Load-Line Calibration Explained: Vdroop, Overshoot, and Picking the Right LLC Level
Every CPU voltage regulator is designed to let output voltage sag under load on purpose, and load-line calibration is the setting that decides how much of that sag the board fights back against—push it too hard and you trade a mild, predictable droop for a voltage spike the CPU never asked for.
Vdroop is not a defect. When a CPU jumps from idle to a heavy all-core load, current draw through the VRM can swing by tens of amps in a fraction of a millisecond, and the inductors and capacitors in the power delivery path cannot respond instantly. Left completely uncorrected, that transition would let voltage sag noticeably below the target the moment load hits, then bounce back once the regulator catches up. Rather than fight this physically unavoidable lag with brute force, CPU vendors specify an intentional, small, controlled voltage droop as load increases, so the effective voltage at the die actually falls slightly under heavy load compared with idle. This gives the regulator headroom to absorb a sudden current spike without the voltage ever exceeding its safe ceiling, because it started a bit lower than nominal to begin with.
What Load-Line Calibration Actually Adjusts
Load-line calibration is the motherboard setting that changes how aggressively the VRM compensates for that designed-in droop. At the lowest LLC setting, the board mostly leaves the natural droop alone, so voltage measured at full load can sit noticeably below the value set in the overclocking software or BIOS. Raising the LLC level tells the regulator to actively boost its output as load increases, clawing back some or all of that droop so the voltage under load stays closer to what was actually requested. This sounds like it should always be a good thing, but the correction only has value up to the point where it exactly cancels the droop; beyond that point, the board is over-correcting, and the failure mode flips from undervoltage to overvoltage.
The Overshoot Problem at High LLC Levels
The trouble shows up specifically at the other end of a load transition, when current draw suddenly drops rather than rises—for instance, the moment a benchmark run ends or a game finishes loading a scene and the CPU relaxes back toward idle. A VRM tuned to aggressively fight droop during the load-up phase tends to overshoot during the load-down phase, briefly pushing voltage above the requested set point before settling back down. At a moderate LLC level this overshoot is small and transient, often just a handful of millivolts lasting a fraction of a second, and mostly invisible outside of an oscilloscope trace. At the most aggressive LLC settings a board offers, that overshoot can spike well past the requested voltage, and a voltage spike above spec, even a brief one, is a more concerning failure mode over time than a bit of droop ever was, since sustained overvoltage transients are one of the more plausible contributors to gradual VRM and CPU degradation on an aggressive daily overclock.
Reading the Actual Voltage, Not the Requested One
A common mistake when tuning LLC is trusting the voltage value entered in the BIOS or shown by monitoring software as if it were what the CPU actually receives. That number is the target, not a measurement. Motherboards with real hardware voltage-read points on the PCB let you confirm actual VRM output voltage with a multimeter under sustained load, and the gap between the requested figure and the multimeter reading at a low LLC setting is exactly the droop that setting is failing to correct. Software monitoring tools report sensor data pulled from the same VRM controller that LLC is adjusting, so at very high LLC levels the reported voltage and the transient overshoot spike can both look deceptively clean in a logging tool that only samples periodically, while a scope would show a brief spike the logger simply missed between samples.
Picking a Sensible Starting Point
| LLC Behavior | Typical Result | Best For |
|---|---|---|
| Low / disabled | Noticeable droop under load, minimal overshoot risk | Stock or mild all-core clocks, stability-first builds |
| Moderate (board's mid setting) | Small droop, negligible overshoot | Most sustained overclocks; the default starting point for tuning |
| High | Droop mostly cancelled, small transient overshoot | Overclocks that are otherwise stable but showing droop-related crashes at full load |
| Extreme / maximum | Droop fully cancelled, measurable overshoot spikes on load-down | Rarely worth it outside of extreme benchmarking sessions with active monitoring |
A practical approach is to start at a board's middle LLC option rather than jumping straight to the highest one, then only step up if a specific instability pattern points at droop—typically a crash that happens right as a sustained load begins rather than during steady-state operation. Because VRM temperature under sustained overclocking load already sets a practical ceiling on how hard the power stages can be pushed, adding an aggressive LLC setting on top of a marginal VRM cooling situation compounds both problems at once: more corrective current draw generates more heat in the same phases that are already running hot. On boards with generous VRM headroom, a higher LLC level is a more reasonable experiment; on boards where VRM temperatures are already a limiting factor, it usually is not worth the added stress for a correction that a moderate setting already handles adequately.
The interaction with core-voltage tuning tools matters here too. Anyone running an undervolt or a per-core offset, such as the approach described in AMD's Curve Optimizer per-core tuning, is deliberately narrowing the safety margin between requested voltage and the CPU's actual operating limits, which makes an overshooting LLC setting more consequential than it would be on a stock or conservatively overclocked chip. A voltage spike that would have been harmless headroom on a stock configuration can matter more once every other margin has already been trimmed. Testing LLC changes in isolation, one step at a time, with a sustained load-cycling stress test rather than a single static benchmark, is the only reliable way to see whether a given level actually helps or has quietly introduced a transient the daily workload never previously produced.