Graphics cards.
Every graphics card we can source. What each one is designed to do.
Manufacturer specifications for 88 graphics cards, with comparisons between them. What a card holds once cooling becomes the limit is the question that decides how a machine feels.
Every graphics card we hold a record for.
Verified against a published specification. The rest render, unindexed, until they are.
78 desktop.
Nearest six per part. Any pair on request below.
Graphics card specifications, checked and compared
This is the SystemCheck graphics card catalogue: manufacturer specifications for 88 graphics cards (memory, clocks, shader counts and rated board power), each sheet checked against a published source, with 323 side-by-side comparisons between close rivals. 78 are desktop parts and 10 are laptop parts. The wider catalogue holds 119 records; the other 31 stay out of the index until their sheets are complete enough to publish. A figure we could not verify reads "Not verified" rather than a guess.
There are three ways to use it. Search for your graphics card below to read its full sheet and its sources. Open a published comparison when you are choosing between two graphics cards, or use the picker further down for any pair we have not published. Or, if you already own the machine, skip the sheet and run the free 3-minute browser test, which loads the processor, memory and graphics in turn and records what each delivers the whole way through.
A specification sheet describes what a graphics card is designed to do. Two machines with the same part can behave very differently once their cooling and power settings come into play, and no sheet records either. That is why every graphics card page here ends with what the sheet cannot tell you, and why the test measures the machine rather than the model number.
Good places to start
Find a graphics card.
88 graphics cards with a full sheet, grouped by family. A check marks a sheet verified line by line against the manufacturer.
AMD Radeon RX 400
AMD Radeon RX 5000
AMD Radeon RX 500
AMD Radeon RX 6000
AMD Radeon RX 7000
AMD Radeon RX 9000
AMD Radeon RX Vega
Intel Arc A-series (Alchemist)
Intel Arc B-series (Battlemage)
NVIDIA GeForce GTX 10
NVIDIA GeForce GTX 16
NVIDIA GeForce GTX 700
NVIDIA GeForce GTX 900
NVIDIA GeForce RTX 20
NVIDIA GeForce RTX 30
NVIDIA GeForce RTX 40 Laptop
NVIDIA GeForce RTX 40
NVIDIA GeForce RTX 50
NVIDIA GeForce RTX 50 Laptop
Compare
Any two. Including pairs we never published.
The published comparisons cover the pairs people search for. This covers the rest, including a laptop part against a desktop one, with a note on why that gap is wider in practice than on paper.
Choose two parts above and their specifications will appear here, side by side.
Questions
From manufacturer specifications. Each graphics card page lists the sources its figures were checked against, and any value we could not check renders as "Not verified" instead of being filled in from elsewhere.
A graphics card is indexed once its sheet is complete enough to publish. The other 31 still have a page if you reach one, but stay out of search and out of this list until the missing figures are verified.
Yes. The 323 published comparisons cover each part's nearest rivals. The comparison picker on this page builds any other pair from the catalogue on request.
The clocks on a sheet are what the part is designed to reach. What it holds under a long load depends on the cooling and power limits of the machine around it. The free 3-minute browser test measures what your machine actually sustains and names the shape of the curve.
No. No browser exposes temperature, fan speed or power draw. Where a SystemCheck report names a likely cause, it is inferred from the throughput curve and labelled as an inference, and a sensor tool confirms it.
Every figure on a specification sheet is a design target. None of it is a measurement.
Nothing above describes how a graphics card behaves twenty minutes into a sustained workload, which is where two chips with nearly identical specifications routinely diverge. A boost clock is a ceiling the vendor permits, not a rate the chip sustains, and whether it holds depends on the cooling, the case, the power profile and the room. None of those appear on a specification sheet.
We deliberately do not reduce these figures to a single score. A composite invented from specifications would look authoritative and measure nothing, and choosing the weights would be choosing the winner.