Modern GPUs from both NVIDIA and AMD are factory-calibrated to hit performance targets across a wide range of silicon quality — which means most individual cards have significant headroom for voltage optimization. Undervolting your GPU can reduce power consumption by 20–40 watts, lower temperatures by 10–15°C, and often actually improve sustained performance by preventing thermal throttling — all without losing a single frame per second. This guide explains how to do it.

Why Undervolt Instead of Overclock?

The conventional wisdom is to push for more MHz. But for most modern GPUs, the performance curve is already very close to its voltage-frequency limit. You can gain a few percent from overclocking, but:

Higher voltage = more heat = thermal throttling under sustained load
Power limits hit before thermal limits on many cards
A GPU running at 85°C will throttle clocks to protect itself, losing the OC gains

Undervolting finds the minimum voltage needed to sustain a given frequency. The GPU runs at the same clocks but draws less power, generates less heat, and sustains those clocks more consistently under load. On many cards, an undervolted GPU outperforms a stock one in long sessions.

Tools Needed

MSI Afterburner — primary tuning tool (works on NVIDIA and AMD)
EVGA Precision X1 — NVIDIA alternative with similar voltage curve editor
GPU-Z — verify readings before and after
3DMark (free tier) or Unigine Superposition — benchmarking
HWMonitor or HWInfo64 — thermal monitoring

NVIDIA Undervolting with Voltage/Frequency Curve

Step 1: Baseline Testing

Before touching anything, run a benchmark and record your stock performance:

3DMark Time Spy score
GPU Package power draw (from HWInfo64)
Peak GPU temperature under load
Clock speeds during the benchmark

Step 2: Open the Voltage/Frequency Curve Editor

In MSI Afterburner, click Ctrl + F to open the Voltage/Frequency Curve editor. This graph shows the relationship between voltage (X axis, in mV) and core frequency (Y axis, in MHz) for your specific GPU.

Each point on the curve represents a voltage-frequency operating point. The right portion of the curve (higher voltages) represents what your GPU uses at maximum load.

Step 3: Find Your Target Frequency

Look at the curve during gaming (let the graph stabilize under a game load, then pause and observe). Note what frequency your GPU boosts to — for example, 2100 MHz. This is the frequency you want to maintain.

Also note at what voltage this frequency is currently running — for example, at 1000 mV.

Step 4: Apply the Undervolt

The goal is to get that same frequency (2100 MHz) at a lower voltage (for example, 875 mV). In the curve editor:

Select all points to the right of your target voltage (select by clicking and dragging)
Hold Ctrl and drag downward to flatten the right portion of the curve
Left-click your target voltage point (e.g., 875 mV) and drag it up to your target frequency (2100 MHz)
This creates a “staircase” — everything to the right of your chosen point is capped at your target frequency/voltage

Click Apply (checkmark in Afterburner).

Step 5: Test for Stability

Run your benchmark or a demanding game for 20–30 minutes. If the system crashes, BSoD, or the display goes black:

The voltage is too low for that frequency
Increase the voltage by 25 mV and test again
Repeat until stable

Most RTX 40-series cards will run their boost clocks stably at 850–925 mV (vs. the stock 1000–1050 mV). Variation between individual cards is significant — a process called the “silicon lottery.”

Step 6: Save the Profile

Once stable, save the profile in Afterburner (select a profile slot and click Save). Enable “Apply overclocking at system startup” in Afterburner settings so it loads automatically.

AMD Undervolting with AMD Software

Method 1: AMD Software Adrenalin (Easiest)

Open AMD Software: Adrenalin Edition → Performance → Tuning:

Set to Manual mode
Enable GPU Tuning
Lower the Maximum Frequency to your tested stable frequency
Reduce Voltage by 50–100 mV below stock
Click Apply

This is a simpler approach than the NVIDIA curve editor — you’re setting a flat frequency/voltage cap.

Method 2: MSI Afterburner Curve Editor (RDNA2/3)

Same process as NVIDIA — Ctrl+F opens the curve editor on modern AMD cards. Some RDNA3 cards respond better to Afterburner’s approach than AMD’s own software.

Typical Results by GPU Generation

GPU	Stock Power	Undervolted Power	Temperature Drop
RTX 4070	~200W	~155W	10–15°C
RTX 4080	~320W	~260W	8–12°C
RX 7900 GRE	~260W	~200W	10–15°C
RTX 3080	~350W	~290W	10–15°C

These are approximate typical results — your card’s response varies by the silicon sample.

Power Limit Tuning

Separate from undervolting, you can reduce the Power Limit slider in Afterburner. Dropping to 80% power limit caps total GPU power draw, which reduces temperatures at the cost of some peak performance. This approach is simpler than curve editing but doesn’t offer the same performance-per-watt improvement because you’re not optimizing the voltage operating point — just capping total power.

For the best result, combine undervolting with a slight power limit reduction: the undervolt optimizes efficiency, and the power limit prevents edge cases where the GPU still hits its TDP.

Verifying Your Results

After tuning:

Rerun your baseline benchmark — scores should be equal or higher than stock
Check HWInfo64 for GPU Power (Package) — should be 20–50W lower than stock at similar frequencies
Check temperatures — should be measurably lower
Run a 30-minute gaming session to validate thermal stability

A well-tuned GPU is quieter (fans don’t ramp as high), runs cooler (better for longevity), draws less from your PSU, and often sustains higher average clocks than a stock card that’s thermally throttling under extended load.