In choosing video cards, the general rule is that you can check the SRP listings of the newest models each generation, and come up with a reasonable comparative for their relative performance. This isn’t very convenient to do locally, because a lot of stores have large delays between price drops here and price drops in the American market, where change is most convenient. It also isn’t useful if you’re looking at the price variances between board makers like Zotac and Palit, who can have striking differences in price for identical GPUs, differing only in cooling block and factory overclocks.
Deciphering model numbers then becomes a key skill needed for choosing a video card. It’s not actually as difficult as it looks, and some seemingly counterintuitive rules (like a “lower” model number, e.g. 6870, being stronger than a “higher” model number, e.g. 7770) are already known to most people who even dip their toes in the sea of consumer hardware. For completeness’s sake, though, all important rules will be listed here.
1) The first number, or highest decimal place, always refers to the generation of the video card. For example, the “7” in HD 7950, and the “6” in GTX 680. Both nVIDIA and AMD have always had comparable generations, even as their product release cycles have become staggered from each other lately. This is an extremely important rule because it immediately explains why lower numbers can sometimes be stronger than higher numbers. Each generation comes with improvements in performance, power consumption, and features for every outing, so a “7870” will almost always be stronger than a “6850.”
Exceptions occur when a company changes their naming convention. For example, in the past, AMD had, as its highest-end single video card for three generations in a row, the “3870,” “4870,” and “5870.” This changed with the “6970,” and the new “6870” was identical to the old “5870.”
2) The second number is the target market. In the “6” and “7” generation of AMD cards, a “9” (as in 7970/7950 or 6970/6950) meant the highest-end models, an “8” (7870, 7850) meant a high-midrange segment, and a “7” was a midrange card. Numbers below these, “6” downwards, were low-end OEM cards that could not be easily found on the market. In the last few generations of nVIDIA cards, which have been relatively consistent in naming convention since the GTX 200 days, a “9” is a high-end dual card, an “8” is a high-end single card, a “7” is a lower-tier high-end card, “5”s and “6”s are midrangers, and anything below are low-end OEM cards that are also hard to find.
3) The third number is typically exclusive to AMD video cards of the past few generations. It represents a further placement within the target market. Usually, only “7”s and “5”s are used – for example in the case of the two mid-rangers, the 7770 and 7750. A “9” is occasionally added to introduce a new video card to the target market that exceeds the “7” series but does not quite match the performance of the next target market. AMD model numbers of these generations always have “0”s as their fourth digits.
nVIDIA has had only three digits for model numbers in the past six generations, with the last digit almost always being a “0.” The only time this rule was broken was with the “GTX 275,” “GTX 285” and “GTX 465” cards. In order to introduce further gradations to their model numbers, nVIDIA has turned to the revival of the “Ti” moniker to denote stronger iterations of a model. For example, the GTX 560 Ti is more powerful than the GTX 560, and the GTX 780 Ti is more powerful than the GTX 780.
4) The current generation of AMD video cards, the R9/R7/R5 series, has a three-digit convention that is pretty much identical to nVIDIA’s. The “R” prefix at the beginning of the model number is, for all intents and purposes, useless, since it only re-indicates the price segment of the video card, which is done well enough by the second digit. For example, the R9 280X is stronger than the R7 260X, but the three-digit number alone already indicates that, with the “8” being higher than the “6”. For clarification, though, “R9” is high-end, “R7” is midrange-to-high, and “R5” is low-tier. In addition, further gradation between parts is accomplished by adding an “X” to the end, which denotes higher performance. For example, an R9 270X is stronger than an R9 270. The third digit is almost always 0, but three subspecies exist with the R7 265, R9 285, and R9 295X2.
This is actually very similar in practice to Intel’s naming convention of i7, i5, and i3, where the numbering after the prefix is ultimately the determinant of performance. However, Intel’s convention makes more sense with the “i” prefix, because each “i” tier actually has more in common than just their target market. Every desktop i7 is a high-end quad-core CPU with multithreading. Every desktop i5 is a midrange to high-end quad-core CPU without multithreading. Every desktop i3 is a dual-core CPU.
