Updates to the graphics card market usually run in bunches. AMD releases a new GPU and Nvidia responds, or vice versa, and then we wait for many moons before another major update rolls along. For most people, there's really only one reason you need a fast graphics card these days: gaming. Sure, professionals and supercomputers can use them as well, but here our focus is squarely on pushing pixels in the latest games.
It might sound crazy to spend hundreds of dollars on a piece of hardware primarily focused on games, but if you already own a decent PC, it's often less expensive than buying a modern console. Both Nvidia and AMD offer good performance at prices to fit nearly any budget, and we've picked the top cards for four categories.
Updated 12/15/2015 with new choices at all price points.
Okay, we know what you're thinking: "Why on earth would you choose the GTX 980 Ti over the GTX Titan X? Isn't the Titan X faster?" You would think so, and in some cases it is, but throughout a large suite of gaming benchmarks, we've never seen more than a three percent difference at reasonable settings (meaning, the cards don't choke on the workload). What's more, there are times when the 980 Ti actually leads by a small amount.
It all comes down to TDP, and with both cards targeting 250W and twice the GDDR5 on the Titan X, sometimes the little upstart wins out. So if you want to game at 4K with a single GPU but you're not willing to shell out a grand for the privilege, the 980 Ti saves you money and provides nearly identical performance. That makes it the better graphics card for almost any sane person, and it still has an impressive 6GB GDDR5 with a 384-bit memory interface.
If you need more than that, the cards we've tested all overclock like champions, often by more than 20 percent over stock clocks. Prices have also fallen since the 980 Ti launched, to the point where the 980 Ti can be had for as little as $600. The Titan X is still there as a step up if you want it, and owners of that card can rest easy knowing they still have the absolute fastest GPU currently around, but there's no beating the two-pronged performance and price punch of the "smaller" GM200 cards. And if you really want to go all-out, add in another card (or two!) for some eye-searing SLI performance.
The 980 Ti has other advantages over the Titan X, as manufacturers are free to deviate from the reference design as they see fit. That means you can find everything from standard dual-slot blower reference cards to monster triple-slot cards with three fans, or even hybrid solutions with factory liquid cooling.
Value-added extras like G-Sync support, HDMI 2.0 connectivity, ShadowPlay, and GameStream further sweeten the deal. Until the next round of 14/16nm FinFET cards launch in 2016, GM200 reigns supreme.
Not willing to spend quite so much money on a graphics card? For the price of a gaming console, or half the price of the 980 Ti, you can step down to the Radeon R9 390. AMD's Fiji cards ultimately couldn't dethrone the GM200, but their old Hawaii cards with some added VRAM and improved clock speeds still pack quite the punch. The 390 takes the existing 290 and doubles the memory, then gooses the RAM clocks from 5GHz to 6GHz for good measure, along with a slight increase in core clocks as well. For less than $300, you can comfortably run most games at 2560x1440 with maxed out settings, and performance is better than Nvidia's competing GTX 970 in nearly every title we've tested.
All is not sunshine and roses, however. For one, the R9 390 can guzzle more power than a GTX 980 Ti while providing less performance—GCN simply isn't as efficient as Maxwell. You'll also need both a 6-pin and an 8-pin power connector (or on some cards, two 8-pin connectors), along with plenty of room in your case, as the R9 390 is a very large card.
On the other hand, there's none of the 3.5GB shenanigans in the R9 390 that you'd get with a GTX 970; no sir, you get a full 8GB of GDDR5 memory here! And while load power of 250W might sound scary, when you're just surfing the web or doing mundane office tasks, the 390 is within a few watts of most other GPUs. AMD has alternatives to most of Nvidia's features as well, e.g., FreeSync instead of G-Sync, but you do miss out on HDMI 2.0 compatibility.
If you're looking for a DX12 card to keep you going until the 14/16nm updates launch, the R9 390 is a great option.
When we talk about entry-level graphics cards, we have to be clear: We still want good performance at 1080p, and we’re willing to pay a bit more for that privilege. If you’re willing to drop to medium quality and you don’t want to spend more than $100, we have one final pick below; here, we’re gunning for high quality 1080p at close to 60FPS.
