How to Undervolt a GPU Safely (and Why It Helps Noise and Temps)

GPU undervolting can cut heat, fan noise, and power use—often with little to no performance loss. This guide shows a safe, repeatable workflow for NVIDIA and AMD GPUs, including what to change, how to stress-test, and an

Undervolting the GPU means reducing the voltage so it needs less power to run at its stock clock speed. Doing so usually means less heat (especially the fan cooling them), a lower fan speed, and therefore, less noise, with the same “real world” FPS.

The safest way to start with GPU tuning is make basic changes in increments: set a baseline, and then change one thing at a time. Adjust voltages, then clock speed, then apply the change and finally, test it by running the same benchmark or game. If you want to go lower after that, go lower.

And if you do go too far dropping the power consumed and the local CPU crashes (the whole thing hangs), the drivers re-set themselves, or you get weird artifacts, nothing “breaks”. You can just go back to your last known stable setting, or hit Reset.

AMD Radeon folks can start with the new “Undervolt GPU” automatic tuning option in AMD Software: Adrenalin Edition. (amd.com) If you have an NVIDIA (and for a lot of cards in general), the most common manual way to do it is to edit a voltage/frequency curve in MSI Afterburner and saving down profiles. (us.msi.com)

What ‘undervolting’ actually means to people (in plain English)

You tune your GPU to run at that stock clock speed at a lower voltage. The intent is to cut back on power draw for less heat and maintain its overall performance up to certain stock limits of the card. So it’s different from ‘underclock’ per se, but also, not just cutting back on the power limit (reducing the allowable power drawn of the GPU, and sometimes lowering clocks). Generally, in all cases, the smaller the performance hit, the better. Most people combine a light undervolt with a lowering in the power limit to make it quietest.

Why undervolting can reduce temps and fan noise

GPU power hits go up quickly as voltage increases (a crude rule of thumb is that power scales roughly with voltage squared). Even a minor voltage drop can trim watts noticeably. Fewer watts means less heat for your cooler to dissipate, which usually equates to lower fan RPM—and that’s where the noise drop comes from.

You can see the same cause-and-effect by lowering a GPU’s power limit: lower power usually lowers temp and fan speed at load (often with a modest performance hit). (pcworld.com)

Before you begin: a safety-first checklist

• Update your GPU driver first (this gets rid of a big variable when you’re hunting for instability).
• Make sure your case airflow is at least reasonable (a great undervolt can’t fully fix a “sealed” hotbox case).
• Shut background overclock/tuning apps so they’re not fighting each other (pick one tool per GPU).
• Decide what you’re optimizing for: (1) quieter fans, (2) lower hotspot temps, (3) lower power draw, or (4) best performance-per-watt.
• Pick 1-2 repeatable tests (one benchmark + one real game) and use the same scene/area every time.
• Plan to change ONE thing at a time: if you change voltage + clocks + fan curve at once, you don’t know what crashed you.

Tools you’ll need (and what each is for)

Tip: Your monitoring overlay should show at least GPU temperature, hotspot/junction (if available), GPU power (watts), fan speed (RPM or %), and core clock. Those five numbers make making undervolting decisions way easier.

The safest undervolt workflow (works for any GPU)

1. Establish a stock baseline (10–15 mins). Run your benchmark of choice and your game of choice at your normal settings. Details to record: average FPS, GPU temp, hotspot temp, GPU power, fan RPM/%.
2. Set a realistic goal. Example “Keep Perf within ~1-3% of stock while reducing fan noise,” or “Reduce hotspot temp by 5-10F (3-6C)”
3. Do the easy win first: lower the power limit slightly (optional but safe). Try -5% to -10% and re-test. This alone often reduces temps and fan noise. (pcworld.com)
4. Move to undervolting (manual or automatic). Make a small change, apply it, then immediately test.
5. Stress-test briefly after each change (5–10 minutes), then do a longer stability run once you’re close to your target.
6. When you find instability, back off to the last stable setting (don’t try to “push through” crashes).
7. Save your stable undervolt as a profile, and keep a ‘Stock’ profile saved too so you can revert instantly. (us.msi.com)

What instability looks like (so you know when to stop)

• Game crash to desktop
• Driver reset / black screen for a moment, then recovery
• System reboot
• Visual artifacts (sparkles, flashing polygons, texture corruption)
• Benchmark hangs or fails to complete

Method 1 (AMD Radeon): Use Adrenalin’s “Undervolt GPU” first

If you have an AMD Radeon GPU, start with the built-in automatic option. AMD Software: Adrenalin Edition includes an “Undervolt GPU” tuning option designed to reduce voltage while maintaining clock speeds to improve performance-per-watt. (amd.com)

1. Open AMD Software: Adrenalin Edition.
2. Navigate to the Performance / Tuning area (wording can differ depending on version).
3. Accept the tuning EULA if prompted (auto-off). (amd.com)
4. Choose Automatic Tuning, then select “Undervolt GPU” (if available for your model). (amd.com)
5. Apply the applied and re-run your baseline tests

If your results are good, you can stop here—or continue reading to learn more about manual tuning to take more control (next section).

