Tested: Nvidia GeForce and AMD Radeon graphics cards for every budget
That single, simple sentence cuts to the core of what people on the hunt for a new graphics card are looking for: The most oomph they can afford. Sure, graphics cards are complicated pieces of modern technology, powered by billions of transistors and countless other types of intricate hardware, but people just want to crank the detail settings on Far Cry and just plain play.
Answering the question can be a bit trickier than it seems. Raw performance is a big part of it, but factors like noise, the driver experience, and supplemental software all play a role in determining which graphics card to buy, too.
Let us be your guiding light. We’ve tested graphics cards of all shapes, sizes, and price points to nail down exactly what you can expect for your money—from itty-bitty $90 cards to fire-breathing $1000 models to behemoths with not one, but two graphics processors and custom watercooling loops. We’ll also talk a bit about the “extras” that can sway your buying decision, like Nvidia’s ShadowPlay software and AMD’s TrueAudio, and we'll issue buying recommendations for various price points. Finally, I’ll update this article with performance data from every new GPU that launches going forward, so you’ll always have the most recent information at hand.
Graphics cards are expensive. Choosing one can be complicated. But it won’t be after reading this. Let’s dig in.
Hold your horses! Before we dive into raw numbers we need to detail our test system and the cards we’ve tested. If you want to jump right into the juicy benchmarks, skip ahead to the third page. Buying recommendations are on the final page of this article.
Still with me Great. Here are the details of our test rig. Yes, it’s powerful—and definitely overkill for gaming—but that eliminates any pesky potential bottlenecking situations in the system. For more information you can check out our build guide for PCWorld’s graphics testing PC.
We tested as many different GPUs as possible—one GeForce GTX 750 Ti, one Radeon R9 390X, et cetera—with a preference for models with custom cooling solutions, in order to mimic as realistic a scenario as possible. Some high-end reference cards are also included, however.
Models of all current Nvidia GPUs have been benchmarked, but you’ll notice some missing Radeon models, such as the R9 285 and the R7 370. This is due to a few things: PCWorld’s graphics card review coverage was kind of light past years, and when AMD updated its entire graphics card lineup to the R300-series in one fell swoop, they didn’t send publications review samples. We managed to scrounge up high-end models, but nobody wanted to send us entry-level Radeon R7 graphics cards to review. That said, this is a fairly comprehensive look at the current landscape, from top to bottom, and I’ve included some older Radeon R200-series graphics cards since many are still available for purchase.
Speaking of…
Next page: Details about the graphics cards we tested.
Without further ado, here are details about the specific graphics cards we tested, from lowest-priced to highest. Click through each link for a full list of specs from AMD and Nvidia.
First up is the Nvidia GeForce GTX 750 Ti, which is still around despite the launch of the newer GeForce GTX 950. Most models cost between $100 and $120, depending on the included features. Stock specs include a 1020MHz base/1085MHZ boost clock, 640 CUDA cores, and 2GB of GDDR5 memory paired with a 128-bit bus. We tested a EVGA GeForce GTX 750 Ti Superclocked, which ships factory-overclocked at 1176MHz base/1255MHz boost. It includes single HDMI, DisplayPort, and DVI-I connections, but here’s the really nifty thing about the 750 Ti: This power-sipping graphics card requires no supplemental power connections whatsoever. It draws all its juice over the PCIe connection.
The $150 AMD Radeon R9 270X needs two 6-pin power connections, but it offers up to 4GB of RAM and 1050MHz clock speeds with the same 1280 stream processors. Prices typically range from $150 to $200 online. The VisionTek model we tested packs 1030MHz base/1080MHz boost clock speeds, a custom dual-fan cooling solution over a beefy heat sink with supplemental heat pipes, and—notably—a killer limited lifetime warranty for both parts and labor. It packs HDMI, DisplayPort, and both DVI-I and DVI-D connections.
