Undervolting and the CPU Efficiency Curve in Benchmarking
This article explores the impact of undervolting a processor on its power efficiency curve during computer hardware benchmarking. By reducing the voltage supplied to the CPU while maintaining or increasing its clock speeds, undervolting significantly lowers power consumption and heat generation. Consequently, this process shifts the processor's efficiency curve favorably, allowing the hardware to achieve higher performance-per-watt metrics, avoid thermal throttling, and sustain peak boost clocks for longer durations under heavy benchmark loads.
Understanding the Processor Efficiency Curve
Every processor operates on a pre-defined Voltage-Frequency (V/F) curve set by the manufacturer. To reach higher clock speeds (frequency), the processor requires more voltage. However, the relationship between voltage and power consumption is not linear; it is exponential.
The dynamic power consumption of a processor is roughly calculated using the formula: Power = Capacitance × Voltage² × Frequency.
Because voltage is squared in this equation, a small increase in voltage results in a massive spike in power draw and heat output. On a graph plotting performance against power consumption (the efficiency curve), efficiency drops sharply as the processor approaches its maximum clock speeds.
Why Manufacturers Use Excess Voltage
Chip manufacturers produce millions of processors, and due to microscopic variations in silicon quality—often called the "silicon lottery"—no two chips are identical. To ensure that even the lowest-quality silicon functions stably at advertised speeds, manufacturers apply a default V/F curve with a safety margin of extra voltage, known as a guardband.
For the vast majority of processors, this extra voltage is unnecessary and only serves to generate excess heat and waste power. Undervolting is the process of manually reducing this guardband to find the minimum voltage required for a specific chip to remain stable.
How Undervolting Shifts the Curve in Benchmarks
Benchmarking software is designed to push computer hardware to its absolute limits, exposing the inefficiencies of the default V/F curve. When you apply an undervolt, you alter how the processor behaves during these stress tests in three distinct ways.
1. Lower Power Draw at Fixed Frequencies
If a processor is locked to a specific frequency during a benchmark, undervolting directly reduces the wattage required to maintain that speed. Because the performance remains identical while power consumption drops, the performance-per-watt metric increases. On an efficiency graph, this shifts the curve toward using less power for the same workload.
2. Higher Clocks at Power Limits
Modern processors from Intel and AMD use automated boosting algorithms that push clock speeds as high as possible until the chip hits a predetermined power limit (measured in Watts) or thermal limit.
During a heavy multi-core benchmark like Cinebench, a stock processor will quickly hit its maximum power limit. When you undervolt the processor, each megahertz of frequency costs less power. Because the chip is no longer hitting its wattage ceiling as quickly, the boosting algorithm uses the newly freed power headroom to push the clock speeds higher. In this scenario, the efficiency curve shifts upward: you get more performance at the exact same power draw.
3. Elimination of Thermal Throttling
Power consumed by a processor is converted almost entirely into heat. During extended benchmarking sessions, a stock processor may exceed its safe operating temperature, forcing it to lower its clock speeds and voltage to cool down—a process called thermal throttling. This causes benchmark scores to drop.
By reducing the voltage, temperatures drop significantly. This keeps the processor below its thermal throttling threshold, allowing it to maintain maximum boost clocks throughout the entire duration of the benchmark.
The Limits of Efficiency: Instability
While undervolting improves the efficiency curve, it has a hard limit. If the voltage is reduced too far, the electrical signals inside the processor will fail to register correctly during complex calculations.
In a benchmark, this manifests as two things:
- Clock Stretching: The processor reports a high clock speed, but internally skips cycles to prevent crashing, resulting in drastically lower benchmark scores.
- System Crashes: The benchmark software will freeze, or the operating system will encounter a fatal error (such as a Blue Screen of Death).
Finding the optimal efficiency curve requires iterative benchmarking—lowering the voltage in small increments and running stress tests until the exact threshold of stability is found.