The graphics card leaderboard built on measured frame rates
The graphics card is the component that decides what your games actually look like, and it is also the component most surrounded by noise: model names that hide giant performance gaps, memory sizes that sound bigger than they perform, and refresh cycles that arrive faster than most people upgrade. This leaderboard cuts through that with one consistent measure. Every graphics card in the database, more than 2,000 of them spanning two decades of releases, is ranked by averaged benchmark results across the major industry test suites, including PassMark, Geekbench's compute tests, 3DMark, and the standard graphics benchmarks, consolidated into a single score you can compare directly across generations and brands.
What separates this ranking from a generic benchmark chart is the measurement behind it. The site's game library holds recorded frame rates for 53 titles across more than 190,000 distinct processor and graphics card pairings, over two million individual results at 1080p, 1440p, and 4K on Low through Ultra settings. When a leaderboard position says a card is fast, that position can be verified against how the card actually ran real games on real machines, and the per-pairing pages show those measured numbers for any combination you are considering. The ranking orders the silicon; the game library shows what the silicon does for you.
How the GPU score is calculated
Each card's score averages its normalized results across every benchmark suite in the database. Averaging protects the ranking from single-test distortions: some suites reward raw memory bandwidth, some reward shader throughput, and some reward ray tracing or compute features that only part of the library exercises. A card that dominates one suite and trails in others lands where its overall capability puts it, which is the fairest default for a general leaderboard. Every card's detail page breaks the score back down by suite, so if your workload maps strongly onto one test, render-heavy work onto a compute suite, for example, you can weight your own decision accordingly.
Two things the score deliberately does not capture. First, price: a card two ranks higher at double the cost is not automatically the better purchase, and the price-performance leaderboard exists precisely to make that trade explicit. Second, feature set: ray tracing hardware, encoder blocks for streaming, and artificial intelligence upscaling features vary widely between cards at similar scores, and the per-card pages document them so you can weigh features the raw score treats as equal.
How to read this page
The rankings are organized the way people actually choose a card. The overall table ranks every tracked card by consolidated score, with NVIDIA and AMD filters for brand-level comparisons. Below it, dedicated sections rank the top 60 desktop cards, the top 60 laptop and mobile cards, and the workstation and professional class, because a 60 watt mobile chip competing against a 350 watt desktop flagship is a comparison nobody shopping for either machine needs. Every section pairs a visual chart of its top ten with a full ranking table, a search box that filters as you type, and a load more button that extends the list without ever leaving this URL. The entire leaderboard lives on one page, so what you see and what search engines index are the same complete picture.
Every tracked graphics card, every class, one consolidated score. Desktop flagships, mobile parts, and professional cards compete directly here; the category sections below then rank each class among its own kind, which is the comparison that actually matters when you know the machine you are building. Use the NVIDIA and AMD tabs above to narrow the field.
For a gaming desktop, the graphics card decision is the build. At the resolutions most people play, the large majority of the 53 games in the measured library are graphics-limited, meaning the card sets the frame rate ceiling and the processor's job is simply not to get in the way. The top of the desktop table is exactly what you would expect: the current flagship parts, ranked by consolidated score, with the measured game data confirming they deliver it where it matters, at 4K Ultra, where nothing else in the system can compensate for a weaker card.
The interesting purchasing action, as with processors, happens further down the table. Previous-generation flagships routinely hold positions ten to twenty ranks below the current leaders at prices that fall faster than their performance does, and the midrange of any current generation is engineered to hit the value sweet spot at 1440p. The leaderboard quantifies all of this directly: find the two cards you are comparing, read the score gap, and the measured pairing pages translate that gap into actual frame rates for the games you play at the settings you use.
Memory deserves a specific note, because it is the most marketed and least understood specification on the box. The VRAM column in the tables below matters when a game's assets genuinely exceed it, at high resolution with high texture detail in the newest open-world titles, where insufficient memory produces sharp, inconsistent frame times rather than graceful slowdown. Below that threshold, extra VRAM buys nothing, and memory bandwidth and GPU architecture matter far more than capacity. The per-card pages list all three, and the honest buying advice is to treat VRAM as a requirement to clear, not a number to maximize.
Laptop and mobile graphics: performance per watt is the game
Mobile graphics chips trade outright speed for the ability to live inside a laptop's thermal envelope, and the mobile leaderboard ranks them on their own terms. The current generation of mobile cards is remarkably capable: genuinely strong 1440p gaming performance now fits in mainstream gaming laptops, and thin-and-light machines with modest dedicated graphics run the full esports catalog at high refresh rates. The same caution that applies to mobile processors applies here with more force: identically named mobile chips ship across a range of configured power, and a card at its maximum wattage in a thick gaming chassis can outperform the same nameplate in a thin machine by a wide margin. Treat the ranks here as the typical case, and check the specific laptop's configured power before comparing prices.