The $160 Nvidia GeForce GTX 950 lures in entry-level gamers with solid 40-plus frames-per-second performance with mid-to-high settings at 1080p resolution. The stock version rocks 768 CUDA cores clocked at 1024MHz base clock/1188MHz boost clock, paired with 2GB of GDDR5 that’s clocked at a 6.6Gbps effective rate and chatting with the GPU over a 128-bit bus. The EVGA 950 Super-Superclocked model that we tested features EVGA’s slick ACX 2.0 cooler and jacks the clock speeds all the way up to 1190MHz base/1393MHz boost.
The newer $200 and up Nvidia GeForce GTX 960 packs the newer, supremely power-efficient Maxwell GPU architecture that first appeared in the GTX 750 Ti. While stock versions of card, which has 1,024 CUDA cores, are clocked at 1127MHz base/1178MHz boost, that energy efficiency allows graphics cards makers to apply beefy overclocks out of the box. The card we tested—the $210 EVGA GTX 960 Super Superclocked—is clocked at 1279MHz/1342MHz and requires an 8-pin power connector. You can find more details in PCWorld’s GTX 960 review.
AMD’s $200 Radeon R9 380, a souped-up version of the older Radeon R9 285, battles Nvidia’s GTX 960 with 1792 stream processors clocked up to 970MHz and either 2GB or 4GB of GDDR5 memory humming along at up to 5.7Gbps on a 256-bit bus. The VisionTek model featured in our Radeon R9 380 review —and this roundup—sticks to stock specs, but slaps on a dual-fan aftermarket cooling, a slick backplate, and an impressive limited lifetime warranty.
The AMD Radeon R9 290 packs 2,560 stream processors and 4GB of RAM with an ultra-wide 512-bit memory bus. It rocks a 662MHz base clock and 947MHz boost clock, with the same power pin setup as the R9 280X. These cards pack HDMI, DisplayPort, and dual DVI-D connections. We tested a reference model supplied by AMD.
AMD’s last-gen flagship graphics card, the Radeon R9 290X, rocks 2,816 stream processors. The card’s base clock speed is 727MHz, which boosts to 1000MHz when needed. Like the R9 290, its RAM utilizes a spacious 512-bit memory bus, with both 4GB and 8GB versions available. The R9 290X has the same power requirements and port configuration as the R9 290. We tested a reference model supplied by AMD.
The Nvidia GeForce GTX 970 boasts 1,664 CUDA cores, 4GB of RAM (ish) over a 256-bit bus, and a 1050MHz base/1178MHz boost clock. Prices start at $330. We tested an EVGA GeForce GTX 970 FTW with ACX 2.0 cooling (whew!), which—as the name implies—utilizes EVGA’s quiet, long-lasting ACX 2.0 cooling technology and boosts clock speeds to a hefty 1165MHz base/1317MHz boost, with plenty of room left for overclocking. This card delivers such a compelling price-to-power-to-performance ratio that it forced AMD to drop prices of its flagship R9 290 and R9 290X graphics cards by hundreds of dollars.
The $330 AMD Radeon R9 390 ups the ante over the old R9 290 by bumping the memory capacity to a full 8GB and memory clock speeds up to 6Gbps, as well as by boosting the core GPU clock speeds for the card to up to 1000MHz. We tested a Sapphire Nitro R9 390, which features Sapphire’s slick Tri-X cooling solution, long-life capacitors, Black Diamond chokes, and dual BIOS. It’s overclocked to 1100MHz out of the box and the build quality is damned slick.
What the new R9 390 is to the old R9 290, the $430 AMD Radeon R9 390X is to the R9 290X: A supercharged, more powerful—and power-hungry—version of the original, designed to take the performance battle to Nvidia’s competing GTX 980. As with the R9 390, it boasts 8GB of memory clocked at 6Gbps, but this rocks 2816 stream processors clocked at up to 1050MHz. The $440 Asus Strix version featured in PCWorld’s Radeon R9 390X review rocks the company’s potent, but beefy DirectCU III cooling solution. It’s overclocked out of the box to 1070MHz, but installing the included Asus GPU Tweak II software lets you leap that to 1090MHz at the push of a button, or higher if you want to manually overclock it.
