How to measure FPS and compare runs fairly

Define the question before collecting numbers

Decide what you want to learn. Testing whether a graphics setting improves a particular scene is different from comparing two complete PCs or reporting typical performance across a game. A short repeatable benchmark can answer the first question more reliably than the last. Choose one primary outcome, such as the average frame rate over the route, and a secondary outcome such as a frame-time distribution. Define the route and duration before seeing which configuration wins.

Record enough context to reproduce the run

Write down the exact CPU and GPU models, memory capacity, game build and driver version. Record output resolution, render scale, upscaling mode, frame generation, frame cap and display mode. Include the benchmark tool and its version because measurement behaviour can change. NVIDIA documents changes to FrameView FPS reporting in its release notes; results from different releases should not automatically be pooled. If a setting is unavailable on another machine, state that limitation instead of claiming identical conditions.

1. Create a named baseline configuration.
2. Record the route and capture duration.
3. Keep game and measurement versions alongside the result.

Capture a controlled workload

Use a built-in benchmark when it represents the problem you care about, or define a repeatable manual route. Online matches introduce player, server and scene variation that limits precise comparisons. Let the machine reach a similar operating state before each capture. Avoid counting menus and loading transitions accidentally. Keep background applications and recording software consistent. If you are investigating first-run behaviour, measure it separately and label it clearly rather than blending it into a warm-run average.

Read matching statistics

Average FPS describes the capture as a whole and can hide short interruptions. Look at frame-time behaviour and a consistently defined low-frame statistic too. Do not compare one tool’s percentile with another tool’s differently calculated low metric merely because their labels look similar. An instantaneous overlay is useful feedback, but a saved capture offers a stronger basis for comparison. Frame generation and presentation behaviour also affect what a reported FPS number represents; record these modes instead of treating every frame counter as interchangeable.

Repeat and publish the limits

Capture several runs per configuration and show the spread. Returning to the original baseline helps reveal drift caused by background work or changing operating conditions. A small difference inside the normal run-to-run spread is inconclusive. Preserve the raw files and explain exclusions before summarising a result. PC Universe stores user-entered measurements and can analyse supported frame-time CSV files, but it does not independently certify a benchmark. Never turn one scene into a claim about every map, game patch or hardware combination.

Verify with the official sources

NVIDIA FrameView official overview ↗

NVIDIA FrameView version 2.0 release notes ↗

Educational workflows, not a claim of hands-on testing across every device. Retest after a game, browser, driver or network change. Editorial policy →