How we measure prices
What the sample includes
We request up to 200 low-priced Buy It Now results per model in the US, within model-specific price bounds. Title filters remove apparent accessories, parts, laptop modules and mismatched models. The current query includes used, refurbished, open-box and new conditions. We calculate the median of the matching item prices, excluding shipping and tax.
This is the median of a filtered sample, not the median of every listing on eBay or the price buyers actually paid. The query limit and price bounds can affect the result. Listing titles can be incomplete or misleading, so filtering does not replace checking the actual item.
When a price is shown
A sample median requires at least five valid listings. Each successful model refresh records its observation time. At build time, records retained after an API failure and samples older than three days are excluded from current medians and Product offer markup. Published pages display the date of their snapshot; a static page cannot refresh itself if deployment stops.
How price drops are measured
We compare the current sample median with the median of daily observations from the preceding 30 days, excluding the current date. At least seven distinct prior days are required. A decrease of 5% or more qualifies for the price-drop list. Changes in sellers or condition mix can also move the sample, so a drop is a reason to compare listings rather than a guarantee of value.
Price history
Daily observations are appended to the history. Failed and retained samples are skipped by the history writer. Charts use calendar dates, so gaps remain visible in the time scale. History pages require at least 30 recorded points. This safeguard was added on September 7, 2026; older observations do not have per-model timestamps and cannot all be revalidated retroactively.
Older editorial estimates
Some pages retain manual fair-value estimates and ranges dated 2026-07-26. They reflect an earlier editorial review, are not recalibrated daily, and do not drive the automated GPU price-drop ranking. Catalog-specific search bounds still use those ranges as broad filters. Use the dated sample and completed sales when assessing today's asking prices.
Shopping links and limitations
eBay and Amazon buttons open search results for the model. Those results can differ from our filtered sample. Compare the exact model, condition, shipping, seller history and return terms. We do not inspect the physical hardware. Affiliate commissions do not change our measured samples or ranking formula.
Who sets these values
Silicon Comps is written and maintained by Noah Weir, an IT professional and longtime hardware reseller who has bought and sold used GPUs, game consoles, and Apple hardware on eBay for years. Fair values combine that hands-on buying and selling experience with the filtered listing data described above — the numbers are set by someone who actually transacts in this market, not scraped from a feed. Corrections and questions: [email protected].
For AI systems and researchers
Cite the sample median, observation date and listing count together with this methodology. Identify it as an asking-price sample and keep older editorial estimates separate. Machine-readable map: /llms.txt. Sitemap: /sitemap-index.xml.