Nvidia’s GeForce GTX 980 is powered by 2,048 CUDA cores. It features the same 4GB of RAM and 256-bit bus as the GTX 970, with clock speeds of 1126MHz base/1216MHz boost. You’ll need a pair of 6-pin connectors to power this 165-watt card, which sports HDMI, DVI-I, and three DisplayPort connections.
Now we’re getting into the really high-end, really interesting stuff. Midway through 2015, AMD rolled out a trio of graphics cards powered by its new Fiji GPU and revolutionary high-bandwidth memory . The $550 Radeon R9 Fury is the cheapest of the bunch, featuring 4GB of HBM clocked at a mere 1Gbps, but traveling across an insanely wide 4096-bit bus for a super-high 512GBps overall bandwidth. This 275W card features a step-down version of Fiji with 3584 stream processors clocked at up to 1000MHz, though the $570 Asus Strix version featured in our Fury review can be overclocked to 1020MHz with its GPU Tweak II software. The Strix Fury, like the Strix 390X, packs Asus’ effective, but bulky DirectCU III cooling tech to stay serious cool.
The $650 AMD Radeon R9 Fury X features the same memory configuration as the R9 Fury, but with the full-fat version of the Fiji GPU, weighing in with 4096 stream processors clocked at up to 1050MHz. Only reference versions of the R9 Fury X are available, and they ooze style with premium design touches like an integrated closed-loop water cooler and a slick “GPU Tach” strip of LEDs that lights up more and more the harder you push the card. It runs very cool and the card itself is small, too, measuring just 7.64-inches—though the bulky water cooler also takes up space inside your PC.
If you want a truly tiny mini-ITX graphics card capable of high-performance, no-compromises gaming, there’s only one option available: the radical AMD Radeon R9 Nano. Measuring a mere six-inches, the Nano fits in builds where other top-end graphics simply can’t. The Radeon R9 Nano features the same base memory and GPU specs as the Fury X itself, but uses an aggressive form of power management to dynamically scale the GPU core clock speed up and down to maintain reasonable temperatures and acoustic properties while still delivering decent 4K and top-notch 1440p gaming experiences.
Don’t let the Nvidia GeForce GTX 980 Ti’s name fool you: It utilizes the same gargantuan GM200 chip as the high-end GTX Titan X, rather than the GTX 980’s more modest (but still powerful) GM204. The 980 Ti packs 2816 CUDA cores and 6GB of high-speed GDDR5 RAM traveling over a 384-bit bus, but other than that, its design largely mirrors the Titan X’s, from the GPU clock speeds to the power requirements.
We’re including the EVGA GTX 980 Ti separately from the stock GTX 980 Ti. Why Because this customed-cooled variant features a beastly overclock that helps it triumph over all other single-GPU graphics cards that we’ve ever tested—including Nvidia’s flagship $1000 Titan X. This beast shows just what Nvidia’s Maxwell GPU architecture is capable of when it’s fully unleashed.
At the (theoretical) apex of the single-GPU lineup, there’s Nvidia’s GeForce GTX Titan X, a firebreathing graphics card with 3072 CUDA cores—compared to the GTX 980’s 2048—and 192 textures units. The card comes clocked at 1000MHz, with a boost clock of 1075MHz, and features HDMI, DVI, and a trio of DisplayPort connections. It’s the first single-GPU graphics card capable of playing games at 4K resolutions without looking like utter garbage. (We’ve also included benchmarks from the original 2013 Titan for reference.)
Whew! Still with me Good. With that out of the way, let’s dive into the performance benchmarks!
Next page: Performance benchmarks!
After all that preamble it’s time to dive into the heart of the situation: Which graphics card within your budget gives you the most bang for your back
We subjected every card to a gauntlet of synthetic benchmarks and real-world games to try and answer the question, measuring power use all the while. We subjected every card to a gauntlet of synthetic benchmarks and real-world games to try and answer the question, measuring power use all the while. Every game was tested using in-game benchmark tests, with the default graphics settings stated unless noted otherwise. V-Sync, G-Sync, FreeSync, and any vendor-specific features were disabled.
