assuming it did.

A successful undervolt typically shows meaningfully lower temperature and power draw with performance within a small margin of the original baseline — if performance dropped noticeably more than expected, the voltage reduction may have inadvertently reduced the GPU’s ability to sustain its boost clock, which is worth investigating further.

Document the final settings that worked for your specific card, since driver updates occasionally reset custom curves, and having a record saves you from re-testing the entire process from scratch.

Keep expectations calibrated to what undervolting can realistically deliver: it’s a refinement that improves an already-functioning system’s thermal and noise characteristics, not a substitute for adequate case airflow or a cooler that’s genuinely undersized for the card. If your baseline temperatures were already uncomfortably high before undervolting, addressing airflow first, covered in our guide to fixing PC overheating, generally yields a larger improvement than voltage tuning alone can achieve.

Saving and applying your settings automatically

Once you have a stable, verified undervolt, save it as a profile within Afterburner and enable the option to apply it automatically at Windows startup, so you don’t need to manually reload the curve every session.

Confirm the profile actually loads correctly after a fresh restart by checking the curve editor’s current state, since a startup application setting that isn’t configured correctly will silently leave the GPU running at stock voltage until manually reapplied.

If you update your GPU driver after establishing a stable undervolt, retest stability afterward — driver updates occasionally change power management behavior in ways that can affect a previously stable curve.

When undervolting isn’t worth pursuing

If your GPU already runs cool and quiet under your current cooling setup, the practical benefit of undervolting is smaller, and the time investment in testing may not be worthwhile unless you’re specifically chasing lower noise for a quiet-focused build.

If you’re not comfortable with the iterative testing process or don’t want to risk occasional crashes during the testing phase itself, this isn’t a required step for good GPU performance — a stock-configured GPU with proper cooling, covered in our guide to checking GPU temperature safely, performs perfectly well without any voltage adjustment.

If your GPU is still under warranty and you’re specifically worried about a manufacturer voiding coverage over a software-level voltage adjustment, it’s worth noting that undervolting through vendor-sanctioned tools like Afterburner generally operates within the GPU’s normal supported adjustment range rather than bypassing a hardware protection, unlike physically modifying voltage regulation circuitry — but checking your specific card manufacturer’s warranty terms directly removes any doubt rather than relying on a general assumption either way.

Troubleshooting undervolt attempts

Driver crashes shortly after applying a new curve: raise the voltage at the clock point you most recently adjusted, since this is the most common sign that specific point is unstable at the current setting.

System is stable in benchmarks but crashes in specific games: test more workloads before considering a setting confirmed stable, since some games stress the GPU differently than synthetic benchmarks and can reveal instability a benchmark alone missed.

Performance dropped more than expected after undervolting: the GPU may not be sustaining its target boost clock at the reduced voltage — try a smaller voltage reduction at that clock point rather than the more aggressive value you tested.

Settings reset after a driver update: this is a known occurrence with some driver updates — reapply your saved profile from Afterburner and retest stability, since the update may have also changed underlying power behavior.

Temperatures didn’t improve despite a confirmed stable undervolt: check that the profile is actually being applied (confirm in the curve editor during a session) rather than assuming it’s active, since an unapplied profile leaves the GPU at stock voltage with no visible benefit.

Curve editor shows the change applied but clock speed doesn’t match what you’d expect: some GPUs adjust their actual sustained boost behavior based on both the curve and the current power limit setting together, so a curve change alone without also reviewing the power limit slider can produce a smaller clock speed shift than the curve graph alone would suggest — check both settings together rather than the curve in isolation.

Frequently asked questions

Does undervolting reduce GPU performance?

A well-tuned undervolt typically maintains performance close to or matching stock, since the goal is finding the lowest stable voltage at each clock speed rather than reducing clock speeds. Poorly tuned or overly aggressive undervolts can cause instability rather than a clean performance trade-off.

How much temperature reduction can I expect from undervolting?

This varies significantly by GPU model, silicon quality, and cooling setup, but a reduction in the range of a few degrees to around ten degrees Celsius under load is commonly reported, alongside potentially lower fan noise from reduced cooling demand.

Is undervolting safer than overclocking?

Generally yes, since you’re reducing voltage rather than pushing the GPU beyond its stock behavior, which carries less risk of instability-related crashes during everyday use. Improperly tested settings on either can still cause crashes, though neither typically causes permanent hardware damage within normal adjustment ranges.

Do I need to test stability after every small change, or can I make several changes at once?

Testing after each small, incremental change is strongly recommended, since it’s the only way to know which specific adjustment caused an instability if one appears. Changing multiple values simultaneously makes troubleshooting a crash much harder.

Will my undervolt settings need to be reapplied after every restart?

Yes, unless you configure Afterburner to launch at startup and automatically apply your saved profile, in which case the settings load automatically each time Windows starts.