How SpecMatchup publishes GPU data

The current catalog covers desktop GPU models that have a reviewed record and a stable public identifier. Missing fields are omitted or described as not verified instead of being guessed.

1. Official NVIDIA or AMD product pages, specification pages and technical documents.
2. Official board-partner pages for partner-specific details or archived variants.
3. A reputable secondary hardware database only when the manufacturer no longer provides the required legacy specification.

Secondary sources are labelled and do not silently override a conflicting first-party value. Search snippets, retailer listings, forum posts and generated summaries are not treated as factual evidence.

Launch-price records use first-party NVIDIA or AMD newsroom releases, product announcements, presentations or archived product pages. Search snippets, retailer listings, forum posts and secondary launch tables are excluded. Historical launch prices are not current offers and are not adjusted for inflation.

Each amount remains in USD and keeps the price basis used by the cited source. MSRP means the manufacturer's suggested retail price while SEP means suggested e-tail price. Starting price, manufacturer direct price and retail price remain separate labels rather than being renamed MSRP.

Each price point also keeps its variant scope. That scope identifies the supported memory configuration, Founders Edition status or board coverage so one launch amount is not presented as the price of every version grouped under a catalog model.

The first-party review verified an initial USD launch price for 92 of 99 GPUs. No verified first-party initial USD launch price was found for these 7 models:

• NVIDIA GeForce GTX 980 Ti
• NVIDIA GeForce GT 1030
• NVIDIA GeForce GTX 1630
• AMD Radeon R9 Fury
• AMD Radeon R9 380X
• AMD Radeon RX 6400
• AMD Radeon RX 7900 GRE

The Radeon RX 7900 GRE's later USD 549 global retail launch is a separate February 27, 2024 event. It is not labelled as the original July 27, 2023 launch price.

The last verified date means a reviewer opened the cited evidence and rechecked the complete set of values displayed for that GPU. It is not the date the page was deployed, imported or automatically scanned.

SpecMatchup preserves the manufacturer's published label. NVIDIA commonly uses Total Graphics Power (TGP) while AMD commonly uses Typical Board Power (TBP). Older or partner material may instead say Board power.

These terms describe vendor-published board-level power guidance but are not guaranteed to be identical lab measurements. Differently labelled values are shown with their original labels and are not presented as perfectly equivalent.

Reference specifications describe a chip vendor's published baseline or reference design. A representative partner image shows one retail board built around that GPU. Its cooler, dimensions, connectors and power limits can differ from the reference design.

A model sold with multiple memory configurations lists each published capacity separately. One capacity is not used as if it represented every version. If another specification varies by memory option but has not been sourced for each version, that difference is omitted.

The fixed cohort is Blender 4.5.0 on Windows with NVIDIA OptiX for NVIDIA GPUs and AMD HIP for AMD GPUs. Each published number is the median of estimated Cycles samples per minute summed across the standard Blender Benchmark scenes. Higher is faster only for this rendering workload.

Established means 20 or more compatible submissions. Provisional means 5 to 19 compatible submissions. Fewer than 5 compatible submissions are recorded as insufficient and their numeric median is not published. A missing exact match also has no numeric median.

Results from different Blender versions, operating systems or backends are not pooled. CUDA and OptiX are not combined. A cross-backend comparison is limited because backend implementation can affect the result. This is not gaming performance or an overall GPU rating.

Comparison pages describe published memory, power, compute and clock fields. The key limit is simple: specifications do not measure gaming performance. Core counts and clock speeds are architecture-specific and are not treated as direct performance equivalents across different GPU designs.

Records are rechecked when credible new manufacturer material appears or when a visitor reports a disputed field. The public value changes only after the evidence is reviewed and the whole displayed record is checked again.

To help with that review, send the exact page and disputed field through the correction process.