How DontLag estimates FPS
Every frame rate on this site is a model estimate, checked against 1,767 measured benchmark points in 84 games. This page explains how the number is made and how far to trust it.
Relative scores for graphics cards and processors
We do not store a frame rate per card per game. Each of the 261 graphics cards and 345 processors in our catalogue has one relative score. The GeForce RTX 4090 is the reference for graphics cards and scores 100. The Ryzen 7 9800X3D is the reference for processors and also scores 100. A card with a score of 50 does about half the graphics work of the RTX 4090. The scores are derived from published relative-performance tables and cross-checked with benchmark databases.
Each game has two demand numbers, one for the graphics card and one for the processor, that say how heavy the game is. The score of your hardware divided by the demand of the game gives a frame rate.
GPU-bound and CPU-bound FPS
A game can be held back by either part. We compute two frame rates and the lower one is the estimate:
- GPU-bound FPS depends on your graphics card, the preset, the resolution and any upscaling. It drops when you raise the preset or the resolution.
- CPU-bound FPS depends on your processor and not on the resolution. Lowering the graphics settings will not lift it. If the game has a built-in frame cap, the cap is a third limit.
When the processor is the limit we say so on the game page, because that changes what advice helps. Having less memory than the game recommends takes 15% off the processor-bound number.
Presets and resolution
Each game has a cost for the Low, Medium, High and Ultra presets, relative to High. By default Low costs 0.55, Medium 0.75 and Ultra 1.4 times what High costs. Games that scale differently carry their own values, fitted to their benchmarks.
Resolution scales the graphics work by the number of pixels drawn, with 1080p as the base. Moving from 1080p to 4K quadruples the pixels, but the frame rate does not fall by a factor of four, so we apply the pixel ratio to the power of 0.8. That exponent was fitted on measured results, not chosen by taste.
Upscaler modes
NVIDIA DLSS, AMD FSR and Intel XeSS render the game at a lower resolution and scale the image up. We model three modes by the share of output pixels that is actually rendered: Quality about 44%, Balanced about 34% and Performance 25%. The upscaling pass itself costs a little, so we take 8% off the gained frame rate. For each card we pick the best upscaler it supports in that game, and we only recommend a mode when the picture stays acceptable at the internal resolution it implies.
Calibration against measured benchmarks
A model is only useful if it matches reality, so every calibrated game carries benchmark points: a measured average frame rate for a named graphics card and processor at a given resolution and preset, with the source. Right now that is 1,767 points across 84 games.
We only take measured averages from reputable outlets that document their test system, such as TechPowerUp, ComputerBase, PCGH, TechSpot, GamersNexus, Digital Foundry and Notebookcheck. We do not use estimator sites, unlabelled videos, or benchmark numbers for a different resolution than the one we label.
We tune only the game's two demand numbers and its preset costs to fit the points. We never edit a measured number to make the model look better. An automated test enforces the result: each point has to land within 30% of the model, and the median error per game has to stay within 15%. A game that misses that bar does not ship as calibrated.
Confidence labels
Game pages carry one of two labels next to the frame rate:
- Based on measured benchmarks (calibrated): the game has at least three published benchmark points and the model passes the check above. The page states how many points it rests on.
- Rough estimate: no usable benchmarks exist yet, usually for a new or niche game. The demand numbers are derived from the publisher's specifications, so treat the number as a ballpark.
The range shown on a game page is a spread around the estimate, not a promise. Drivers, patches, background programs, cooling and the exact scene all move the real number.
Why spec targets are not benchmarks
A publisher's minimum or recommended specification says something like "1080p, Low, 60 fps" for a given card. That is a floor the publisher is willing to promise, not a measurement. The same goes for thresholds on other sites, such as "reaches 60 fps". We never enter such figures as benchmark points. For a game without real benchmarks we use them only as a floor, with headroom, and label the result a rough estimate.
How often the data is refreshed
Settings and benchmark data are checked by hand and each game file records when it was last verified. When a patch changes how a game performs, or when new benchmarks are published, we update the game's demand numbers and points. Meta loadouts and news follow the games' patches and season changes. The catalogue of graphics cards and processors grows as new hardware is released.
Report a wrong number
If a frame rate is far from what you see, mail [email protected] with the game, your graphics card and processor, resolution, preset and the average FPS you measured, plus the tool you used. A measurement from your own PC is the most useful kind of report. See About DontLag for how we handle corrections.