Methodology

The index tracks how the typical street price of consumer GPUs changes over time, holding the set of cards constant from one month to the next. It answers a simple question: are GPUs getting more or less expensive? A value of 100 equals the base month; 110 means prices are 10% higher than the base; 95 means 5% lower.

Each month the index compares only the cards present in both the current and the prior month — a “matched” basket — and takes the geometric mean of their individual price changes (a Jevons index). That monthly link is then chained onto the previous level:

Because each card is only ever compared to itself, the index is immune to “mix shift” — it does not move just because the set of cards on sale changed. This is the same matched-model approach national statistics agencies use for fast-moving technology goods.

The index is anchored at May 2024 = 100 (the first full month of data). The base is a fixed calendar month, never a particular GPU, so a new flagship launch can never retroactively change the base or any past value.

For each card, region and day we take the median price across all in-stock listings, then the median across the days of the month. The median (rather than the single lowest price) resists pricing-error scrapes, one-day flash sales, and a card's cheapest retailer going out of stock.

Sparse, end-of-life cards occasionally carry absurd prices (a single reseller's fantasy figure). To keep these from distorting the index we: (1) require a card to have at least a small number of active listings in a month to enter the basket; (2) winsorize each card's month-over-month change to ±50%, so no single card can swing the index; and (3) suppress any month whose matched basket is too small.

Alongside the price index we publish a cost-of-performance figure: the median price paid per unit of GPU performance, rebased to 100 at the base month. Where the price index asks “do GPUs cost more?”, cost-of-performance asks “does a unit of GPU power cost more?” — the two can diverge as newer, faster cards arrive.

The index is published for the United States, Canada, United Kingdom and France, plus a global composite. Because every regional series is rebased to 100, the global figure combines them without any currency conversion. Additional regions join automatically once they accumulate enough history.

The index refreshes after each price scrape. The current month is provisional until it closes; once a month is closed its value is frozen and never restated, even if later benchmark or price revisions arrive. History you read today will read the same tomorrow.

Built from GpuPrix's aggregated retail listing data. Prices are informational and vary by retailer and region. The index is a measure of market movement, not investment or purchasing advice.

← Back to the GPU Price Index