For buyers choosing between a laptop with a stronger mobile card and a desktop with a midrange card at the same money, the measured data settles it cleanly: the desktop wins on frame rates nearly every time, and the laptop wins on everything else. Both conclusions are visible in the same leaderboard, because mobile and desktop cards are scored by the same method even though they are ranked in separate sections.
Workstation and professional cards: certified throughput
The professional section ranks the cards built for certified drivers, sustained compute, and memory capacity: the Quadro, Radeon Pro, Arc Pro, and accelerator families. For 3D rendering, video production, simulation, and machine learning workloads, these cards trade gaming performance per dollar for reliability, validation with professional applications, and VRAM capacities consumer cards do not offer. The ranking method is unchanged, so a shopper can directly compare a professional card against a consumer flagship and see the real performance relationship rather than the marketing separation. The used-market note applies here even more strongly than with processors: previous-generation professional cards retain large memory capacities and driver support long after their compute rank is surpassed, and the leaderboard makes their current standing against modern hardware plain.
One practical rule does more work than any other on this page: the resolution you play at determines how much of your card's ranking you can actually use. At 4K, differences between neighboring cards translate almost directly into frame rates, and buying up the table pays exactly what the score gap promises. At 1440p, the field compresses and mid-table cards deliver most of the flagship experience. At 1080p, even a modest modern card clears 60 frames per second in most of the measured library, and above roughly 100 frames per second the frame rate race only matters on high refresh monitors. Before paying for rank, check what your monitor can display; the measured pairing pages make the same point with numbers for any specific combination.
Ray tracing, upscaling, and the feature gap
Two feature families now separate cards that score similarly in traditional benchmarks. Ray tracing hardware determines whether the games that use it run well with the feature enabled or merely run at all, and in the measured library the ray traced titles show some of the widest gaps between same-generation competitors. Upscaling technologies, which render at a lower resolution and reconstruct the image with hardware assistance, effectively add a performance tier to every card that supports them well, and support quality varies by architecture generation. The consolidated score includes ray tracing tests where the suites provide them, but upscaling is a settings-level feature the score cannot see, so treat strong upscaling as bonus value on top of a card's rank rather than something the leaderboard already counts.
Cooling, power, and case reality
A graphics card is the largest, hottest, most power-hungry component in most builds, and its rank on paper assumes it can run at full speed indefinitely. The TDP column in each table is the guide: triple-fan coolers on flagship-tier cards exist because those cards need them, power supplies should be sized to the card's requirement plus the processor's with comfortable headroom, and compact cases need verified clearance for both the card's length and its slot width. None of this changes a card's rank; all of it changes whether the card delivers its rank in your machine.
The measurement backbone behind this ranking
The benchmark database aggregates results from the industry's standard graphics test suites across every card tracked, normalized so differently scaled suites contribute comparably, and refreshed continuously as new results and new cards arrive. On top of the synthetic layer sits the measured game library: 53 titles, more than 190,000 processor-and-card pairings, over two million recorded frame rate results covering three resolutions and four settings presets each. That second layer is what turns a ranking into a buying tool, because the question a graphics card purchase actually asks is not which silicon scores highest but which card runs your games at your settings on your monitor. Every leaderboard position here can be checked against that measured record, and the pairing pages exist precisely so you can run that check for any combination under consideration.
Pairing the leaderboard with the rest of the tools
A graphics card never performs in isolation; it performs inside a machine. After shortlisting cards here, the comparison tool puts any two cards head to head on their shared suites, the Can I Run It tool tests a specific card and processor against a specific game using measured results where they exist, Rate My PC scores the complete build and names the limiting component, and the price-performance leaderboard converts every score on this page into performance per dollar at launch pricing. The measured pairing data answers the question the leaderboard cannot, how a card behaves beside a specific processor, and together the two answer the only question that matters: which card belongs in your next machine.
Frequently asked questions
How often is the leaderboard updated? Continuously, from a database that ingests new results and new cards as they appear, with this page rebuilt on a short cache cycle so launches are reflected quickly.
Does a higher score always mean more frames in games? Usually, and the margin depends on resolution and settings. The measured pairing pages show actual game frame rates for the specific card and processor combination you are considering.
Why do some cheaper cards outrank expensive ones? The ranking is performance only. Value is a separate calculation, and the price-performance leaderboard covers it directly.
Are NVIDIA and AMD cards scored the same way? Yes, identical suites and identical averaging, so cross-brand rank comparisons are valid.
Is VRAM the most important specification? It is a requirement to clear for the games and settings you play, not a number to maximize. Bandwidth and architecture matter more once the requirement is met, and every card's detail page lists all of them.