When to use AMD manual tuning instead of automatic undervolt

• You have a specific noise/temperature target and want to dial in fan behavior and/or frequency ranges.
• Auto undervolt is doing well, but you suspect you can go lower for more power savings.
• Auto undervolt isn’t available on your specific GPU model (AMD indicates it’s only present on select GPUs). ( amd.com )

Method 2 (Most NVIDIA cards): Undervolt with MSI Afterburner’s curve editor

On many NVIDIA cards, the most repeatable manual undervolt is to set a “cap” voltage and then tell the GPU what clock it should aim for at that voltage, using the voltage/frequency curve.

MSI Afterburner includes a Curve Editor and can save multiple profiles letting you A/B your results. ( us.msi.com )

1. Install MSI Afterburner from an official source and fire it up.
2. Locate and open the Curve Editor from within Afterburner (MSI’s changelog mentions using the Curve Editor from the main window). ( us.msi.com )
3. Fire up a game for a few minutes at stock so you have a ’typical’ boosted clock and voltage to reference from load and to use to help pick a sensible target point.
4. Pick a target voltage point on the curve (e.g. 900–975 mV is a typical range most people begin testing at on modern cards, but your GPU may need higher)
5. At that voltage point, set a frequency that’s close to your real, sustained in-game clock from your baseline (not the single highest spike clock)
6. Flatten the curve to the right of your chosen voltage point so the GPU doesn’t try to boost higher voltage states
7. Click Apply
8. Immediately save the setting to a profile slot so you can revert quickly. (us.msi.com)
9. Test for stability (see the stress-testing section below). If unstable, raise voltage slightly or lower the target frequency, then test again.

A practical example (numbers are illustrative, not a promise)

Workflow example: If your card typically sustains ~2,700 MHz while bouncing round ~1.00–1.05 V in a demanding game, you might try 2,700 MHz at 0.975 V first. If stable, try 0.950 V at the same clock, if that crashes, revert to 0.975 V or drop the target clock slightly (for example, -15 to -30 MHz) and re-test.

How to stress-test an undervolt (and what “pass” means)

Stress tests are designed to reveal instability quickly running a sustained heavy GPU load. UL (3DMark) notes that stress testing is useful after upgrading a GPU or overclocking/tuning, and 3DMark Stress Tests loop a graphics test continuously. (support.benchmarks.ul.com)

1. Do a short ‘sanity’ test first (5–10 minutes). If it crashes quickly, your undervolt is too aggressive.
2. Run a longer test once you’re close to your final settings. In 3DMark, the default stress test is 20 loops and takes about 20 minutes. (support.benchmarks.ul.com).
3. Check for artifacts, driver resets, or crashes while the test runs.
4. If using 3DMark results, UL explains that the stress test reports Frame Rate Stability (%) and that passing requires at least 97% plus completing all loops. (support.benchmarks.ul.com).
5. After the benchmark, do a real-game test for at least 30–60 minutes in the titles you actually play (some games are more sensitive than synthetic tests).

What to record (so you can prove the undervolt helped)

• Average FPS (and 1% lows if you track them)
• GPU power (watts) during the same repeatable scene
• GPU temp and hotspot/junction temp
• Fan RPM/% at steady state (after a few minutes)
• Noise (optional): a phone dB app at a consistent distance can show relative improvement

Undervolting vs. power limit vs. underclocking (quick comparison)

Common mistakes (and how to avoid them)

• Changing too many sliders at once: make one change, then test.
• Using an unrealistic test only: include at least one real game you play.
• Forgetting to save a known-good profile: keep ‘Stock’, ‘Daily Quiet’, and ‘Benchmark’ profiles.
• Assuming lower voltage always equals lower temps: if your GPU boosts higher clocks because it’s cooler, temps may drop less than expected (but power/noise can still improve).
• Over-optimizing fans: an ultra-low fan curve can make temps spike and cause clock throttling. Tune voltage first, fan curve second.

Troubleshooting: quick fixes when things don’t behave as expected

If the GPU crashes or shows artifacts

1. Revert to your last stable profile (or hit Reset)
2. Slightly increase the voltage at the target frequency, OR slightly reduce the target frequency at the same voltage.
3. Re-test with the same benchmark loop so that the results are comparable.
4. If the crashes only happen in one game, make a per-game profile (AMD supports per-game tuning profiles; other tools can be used in a similar manner). (amd.com)

If you want a simpler NVIDIA option than manual curve tuning

NVIDIA’s newer NVIDIA app has some automatic tuning and allows you to tune targets directly, such as voltage, power, temperature, fan speed, etc. It’s not a “classic undervolt curve guide” but if you want a lower-effort way to reduce the heat/noise of your graphics card without diving into curve editing, you can check it out. (nvidia.com)

FAQ