AMD’s Graphics Core Next architecture has been immensely successful. Originally launched all the way back in late 2011, even four years later many of those old GPUs remain game worthy. GCN has seen a few updates over the years, however, and the Tonga architecture at the heart of the R9 380 is nearly the latest revision—Fiji has a few extras, but nothing critical. Now sporting 4GB VRAM (avoid the 2GB VRAM models!), Tonga is more than capable of playing games at 1080p, and more importantly, it’s almost always faster than Nvidia’s competing GTX 960, and it easily bests the GTX 950.
Fundamentally, R9 380 delivers nearly the same experience as the original GCN card, the HD 7970 (you can compare performance in our R9 380X review). That was a $550 card four years ago, so spending $180 for the same performance today—slightly higher in most cases, and with substantially lower power draw and noise levels—strikes us as great progress.
Note that this isn’t the fully enabled Tonga chip (that’s the R9 380X), but you’re only missing out on the last five percent or so of performance, and you’re saving nearly twenty percent. Meanwhile, the R9 380 bests the GTX 950 by 30 percent on average while leading by as much as 60 percent in games like Shadow of Mordor, all while costing 20 percent more, giving it the net victory. Having twice as much memory is also more forward looking, though the GTX 960 at least is available in 4GB trim as well.
We know where you’re coming from: $600 for a graphics card is fantasy land, $300 is way too much, and even $180 would break the bank. How about something closer to $100? Why yes, we can go that low. To be clear, there’s compromise involved. You’re not going to be gunning for 1080p at high quality at this price point, but 1080p at medium quality should be doable in most games, or you can dig into your backlog of games and play titles from a few years ago at higher quality settings.
There’s some good news, however, like the fact that the MSI GTX 750 Ti we’ve selected doesn’t require any form of external power. If you have a PCIe x16 slot on your system, you should be able to plop in the 750 Ti and start gaming. The card pulls all of its whopping sub-75W power over the PCIe slot, which means cooling isn’t much of a problem either.
We’re aiming pretty far down the totem pole at this point, so where the top card has 2880 CUDA cores and an entry-level GTX 950 still packs 768 cores, the 750 Ti only has 640 cores. These are Maxwell 1.0 cores as well (GM107 if you prefer), which means you lose some of the newer features like lossless delta color compression. It's still technically a DX12 part, but we're skeptical that will prove of much use on these cards. Thankfully, it’s going to be way faster than any integrated graphics chip, and you get all of this along with 2GB GDDR5 memory for $112.
If you’re hoping to go under $100, we have to advise caution. Yes, there are cards available, but many of them start cutting so many features that you’re only going to end up slightly faster than a modern APU like AMD’s A10-7800. Maybe you have an older CPU, though, in which case you could pick up something like a GT 740 1GB or R7 240 2GB, but at that point you’re really scraping the bottom of the barrel. Do yourself a favor and try to save up for at least a $100-class graphics card.
These are some great graphics cards and we have no qualms recommending them, but new technology is always just around the corner. And that corner right now is looking like Q1/Q2 2016, when we'll see the first 14nm and 16nm FinFET GPUs produced. Why should that matter? A short bit of history and some physics will help us out.
Our current GPUs are all being made using TSMC's 28nm process technology. In the past, every two years or so we'd see a shrink that would help improve performance by allowing for more transistors within the same space along with improving power and clock speeds. Two years ago, TSMC hit 20nm, but it was a process geared more towards smartphones than GPUs, and it was late. Apparently it was so underwhelming that both AMD and Nvidia decided to stick with 28nm and wait for the next process node.
That node is coming via GlobalFoundries (14nm) and TSMC (16nm) in 2016, and it's not just a process shrink as it will also add FinFET. Intel did this at 22nm and gave us 65W desktop Ivy Bridge with some impressive performance, but it's taken longer for other chip foundries to make the switch. The short summary is that the combination of a 1.5X process node jump and FinFET could very possibly double graphics performance at every price point next year.
So if you're willing to play the waiting game for a bit longer, the move to 14/16nm FinFET in 2016 will likely give us the biggest single jump in graphics performance that we've seen in more than a decade. Even if you don't grab a newer part, it should also push prices down quite a bit on existing GPUs.