Overclocking raises a processor's operating frequency beyond its rated speed. The performance gained is real, and it is paid for in three specific currencies.
Power rises faster than speed
A chip's power draw scales with frequency and with the square of the voltage it needs. Higher frequencies usually require more voltage to stay stable.
That combination means a modest frequency increase can demand a disproportionate increase in power. The relationship is not linear at the top of the range.
This is why gains flatten quickly. The last few percent of frequency costs far more than the first few, and the cost is paid continuously.
Heat is the practical ceiling
All that extra power becomes heat, which must be removed faster than it is produced. Cooling capacity therefore sets the real limit rather than the silicon.
When temperature rises past a threshold the chip reduces its own frequency to protect itself. An overclock that cannot be cooled produces a lower sustained speed than leaving it alone.
This is why cooling upgrades often deliver more than frequency adjustments. Removing the thermal ceiling lets the chip's own boost behaviour run longer.
Stability margin is being spent
Manufacturers rate a chip conservatively so it works across a range of temperatures, workloads and years of use. That margin is deliberate.
An overclock consumes it. The system may pass a stress test and still fail hours into a specific game that exercises an unusual instruction pattern.
Instability presents as crashes, corrupted saves or graphical faults that look like game bugs. Diagnosing them is difficult precisely because they are intermittent.
Modern chips already overclock themselves
Processors and graphics cards now run boost algorithms that raise frequency automatically whenever power, temperature and current allow. The chip is continuously seeking its own limit.
Manual overclocking therefore competes with a system that has better information than the user does, including per-chip characteristics measured at the factory.
The remaining headroom is consequently small on recent hardware. Large manual gains belong to an earlier era when boost behaviour was crude or absent.
Undervolting inverts the trade
Reducing voltage while holding frequency lowers heat, which lets the boost algorithm sustain higher speeds for longer. The result can be more performance, not less.
It carries the same stability risk, since the chip is again running outside its validated envelope, and it requires the same patient testing.
For laptops and small cases this is usually the more productive direction. The limiting factor there is heat removal, and undervolting attacks it directly.