There are plenty of great graphics cards out there, no matter what you're looking for. Thing is, the odds are seemingly stacked against you ever finding the right one. It doesn't have to be that hard.
Whether you're buying a new computer, building your own or upgrading an old one, the process of choosing a new graphics card can be daunting. Integrated graphics solutions—the kind that come standard with many PCs—have trouble playing games from three years ago, let alone today, and will put you at a disadvantage when future technologies like GPGPU computing, which essentially uses your graphics card as an additional processor, finally take hold. The point is, you'll want to make the right choice. But how?
Set Specific Goals, Sight Unseen Your first step to finding the right graphics card is to just step back. Just as graphics card specs are nigh-on impossible to understand, naming conventions and marketing materials will do nothing except give you a headache. The endlessly higher numerical names, the overlapping product lines, the misleadingly-named chip technologies—just leave them. For now, pretend they don't exist.
Now, choose your goals. What games do you want to play? What video output options and ports do you want? What resolution will you be playing your games at? Do you have any use for the fledgling GPGPU technologies that are slowly permeating the marketplace? And although you may have to adjust this, set a price goal. Ready-built PC buyers will have to consider whatever upgrade cost your chosen company is charging, and adjust accordingly. For people upgrading their own systems, $US150-$200 has been something of a sweet spot: It'll get you a card with a new enough GPU, and sufficient VRAM to handily deal with mainstream games for a solid two years. If you want to spend less, you can; if you want to spend more, fine.
These are the terms that matter most. Seriously, disregard any allegiance to Nvidia or ATI, prior experiences with years-old graphics hardware or some heretofore distant, unreleased and unspec'd game franchise. Be decisive about what you want, but as far as hardware and marketing materials go, start blind.
Don't Get Caught Up In Specs Now that you've laid out your ambitions, as modest or extreme as they may be, it's time to dive into the seething, disorienting pool of hardware that you'll be choosing from. The selection, as you'll find out, is daunting. The first layer of complexity comes from the big two—Nvidia and ATI—whose product lines read more like Terminator robot taxonomies than something generated by humans. Here's Nvidia's desktop product line, right now:
It seems like you ought to be able glean a linear progression of performance (or at least price) out of that alphanumeric pile, right? Not at all. How in the world are we to know that the 9800GTX is generally more powerful than the GTX 250, or that the 8800GTS trumps a 9600GT? A two letter suffix can mean more than a model number, and likewise, a model number can mean more than membership in a product line. These naming conventions change every couple years, and occasionally even get traded between companies. For example, I've personally owned two graphics cards that bore 9x00 names—you just won't see them on the chart above, because they were made by ATI. Point is: You don't need to bother with this nonsense.
The next layer of awfulness comes from the sundry OEMs that rebrand, tweak and figure out how to cool offerings from the big two. This is what Sapphire, EVGA, HIS, Sparkle, Zotac and any number of other inanely named companies do. They can, on occasion, cause some sizable changes to the performance of the GPUs they're built around, but by and large, the Nvidia or ATI label on the box is still the best indication of what to expect from the product, i.e., a Zotax Gtx285 won't be that much better or worse than an eVGA or stock model. You'll get a different cooling solution, different hardware styling, and possibly different memory or GPU frequency specs, but the most important difference—and the only one you should really concern yourself with—is price.
Graphics cards' last, least penetrable line of defence against your comprehension is hardware jargon. Bizarre, unhelpful spec sheets are, and always have been, a common feature in PC hardware, from RAM (DDR3-1600!) to processors (12 MB L2 cache! 1333MHz FSB!).
Graphics cards are worse. Each one has three MHz-measured speeds you'll see advertised—the core clock, the CPU (shader) clock and the memory frequency. VRAM—the amount of dedicated memory your card has to work with—is another touted specification, ranging from 256MB to well beyond the 1GB barrier for gaming cards. On top of frequency, memory introduces a whole slew of additional confusing numbers: memory type (as in, DDR2 or DDR3); interface width (in bits, the higher the better); and memory bandwidth, nowadays measured in GB/s. And increasingly, you'll see processor core numbers trotted out. Did you know that Nvidia's top-line card has 480 of them? No? Good.
The best way to approach these numbers is to ignore them. Sure, they provide comparative evaulation and yes, they do actually mean something, but unless you're a bonafide graphics card enthusiast, you won't be able to look at a single spec—or a whole spec sheet—and come to any useful conclusions about the cards. Think of it like cars: horsepower, torque and engine displacement are all real things. They just demand context before they can be taken to mean anything to the driver. That's why road tests carry so much weight.
Graphics cards have their own road testers, and they've got the only numbers you need to worry about.
Respect the Bench, or Trust the Experts In the absence of meaningful specs, names or distinguishing features, we're left with benchmarks. This is a good thing! For years, sites like Tom's Hardware, Maximum PC, and Anandtech have tirelessly run nearly every new piece of graphics hardware through a battery of tests, providing the buying public with comparative measures of real-word performance. These are the only numbers you need to bother yourself with, and where those goals you settled on come into play.
Here's how to apply them. Say you just really want to play Left 4 Dead, and have about a hundred dollars to spend. Navigate over to Tom's, check their benchmarks for that particular game, and scroll down the list. You're looking for a card that is a) an option on whatever system you're buying and b) can handle the game well—at a high resolution and high texture quality—which, generally speaking, is a comfortable 60 frames per second. Find the card, check the price and you're practically done.
From there, your next worry will be buying for the future. You shouldn't buy the bare minimum hardware for the current generation of games—there's no need to spring for a card that'll be obsolete within a few months, no matter how cheap it is. But buying the latest, greatest graphics cards is an equally bad value proposition. As generations of video hardware have come and gone, one thing has remained constant: A company's midrange offerings, usually pegged at about $US150-$200, are your best bet, period. Sometimes they'll be new products, and sometimes they'll have been around a while. What you'll be buying, basically, is the top end of the last generation. This is fine.
Your alternative route is to just trust the experts. Sites like Ars Technica and Maximum PC regularly assemble system guides at various pricepoints, in which they've made your value judgments for you. At given price points, the answer will often be obvious, and these guys know what they're talking about.