How the numbers are made.

Review sites rank graphics cards by raw performance. A card that is 20% faster but 60% more expensive is not a better buy, just a more expensive one. This index ranks by what a frame costs.

The score

For each card and each view (a resolution and a rendering mode), cost per FPS = current price ÷ average FPS. Lower is better. There is no weighting and no opinion in the number. The default view is 1440p, rasterization.

Value zones on the 1440p, rasterization board are fixed dollar thresholds: great under $8, fair up to $12, poor above that. Other views use percentile bands (the cheapest third, the middle third, the rest), because a frame at 4K costs more than a frame at 1080p and fixed thresholds would not mean the same thing.

Benchmarks

FPS comes from Tom's Hardware's GPU benchmarks hierarchy, which tests every card with the same games at the same settings. We use their rasterization and ray-tracing averages at 1080p, 1440p and 4K "ultra". Upscaling and frame generation are not included.

Cards from an older test suite

Tom's replaces its game suite every couple of years and does not retest every old card. Newer games are heavier, so an older card's FPS from the old suite would look inflated next to a current card's FPS from the new suite.

To compare them we look at the cards that were tested in both suites and fit a straight line from old-suite FPS to current-suite FPS. Cards that only exist in the old suite are translated along that line and marked est.on every page. Cards present in the current suite always use their measured number.

Current fit for 1440p, rasterization, 2026-03 to 2026-09-16:current = 1.116 × older + 4.2, fitted on 28 of 30 overlapping cards, R² 0.96. Excluded as outliers: RX 9070, RTX 4070 Ti Super.

Each point is a card tested in both suites. The solid line is the transform in use; the dashed line is the single ratio (1.170) the index used before, which over-credited slower cards and under-credited faster ones.

Cards currently shown with estimated FPS (16):GTX 1080 Ti, RTX 2060, RTX 2060 Super, RTX 2070, RTX 2070 Super, RTX 2080, RTX 2080 Super, RTX 2080 Ti, RTX 3080 12GB, RX 6500 XT, RX 6700 10GB, RX 6800, RX 7900 GRE, Intel Arc A580, Intel Arc A750, Intel Arc A770.

Prices

Each card's price is the lowest valid retail price we can find for a new, in-stock unit. A price is rejected when it is for a bundle, a laptop part, an open-box or refurbished unit, or when it falls outside 35% to 300% of that card's recent median (a sign of a mislabelled listing). Every price shows its retailer and the date it was checked. Prices are stored as history, so each card page shows how it has moved.

Performance and price describe the same GPU, not the same box. The FPS figures are Tom's Hardware's results for the GPU, usually measured on a reference or Founders Edition card. The price is the cheapest qualifying card built on that GPU by any board partner (an MSI or ASUS model, for example), so a particular card can run slightly faster or slower than the benchmark. Configurations that differ in hardware are always separate entries: the RTX 5060 Ti 8GB and 16GB are ranked apart and never share a price.

What this does not tell you

Averages hide game-specific behaviour. A card can be great value on average and weak in the one game you play.
Memory matters beyond FPS. An 8 GB card can post a good average today and struggle in newer titles.
Power draw, noise, size and features (encoders, upscaling quality) are not in the score. They are listed on each card's page.

The pipeline, data and this site are open source: github.com/SiliconValueIndex/siliconvalueindex.