5) As a rule, a comparison with two video cards that have the same third digit/”X”/”Ti” convention and same second digit, but different generation, will always favor the newer card. For example, a GTX 980 is stronger than a GTX 780, and a GTX 660 Ti is stronger than a GTX 560 Ti. The exception is found in comparing to AMD’s HD 5000 and 6000 series, where a 6870 is actually a little weaker than a 5870, and this holds true across the entire product line below HD 6900.
This can be very clearly demonstrated when you consider the concept of “rebrands,” which are older video cards repackaged as iterations of higher generations. This sounds like a shady practice, but when actually implemented, the rebrands are slotted into lower-tier price segments in higher generations, where they would perform exactly as a hypothetical new release would. The HD 7870, for example, is rebranded in the form of the R9 270X, only improved slightly through better manufacturing processes, and a higher memory clock speed. As the HD 7870, it was one step below the highest tier, but as the R9 270X, it was pushed down to two slots below the highest tier, and was also given a lower price point.
It gets interesting when you consider the rebranding history of the GT 435M, a laptop video card from 2010. The card was rebranded into the famous GT 540M, the GT 630M, the GT 720M, and finally into the modern GeForce 820M, all of which are differently-labeled with almost-identical performance. Its target market number is shown to go down steadily with each generation, until it finally loses the GT, which is a convention for very low-end graphics cards.
6) A growing trend is to buck the numbering system for special or very high-end parts. The GTX Titan series represents the strongest possible video cards that can be purchased in a certain generation. The current single-GPU crownholder, the GTX Titan X, is a veritable beast that is 50% more powerful than the GTX 980 flagship. On the AMD side, the R9 Fury and Fury X are Team Red’s flagship graphics cards. Finally, the upcoming R9 Nano is a very small form-factor GPU that promises high performance per watt squeezed into a mini-ITX board.
7) Laptop graphics card numbers are not comparable to desktop parts. There is no reliable way to determine the relative performance of a laptop graphics card and a desktop graphics card based on their model numbers. Right now, the GTX 980M is between the desktop GTX 970 and GTX 960, while a GTX 970M is about as good as a GTX 960. The next part number after this, the GTX 960M, gets a precipitous drop towards the performance levels of a GTX 750 Ti. These three graphics cards all represent substantial advances over their predecessors in the GTX 800M series. However, nearly all the graphics cards below them are rebrands with almost no change to performance over the last generation. The GTX 950M is equal to an 850M, the GeForce 940M is equal to a GeForce 840M, and the GeForce 920M is equal to a GeForce 820M.
Laptop graphics cards numbers are among the most annoying to decipher because they actually can’t be, without being benchmarked. It’s unpredictable whether a manufacturer will decide to rebrand an entire graphics card line and sell them as “next-generation” parts as in the lower-end GeForce 900M series, or whether a massive performance boost will occur at the top level like in the case of the 980M and 970M. To make matters worse, there are often differences in the kind of RAM used – some individual graphics card models have separate versions that use either DDR3 and GDDR5 memory, but this distinction is not revealed in the name, or indeed even in common documentation. To figure this out, one must find out through GPU identification tools like GPU-Z, or through tech website articles.
8) nVIDIA has skipped a generation number twice already, most recently with the GTX 980/970. Don’t consider this to be anything more than a marketing tactic; the GTX 900 series is a predictable and expected performance increase over the GTX 700 series.
Intel’s apparent generation skip is more complicated – Broadwell, the generation after Haswell (i3/i5/i7-4000), was to be the i7-5000 or i5-5000 desktop part, but it was extensively delayed, to the point that when it was released, Skylake – the next generation – was mere months away. Today, the i5-6600K and i7-6700K are brand-new parts that are good upgrades over Haswell.
It’s a long list to read through, but the rules are easily-enough followed. Learning new conventions that are adopted by your favorite GPU manufacturer will also be trivially deciphered through a comparison of price points. The key has always been to select what suits your needs the most, for the best possible price. Some people would prefer to buy the cards one step below the highest tier, every two generations or so, for instance going from a 470 to a 670 to a 970. Other, more affluent buyers would like to stick to the highest tier every year. Upgrades don’t have to be expensive, and they don’t have to be hard to choose.