Methodology

We do not test hardware

GPURanker owns no test bench and runs no benchmarks. Every performance figure here is published by someone else and re-used with the source named. If you want first-hand testing, read the reviewers we cite — they did the work. What this site adds is the part they do not: putting 77 cards on one comparable scale, keeping the price next to the frame rate, and doing the arithmetic that turns the two into a purchase decision.

Where the benchmarks come from

The primary source is Tom's Hardware GPU hierarchy. Geometric mean of 15 games, ultra settings, no upscaling. Ray tracing figures are DXR-enabled at 1440p. Ray tracing and productivity columns are cross-checked against 3DMark Steel Nomad / PassMark and Technical City's GPU rankings; where two sources disagree materially we keep the one whose test conditions are documented.

Frame rates are absolute, not indexed — a 1440p figure of 90 means roughly 90 fps at ultra settings with upscaling off, not "90% of the fastest card". That is why the numbers can be divided by a price and still mean something.

The Value Score, exactly

Value Score is deliberately a price/performance measure, not an overall rating. A $2,000 flagship is the fastest card on the site and still scores poorly here — that is the intended behaviour, not a bug.

raw   = (fps_1440p_ultra / price) × 100 × ageMultiplier
score = min(raw / 30, 1) × 100        // 0–100, one decimal

ageMultiplier   1.00   released 2024 or later
                0.90   2022–2023
                0.75   2020–2021
                0.60   2019 or earlier

The age multiplier is a discount for things the frame rate does not capture: shorter driver support, missing upscaling generations, older encoders, and less remaining warranty. The divisor 30 is a fixed ceiling chosen so that the best performance-per-dollar card in the database lands near 100; it is fixed rather than derived so that one bad price entry cannot rescale every card on the site. A card with no usable price or no 1440p measurement scores 0 and is labelled Not Rated rather than ranked last.

Prices — read this before trusting a Value Score

This is the weakest part of the data and we would rather say so than bury it. Of 85 active cards, 1 currently carry an observed retail or used price; 84 fall back to MSRP, the manufacturer's launch price.

MSRP is not a street price. For sold-out flagships the real price is higher; for cards two generations old it is usually much lower. Where a card's price is an MSRP, its Value Score is a launch-day value score. Cards labelled with an observed price show when that price was captured. Used prices are eBay sold-listing averages, not asking prices.

Live pricing is not yet enabled. It will come from Amazon's official product API, not from scraping, which means it stays switched off until the account qualifies for API access. We would rather show a labelled MSRP than an unlicensed number.

Which pages exist, and which we ask Google to skip

A comparison page is published for any two cards, but only pairs within 30% of each other at 1440p — and released 2020 or later — are marked indexable. Two cards three tiers apart are not in the same purchase decision, and a page that exists only because two names can be concatenated is not worth a search result. Cards we hold no benchmark data for are kept out of the rankings and their pages are noindexed rather than padded with estimates.

What this data cannot tell you

• Noise, coil whine, thermals, and physical card size — these are per-model, and we track chips, not the dozens of board partner models per chip.
• Driver stability and per-game bugs, which change monthly and are not visible in an aggregated average.
• Frame-time consistency. A card can win on average fps and still stutter; only reviewers with frame-time captures can tell you that.
• Whether upscaling or frame generation would change the answer for you — all figures are native, no DLSS/FSR/XeSS.
• Regional pricing, tariffs, and availability. All prices are USD.

Corrections

If a number here is wrong, it is worth more to us than the traffic on the page that contains it. Send the card, the figure, and the source you trust to contact and it gets fixed at the data source, not just on the page you found it on. Affiliate relationships are disclosed on the about page and have never changed a ranking — the ranking is a formula, so changing one would mean changing the formula in public.