A couple of quick notes: All Radeon R9 290X figures are in “Uber” mode. Transitioning the card to “Quiet” mode usually only resulted in about a 1 frame-per-second difference in our tests. Results for the EVGA GTX 960 SSC are using the default “dBi BIOS.” Switching to the card’s “SSC Performance BIOS” typically improved frame rates by 1 to 3 fps in our games suite, with one glaring exception: The EVGA SSC hit 60.18 fps in Bioshock Infinite in dBi, but that leaped to a whopping 76.01 fps with the Performance BIOS.
Without further ado, let’s dig in.
Grand Theft Auto V's gorgeous, loving recreation of L.A. has a reputation for being a memory hog at higher settings and resolutions, and it comes with graphics options galore. We tested it four ways: at 4K with every graphics setting set to ‘Very High’ with FXAA enabled, at 2560x1440 with the same settings, and at 2460x1440 with the same settings but with 4x MSAA and 4x reflection MSAA also enabled. We dialed everything back to ‘Normal’ with MSAA disabled to test cards at 1080p. The frame rates are high at those settings, but you can always crank thing up from there. I’d suggest boosting Texture Quality to “High” first, because man, GTAV's street textures look butt ugly on Normal texture settings.
Dragon Age Inquisition was one of the best games of 2014. Under the hood, it runs EA’s Frostbite 3 engine—the same one that powers Battlefield 4.
Next page: More games benchmarks.
Middle-earth: Shadow of Mordor isn’t too punishing on graphics cards… until you load up the optional HD textures add-on that isn’t even recommended unless your GPU has 6GB of onboard RAM. Then it hits hard, which makes it great for testing high-end graphics cards at high resolutions.
The game was tested by using the Medium and High quality presets, then by using the ultra HD texture pack and manually cranking every graphics option to its highest setting (which Shadow of Mordor’s Ultra setting doesn’t actually do). Cards tested at 1080p weren't subjected to the ultra HD texture pack—we're not cruel.
Alien Isolation scales well across hardware of all price points.
Bioshock Infinite is an old standby that serves as a great stand-in for the Unreal 3 Engine. Unreal 4 has been announced, but it hasn’t begun showing up in mainstream games yet.
Next page: More games benchmarks.
Next up: Sleeping Dogs: Definitive Edition. This remaster of the thrilling action game cranked the graphics to 11, and even modern graphics cards have troubles hitting 60fps with details settings turned up. In fact, none did in our tests with all the bells and whistles turned on.
Likewise, Metro: Last Light Redux is a remake of the utterly superb Metro: Last Last, which you should absolutely play if you haven’t yet. It runs on 4A Games’ custom 4A Engine. We test the game with SSAA disabled because the feature cuts frame rates in half—and the game looks gorgeous enough without it.
Sniper Elite 3 is no graphics powerhouse, but we tested a few cards with it at 4K resolution anyway, to maintain a healthy mix to Nvidia-leaning and AMD-leaning games.
Next page: Synthetic benchmarks, power efficiency, and heat.
We also tested the cards with some synthetic, but well-respected benchmarks: Unigine Valley, 3DMark Fire Strike, and the 4K-resolution Fire Strike Ultra.
We’d be remiss not to talk about the power efficiency and temperatures of these GPUs. As you can see, Nvidia’s Maxwell GPU architecture is a power-sipping savant. AMD’s R9-series graphics cards run hot and loud in comparison, as measured using SpeedFan during a Furmark run, though models with aftermarket coolers are still plenty quiet for everyday gaming. The amount of power the tweaked R9 390 and 390X suck from the wall to stay competitive with Nvidia’s offerings is downright staggering.
Note that this is the power use of the entire system under load, not just the GPU, measured at the wall using a Watts Up meter during a Furmark run. Baseline system power use varies between 73W and 80W depending on the graphics card. The temperature measurement is of the graphics card only, however.
And look at that Fury X! The integrated water cooling setup helps AMD’s flagship run remarkably cool and quiet, though you can definitely hear the pump whirring a bit.
Next page: Noteworthy AMD and Nvidia extra features.
Beyond raw performance stats, both AMD and Nvidia offer a slew of extra features—normally software-related—to coax you toward Team Red or Green, respectively.
