Average FPS Is Not the Whole Story

The average answers one question

Average frames per second summarizes how many frames were presented over a period of time. It is useful for comparing overall throughput when the test conditions match. But it compresses a sequence of frame durations into one figure. The same average can come from a steady cadence or from a run with brief slowdowns balanced by faster moments elsewhere.

Imagine two hypothetical captures with the same total frame count and duration. In one, each frame arrives at a similar interval. In the other, most frames arrive quickly but a handful take much longer. The average can match while the second capture contains more obvious pauses. This is a teaching example, not a measurement from a game benchmark.

Frame time makes the gaps visible

Frame time is the time between frames, usually expressed in milliseconds. A consistent series is easier to interpret than a single FPS average because it shows when delivery changes. A sudden long interval can correspond to a hitch or visible pause; repeated variation can make motion feel uneven even when the average looks strong.

Capture tools such as PresentMon record per-frame events and metrics, which makes it possible to inspect a distribution rather than only a top-line number. The quality of the conclusion still depends on how the capture was collected: scene, route, game version, driver, background tasks and warm/cold state all matter.

Percentiles and “lows” are related, but not interchangeable

Percentiles describe a point in a distribution. A 99th-percentile frame-time threshold, for example, marks a boundary below which most recorded frames fall; it does not itself tell you why the slow tail exists. Tools also use different definitions for “1% low.” NVIDIA’s FrameView guide describes its 1% Low FPS as the average of the slowest one percent of frames, while percentile FPS metrics are thresholds. Similar labels can therefore summarize different calculations.

Always check the tool’s formula before comparing charts. A “low” based on averaging the slowest frame-rate samples is not necessarily the reciprocal of a frame-time percentile, and a short capture can make the tail sensitive to one unusual event. Percentile values are helpful summaries, not a replacement for looking at the trace and test conditions.

Smoothness has more than one cause

Stutter can arise from shader compilation, asset streaming, CPU scheduling, memory pressure, storage delays, driver behavior or an engine’s own workload. In our PC gaming evidence series, the SSD review discusses why storage can affect particular streaming tasks without predicting every game’s frame pacing. The VRAM review likewise explains that memory capacity is only one part of performance. Neither subsystem alone determines smoothness.

For a useful comparison, record the average and a frame-time trace, report a clearly defined low-percentile metric, and repeat the same route under the same settings. Separate loading screens from active play. Note unusual stalls instead of allowing one summary statistic to conceal them.

The reader’s rule of thumb: average FPS describes broad throughput; frame-time consistency describes how evenly that throughput arrives. Neither number alone says whether a game feels smooth to every player, on every display, in every scene.