What is RunBenchmarks

One place for all your benchmark data.

The idea behind RunBenchmarks is simple: one place to check real FPS performance across different hardware and graphics settings, before making hardware or settings decisions.

Real Gameplay Data

Every result comes from actual in-game sessions, not synthetic benchmarks or cutscenes.

No Sponsored Results

No one influences what gets tested or how results are presented. Benchmark data reflects real performance — nothing more, nothing less.

Free & Open

No account required. No paywalls. All benchmark data is freely accessible to everyone.

Methodology

How I benchmark.

Every benchmark session follows a strict, repeatable process designed to produce consistent, reliable results. Every run is based on real gameplay — played within a broader area like a city district or a forest, keeping the gameplay natural and representative.

01

Settings changed on the fly — no restart between configurations

The game keeps running between configurations. Only the graphics or hardware settings are changed, then the next run starts. A full restart only happens when the game itself requires it to apply a setting — never as a routine step.

02

Getting into position before recording

Before hitting record, I play through the test route informally for a short while to get into position and let the scene settle. This isn't a separate measured phase — it's just preparation that happens before the actual benchmark starts.

03

6 benchmark runs — 60 seconds each

Once recording starts, 5 consecutive benchmark runs are captured back-to-back, each lasting 60 seconds, all within the same session.

5 × 60s runs

04

Deviation filtering — ±5% threshold

The system calculates the median across all 6 runs and selects the 3 most consistent ones — those with a deviation below ±5% from the median. Runs outside this threshold are automatically rejected.

±5% deviation threshold

05

Final result — average of 3 selected runs

All metrics you see on the platform — Avg FPS, 1% Low, CPU Usage, Frametime, histogram and more — are calculated from the 3 accepted runs by CapFrameX. The most representative of the 3 runs is the one available to watch as the reference video.

Consistency

Action and location — the two rules.

Running 6 benchmark passes per configuration only works if the gameplay itself is consistent. That's why every session follows two core rules:

Action: I always try to perform the same type of in-game activity during each run. Mixing combat and exploration, or walking vs. driving, leads to results that aren't comparable. Keeping the action consistent means the GPU load stays comparable across runs.

Location: I stay within the same general area of the game — same city district, same environment type. I don't need to follow the exact same path pixel-for-pixel, but I don't switch between completely different environments mid-run either.

Together, these two rules make the benchmark results repeatable, fair, and directly comparable across different hardware configurations and graphics settings.

Data Collection & Analysis

How data is collected and analyzed.

Two dedicated tools handle all data collection and overlay management during every benchmark session:

RivaTuner Statistics Server (RTSS) manages the real-time performance overlay visible during gameplay, without interfering with the benchmark results.

CapFrameX is responsible for capturing, analyzing, and aggregating all performance metrics across every run. All results you see on the platform come directly from CapFrameX analysis.

Recording Setup

Why a second PC for recording?

To ensure 100% accurate benchmark results, gameplay footage is captured using a dedicated second PC connected via an Elgato 4K Pro capture card. This approach completely separates recording from gaming — the gaming PC runs only the game, with no encoding overhead affecting frame times or CPU load.

Elgato 4K Pro — dedicated capture PC

All gameplay footage is captured on a separate machine using the Elgato 4K Pro capture card. The gaming PC is completely free of any recording software, encoding processes, or background tasks that could influence benchmark results. What you see on the platform is what the GPU actually delivered — nothing more, nothing less.