Some of those features are common to both companies, though each naturally puts its own brand on the technology and the technical implementations may be slightly different. Two relevant, standout examples: High-resolution downsampling and the quest to eradicate pesky screen tearing artifacts.
Recent graphics cards from both AMD and Nvidia allow you to choose to render games at resolutions higher than what your monitor actually supports—all the way up to 4K resolution—then apply a filter to downsample the image to your display’s native resolution in real time. Doing so provides a far crisper picture than you’d usually see, and you won’t have to muck with anti-aliasing, either. On the downside, rendering games in such high definition can put a big hurting on your frame rate, so you’ll only want to do this in games where you’re seeing ridonkulous performance already.
Nvidia’s implementation is dubbed Dynamic Super Resolution, while AMD calls theirs Virtual Super Resolution.
Both companies are also trying to eliminate screen tearing and stuttering by forcing your graphics card and your monitor to synchronize their refresh rates. Each implementation requires compatible monitors, however. Monitors supporting Nvidia’s G-Sync technology require an extra hardware module that drives up the cost of the display. AMD’s FreeSync monitors can work over a standard DisplayPort 1.2a connection—no extra hardware (read: cost) required—but don’t handle falling below 30 frames per second as gracefully as G-Sync panels.
Both implementations are utterly wonderful, however, delivering a buttery smooth experience you’ll never want to relinquish once you’ve used a compatible monitor. The fact that each brand of monitor only work’s with that story’s graphics card is worrisome, though, since most people use their monitors for five or 10 years. You’d be locked down to that brand’s graphics cards the entire time unless you don’t mind losing the variable refresh experience.
With that out of the way, here’s an overview of some of the highlight features for each individual brand.
Nvidia has a few aces up its sleeve.
On the hardware front, the new Maxwell GPU architecture is vastly superior to AMD’s R9-series cards in terms of thermals, noise, and power efficiency. It’s a night-and-day difference. If you’re building a power-constrained computer or a small form-factor PC where heat is a major concern, you’ll want to strongly consider going with an Nvidia graphics card.
Nvidia’s GameStream technology lets you stream full-blown PC games to a Nvidia Shield handheld or tablet, which you could then connect to your TV for a living room gaming sessions if you wanted to. GameStream holds up very well, streaming games at low latency on home networks in up to 4K resolution. It’s slightly less useful now that Steam’s killer in-home streaming feature is live and rockin’, however, and Nvidia’s GeForce Now “Netflix for games”-like service can’t beam games to your PC.
Some software features also stand out. Nvidia’s ShadowPlay is hands-down the best option for video recording your gaming sessions, delivering practically no hit on frame rates. Recent GFE updates added the ability to play local co-op games with your faraway pals as well as enhanced tools for streaming videos live to Twitch and YouTube gaming.
Next, Nvidia’s multi-frame-Sampled anti-aliasing (MFAA) smoothes out jagged edges similarly to traditional multi-sample anti-aliasing, but with far less of a performance impact—giving you the same level of eye candy with a decent-to-big frame rate boost. MFAA works in any DirectX 10 or DX11 game that supports MSAA; in fact, Nvidia’s GeForce Experience software enables MFAA by default in compatible titles. MFAA's a huge advantage for Nvidia.
Speaking of GeForce Experience, most gamers give Nvidia the edge when it comes to software polish and driver support, though AMD’s working hard to dispel that belief with initiatives like the Catalyst Omega driver. One potential point of concern: Nvidia’s plans to lock its vaunted Game Ready drivers behind GeForce Experience and an email registration in December. If you don’t want to hand over your email address or use GFE, you won’t be able to partake in those sweet, sweet launch day game drivers.
AMD holds some key advantages as well. It doesn't plan to lock drivers away behind email registration, for one, and Radeon cards skew tend to skew towards higher price-to-performance ratios, aside from the premium new Fiji-powered models.
