Undervolting Mobile GPUs: Cooler, Quieter, Often Faster
Laptop GPU silicon is voltage-binned conservatively at the factory. The chip runs at more voltage than it needs to maintain stable clocks, which generates excess heat. Pulling the voltage curve down by 50 to 80 millivolts typically reduces sustained junction temperature by 6 to 10 degrees, cuts fan noise significantly, and in thermally-throttled laptops can actually increase sustained performance.
The logic behind undervolting is counterintuitive until you understand what happens in a thermally-constrained laptop. The GPU has a fixed thermal budget. If it reaches its temperature limit, it reduces clock speeds to bring heat generation down. By reducing the voltage at a given frequency, you reduce the heat produced at that frequency, which means the GPU can sustain higher clocks before hitting its thermal limit. Less voltage, same clocks, less heat—or same heat, higher clocks.
On a well-cooled desktop GPU with ample thermal headroom, undervolting delivers only noise and power savings. On a thin gaming laptop that thermal-throttles under sustained load, undervolting can be the single most effective performance improvement available, sometimes delivering more than an overclock would.
Tools: MSI Afterburner Voltage/Frequency Curve
For NVIDIA mobile GPUs, MSI Afterburner's voltage/frequency curve editor is the standard tool. Open Afterburner, click the Voltage/Frequency Curve icon (or press Ctrl+F) to open the curve editor. The X axis is voltage in millivolts, the Y axis is GPU clock in MHz. The default curve climbs from left to right—more voltage, higher clock.
The undervolting approach is to select a point on the curve that corresponds to your target maximum clock speed, shift that point down to a lower voltage, then flatten all points to the right of it by holding Shift+L after clicking your target point. This tells the GPU to run at your target clock while capped at the lower voltage, rather than stepping up to higher voltages for marginal clock gains.
Finding your target voltage point
Run a 10-minute sustained gaming load or OCCT GPU:3D test and read your GPU voltage and clock from HWiNFO64 while it is running under load. Note the sustained voltage and clock—not the peak, but the value that persists for several minutes. This is your baseline operating point.
Subtract 50 mV from that voltage as your first undervolting attempt. If your GPU is sustaining 1000 MHz at 900 mV during gaming, set your target point to 900 MHz at 850 mV, flatten the curve, and test. Use the stock clock as the target frequency (or slightly above it) to avoid a performance loss.
Stability testing
After each voltage change, test stability with a 15-minute OCCT GPU:3D run or a 20-minute gaming session. Instability from undervolting typically manifests as a black screen followed by a driver crash/recovery, or as hard system lockup. If the system is stable, reduce voltage by another 25 mV and test again. Continue until you find the instability point, then add 25 mV back as your safety margin.
Most Ada Lovelace laptop GPUs can be undervolted 50 to 100 mV below their factory operating point. Turing-generation mobile GPUs tend to have less flexibility, typically 25 to 75 mV. AMD RDNA2 and RDNA3 mobile GPUs have more voltage/frequency adjustment points built into their driver and respond well to undervolting via AMD Software Adrenalin.
AMD mobile GPU undervolting process
On AMD RDNA2 and RDNA3 mobile GPUs (RX 6000M and RX 7000M series), the approach is through AMD Software Adrenalin. Under the Performance tab, select Tuning, then Manual. Enable GPU tuning and access the Voltage/Frequency Curve. The process is conceptually similar to NVIDIA: find the sustained operating point under load, reduce voltage at that frequency point, flatten the curve.
AMD mobile GPUs in many laptop configurations also respond to SmartShift adjustments, which affect how power is distributed between CPU and GPU. If your AMD laptop has SmartShift, experimenting with the SmartShift balance slider in Adrenalin can yield additional performance in GPU-heavy workloads by allowing more power budget to the GPU without changing its voltage curve.
Measuring the results
Run your standard benchmarks before and after with full thermal soak, logging HWiNFO64 data. Compare GPU junction temperature at peak sustained load, GPU clock at thermal steady state, and benchmark score. On a thermally-throttled laptop you should see junction temperature down 6 to 10 degrees, sustained clock up 50 to 150 MHz, and benchmark score improved by 3 to 8 percent.
Fan noise improvement is harder to quantify but immediately audible. The GPU running cooler means the fan spins at a lower RPM to maintain the same temperature. On laptops where fan noise is a significant annoyance, undervolting is often worth it for the acoustic improvement alone, even before considering performance.
Combining undervolting with other modifications
Undervolting and thermal repaste are complementary. The repaste reduces the thermal resistance from die to heatsink; the undervolt reduces heat generation at the die. Together, they typically produce larger temperature reductions than either alone. On a laptop that was thermal-throttling significantly, the combined effect can be substantial: 15 to 25 degrees Celsius GPU junction temperature reduction at steady state, translating to clock speeds that no longer drop below the GPU's rated boost clock during extended gaming sessions.