Methodology

Every page leads with measured numbers and links to where they come from. Our own estimate fills the gaps, and this page says how it is made and how often it misses.

Measured results

We collect published average-FPS results: one row per game, graphics card, resolution and preset, with the URL of the chart it was read from. There are 2,137 of them across 150 games right now. Each row is matched to a game by name and the source's release year, and to a GPU by its model name. Rows we cannot match with confidence are left out rather than guessed.

When a source says the result hit a frame cap, was CPU-bound or ran out of video memory, we still show it, hatched and marked as a lower bound: the card can do at least that much. Results at 720p and other resolutions outside 1080p, 1440p and 4K are not shown.

Sources

• GameGPU: Per-game GPU charts of new releases, 1080p to 4K, average FPS at the maximum preset with ray tracing off.
• TechPowerUp: Game performance reviews, native resolution, no upscaling or frame generation.
• Notebookcheck: Per-GPU game tables, including integrated and laptop graphics and older games. Results come from many different test systems.
• YouTube and Reddit: Gameplay recordings and posts with an FPS counter, used for light 2D games on integrated graphics. Lower quality than the review sites: many are frame-capped and count as lower bounds.

Player reports are FPS figures submitted by players for a game, card and setting. We show the median per setting. Reports outside a quarter to four times our estimate, and reports for games without an estimate, are held back for review and not shown.

How the estimate works

The estimate is one multiplication, the same for every game:

FPS = game baseline × GPU factor × preset × resolution, then capped at the game's frame limit (or 360)

• Game baseline. A starting frame rate for the kind of game (about 40 for a typical big-budget game, 52 for the most demanding ones, 56 for indie games, 71 for esports titles, 110 for light 2D games), divided by the strength of the minimum graphics card on its Steam page. A game that asks for a strong minimum card gets a low baseline. The minimum is kept within 60–100% of the recommended card, so a stale or inflated listing cannot swing it far.
• GPU factor. The card's PassMark G3D score, stretched at the high end (G3D understates how much faster a flagship is in games) and flattened below 16,000 (integrated and entry cards). A few cards PassMark ranks out of line with game results, such as AMD's RX 9070 series, use a corrected score set from the measured data.
• Preset. Low ×1.25, Medium ×1.00, High ×0.80, Ultra ×0.65.
• Resolution. 1080p ×1.00, 1440p ×0.72, 4K ×0.39.

Constants as of October 2026. They change only after a new calibration run against the measured data.

How accurate it is

We run the model against every measured result it can be compared with (1,990 results in 142 games on 66 GPUs, last run 2026-10-06; 40 results where the model hit a frame cap are left out).

Mean error is geometric: ×1.43 means a typical estimate is off by that factor in one direction or the other. A bias near ×1.00 means the misses go both ways about equally. The ±30% band drawn around every estimate covers 48% of the measured results, so treat it as the likely range, not a guarantee.

Limitations

• The model knows the GPU only. Processor, memory, drivers and game patches are not in it; a slow CPU caps fast cards, most of all at 1080p.
• No ray tracing, upscaling (DLSS, FSR, XeSS) or frame generation. Turning them on changes the numbers a lot.
• Each source tests its own scene. Two reviews of the same game and card can differ by 20% or more, and we show both.
• Notebookcheck tables mix laptops with different power limits; laptop results vary more than desktop ones.
• Games without a usable minimum GPU on Steam get no estimate.
• Average FPS says nothing about stutter or 1% lows.

Found a wrong number or a benchmark we should add? Write to the address on the about page.