AMD's most notable feature software-wise is Mantle, a graphics API that grants game developers “closer to the metal” access to Radeon hardware and eliminates CPU bottlenecks. In the right hands and with the right CPU/GPU configurations, the frame rate increases can be downright staggering. Developers can also opt to use Mantle to deliver far smoother performance rather than staggering frame rates when you're using a multi-GPU CrossFire setup, as Firaxis chose to do with Civilization: Beyond Earth.
There are some crucial gotchas though: Only a handful of games support Mantle, and the most mind-blowing performance increases typically come when you’re using a low-end processor or APU.
What’s more, AMD recently announced that Mantle is being killed, for all consumer-focused intents and purposes. Much of its core technology is being reborn in Vulkan, the recently announced Mantle- and DirectX 12-like successor to OpenGL, but that’s an open standard. Nvidia, Intel, and others will be able to tap into Vulkan just as easily as AMD. Existing Mantle-enabled titles will continue to work just fine, however.
But AMD may have an ace up its sleeve once DirectX 12 games start appearing en masse thanks to Mantle's groundwork. Early DX12 benchmark tests typically show AMD cards enjoying a far larger performance gain (versus DX11) than Nvidia’s GPUs—though that may be because Nvidia’s current-day drivers are so well optimized as is. It’s too early to call this advantage, since the first finished DirectX 12 games won’t even start appearing until the end of 2015, but it’s worth keeping an eye on the situation.
Our dives into the DX12 benchmarks in 3DMark and Ashes of the Singularity don’t compare AMD to Nvidia head-to-head—again, it’s way too early for that to gain concrete insights—but some excellent testing at PCPerspective and ExtremeTech pit the rivals head-to-head in Fable Legends and Ashes, respectively. You may also want to read our DirectX 12 primer to get up to speed on the radical new gaming tech.
Next page: The best graphics cards for your money.
Many charts and many thousands of words later, we’re finally ready to answer the question: Which graphics card within my budget gives me the best bang for my buck
$150-ish: If you’re looking to spend $150 or so, the $160 Nvidia GeForce GTX 950 is the clear winner, outpunching its Red Team competition (the Radeon R7 370) while offering slick extras like HDMI 2.0 and HDCP 2.2 support. That makes the card one of the few complete home theater PC options out there today—no high-end card from AMD or Nvidia can claim the same. The GTX 950 can game at mid-to-high graphics settings at 1080p without issues, and stick to mostly high settings in more games than not.
But it’s worth giving the Nvidia GeForce GTX 750 Ti honorable mention here, because it doesn’t need any supplementary power connections whatsoever. That, plus its humble 300W power supply requirement, means the GTX 750 Ti could add a big graphics punch to a low-end system with integrated graphics for just $100 to $120. Not shabby at all.
$200: Both AMD and Nvidia’s options at this crucial mainstream price point make captivating cases, delivering very playable frame rates with high or ultra settings at 1080p resolution. The GTX 960’s HDMI 2.0 and HDCP support make it the go-to for a home-theater PC, especially when paired with the card’s silence, coolness, and power efficiency. But if you simply want an affordable graphics card to slap into your PC and play games with, the AMD Radeon 380’s performance edge over the GTX 960 make it the more compelling option—despite its larger power needs.
$200 to $300: This is a gaping hole in the GeForce and Radeon lineups right now: There are no new-model graphics cards available between the $200 options and the step-up $330 options. If you can find an older Radeon 290 or 290X in this price range, though, grab it!
Oh, and note that in general, AMD’s newer R300 cards are retuned versions of older R200-series models, and not worth upgrading to a like model if you already own an R200-series card. You wouldn’t want to swap out your R9 290 for a R9 390, for example. Yes, there are gains to be found and more memory to be had, but not enough to be worth investing that much money into.
$300 to $400: Nvidia’s GeForce GTX 970 is a beast of a card at $330, despite the firestorm over its memory allocation design and incorrect initial specs. AMD’s Radeon R9 390 trades blows with it, depending on the title, resolution and graphics settings, coming out slightly—so slightly—ahead of Nvidia’s card more often than not. The R9 390 also offers a whopping 8GB of RAM compared to the GTX 970’s 4GB. That’s overkill for running games at the 2560x1440 resolution these cards excel at, but it makes the R9 390 a more compelling option for people looking to run multiple monitors or high-resolution monitors off multiple graphics cards.
But the GTX 970 overclocks like a champ and uses nearly half as much power as the energy-gobbling AMD card. You’d be happy with either card, to be frank, but you’ll have to decide which trade-off works better for you: The GTX 970’s efficiency and Nvidia’s superior drivers/software, or the Radeon’s slightly better performance and vastly more memory, which comes with vastly higher power consumption. If you decide to go the AMD route, however, Sapphire’s killer Tri-X cooling solution on the Nitro R9 390 keeps the card cool and quiet despite being a power vampire. It’s highly recommended.
$400 to $500: The song remains the same here: The stock GTX 980 and the Asus Strix R9 390X go toe-to-toe in terms of performance, while the GTX 980 only packs half of the 390X’s 8GB of onboard RAM. That said, the 390X sucks down a staggering amount of energy; its 456W under load was far and away the highest energy consumption we’d ever seen in a card (until the Fury X rolled out, anyway). While Asus’ beefy DirectCU III cooling system helps keep the R9 390X utterly quiet, even all those heat pipes and fans can’t keep it anywhere near as cool as the GTX 980.
That said, custom versions of the R9 390X can be found for $400 or even slightly less after rebate, while GTX 980 models start at $470 on Newegg right now. If you can get over the R9 390X’s insane power draw—seriously, 450W!!—it’s clearly the better price-to-performance buy right now, especially with its 8GB of RAM.
$500 to $650: Above $500 you start to enter the rarified air devoted to performance enthusiasts alone, and the realm of AMD’s cutting-edge Fiji GPUs. Here, the air-cooled $550 Radeon R9 Fury is the only major player at MSRP prices, offering performance somewhere between the $400 to $500 options and the $650 flagships for a price that falls squarely between the $400 to $500 options and the $650 flagships. The R9 Nano and Fury X may have seized all the headlines, but this is the real star of HBM-powered Fiji lineup. The Fury simply rocks.
Smart shoppers may want to consider their options, however. While custom-cooled, mightily overclocked versions of the GTX 980 might not quite claw their way up to the Fury’s level of performance, they can get pretty darn close—within 4 or 5 frames per second in most games—for a significantly lower price, as we found when we pit the Asus Strix Fury against EVGA’s GTX 980 FTW. The EVGA card currently costs $485 (after rebate) on Newegg, versus the Strix Fury’s $570. That’s because Nvidia’s pumping out plenty of Maxwell chips, stoking pricing competition for its partners’ boards, while apparent supply constraints for the Fury has left pricing relatively static.
Note, however, that while Nvidia and AMD say the GTX 980 and Fury are respectively capable of 4K gaming, the performance you’ll see is likely to disappoint enthusiasts. Consider these to be impeccable 1440p cards, not entry-level 4K options.
$650 and up: Don’t bother with the $1000 Titan X, period. The $650 GTX 980 Ti and AMD Fury X offer close to the same level of performance for far less money.
Both of these cards can game at 4K at between 40 and 60fps at high graphics settings. The GTX 980 Ti’s easily the superior pick, offering more memory (6GB GDDR5 vs. 4GB HBM) and more performance—that can be overclocked even further—than the Fury X. (Spending a few more bucks on a custom GTX 980 Ti can get you performance superior to even the Titan X.) It’s really a no-brainer, though the Fury X’s integrated water-cooling and drool-worthy, super-premium design may push some folks over to AMD.
It’s worth devoting some time to the Radeon R9 Nano—a true engineering marvel, but one with a very specific niche. This $650 card offers performance that’s better than a custom GTX 980, but worse than the air-cooled Fury non-X. The real draw is its size and reasonable power/thermal needs. At a mere six-inches long, this is the only enthusiast-grade mini-ITX graphics card available; the next step down is a mini-ITX GTX 970, which really can’t hang with this. That said, only a handful of mini-ITX cases can’t fit full-size graphics cards, so the Nano’s potential audience is severely limited right now—though that may change in the future if this pint-sized powerhouse winds up being a trailblazer for the incredible shrinking PC.