The CPU may be the brain of your PC, but when it comes to gaming, the graphics card is the beating heart that pumps pixels out of your obelisk of a tower and into your monitor. A graphics card consists of dedicated video memory and a graphics processing unit (GPU) that handles all sorts of calculations, like mapping textures and rendering millions of polygons. Simply put, the graphics card is the most vital component of your gaming PC. And these are the ones worthy of your next PC, whether it’s a savvy middle-of-the-road build, a budget rig, or a 4K monster.
Earlier this year, Nvidia unveiled their upcoming GTX 1080 and GTX 1070 cards, along with providing release dates: May 27 and June 10, respectively. We’ve now posted our full review of the GTX 1080, and the cards are available at retail…sort of. They keep going out of stock at all the major resellers, and they’re mostly the $700/£619 Founders Edition; other places have inflated prices of $900/£650 or more. Meanwhile, we just reviewed the GTX 1070, and the card launched to a similar state of affairs on June 10, selling out more or less immediately—unless you want to pay the early adopter price premium. Like the GTX 1080, all the initial 1070 cards are Founders Edition models priced at $450/£399, with custom cards on the way.
Considering these Founders Edition cards carry a price premium and don’t come with a factory overclock or any other extras, most users will be more interested in custom versions from Nvidia’s partners. We’ll hopefully start seeing these in the near future, and Newegg at least is showing several non-FE GTX 1080 cards. Or as another example, EVGA lists five 1080 cards, only one of which is the Founders Edition, and the non-FE models range in price from $610-$680 (£428-£477, using exchange rates). That’s promising and our prices should update automatically as the less expensive cards show up. Finally, while Nvidia has officially launched their GP104 Pascal cards, we still have to wait for the end of June to see what AMD’s RX 480 has to offer.
With all the next generation cards coming, the result has been a freefall in GPU prices. GTX 980 Ti can now be found for as little as $490/£470 (after mail-in rebate / MIR), over $100/£70 lower than in April, and other cards are showing similar drops. However, Nvidia appears to be ceasing production on many of their 28nm GPUs, and AMD will likely do the same for the R9 300 series, replacing those with RX 400 series parts that perform better and cost less. Short-term, we expect to see some decent sales to clear out inventory, after which prices may start to creep back up on the older parts (assuming you can find them). Whatever card you’re looking at, it will definitely pay off to check prices before pulling the trigger.
The best graphics card
- Excellent performance for the price
- New features like SMP
- Second fastest GPU
- Availability and Founders Edition pricing
Was there any question about which card would claim the top spot? AMD has a chance to dethrone the GTX 1070 with the RX 480, depending on what actual performance looks like, but until we have hardware in hand, the GTX 1070 is the card to beat. As we show in our full review, the 1070 manages to match or exceed the GTX Titan X and GTX 980 Ti in every game at every setting we tested…and it does so at a much lower price point.
The performance boost is thanks mostly to the move from 28nm planar transistors to TSMC’s current 16nm FinFET transistors, which shrinks the GPU size, reduces power leakage, and allows for higher clocks. Where the base clock of 980 Ti and Titan X is 1000MHz (stock), the 1070 runs at 1506MHz. Even with fewer CUDA cores—1920 on the 1070 vs. 2816 on the 980 Ti and 3072 on the Titan X—the added clock speed and architectural enhancements keep the 1070 in the lead.
Which isn’t to say that you’d necessarily want to upgrade from a 980 Ti or Titan X to the 1070; the performance is close enough that it’s really about the change in price for the same level of performance; the same goes for cards like the R9 Fury X, Fury, and Nano. Regardless of whether you’re looking at a custom card or a Founders Edition model, that’s a lot of money to spend on a graphics card. Our advice for gamers is to try and skip a generation or two of hardware, or more generally, only upgrade when your current hardware becomes ‘too slow.’ That means different things to different people, of course, so high-end gamers are likely to jump at the slightest enticement while frugally minded might get by with high or even medium/low quality.
But even though we just said skipping a generation makes sense, owners of 970 and lower hardware—or R9 390 and lower on the AMD side—will certainly be tempted to upgrade, and the older your current hardware, the bigger the performance gains. Even the single generation move from a 970 to a 1070 will get you a solid 70 percent increase in frame rates, or 45 percent more performance than an R9 390. And if you happen to be running something from the GTX 700 era, well, that’s where upgrading delivers the biggest returns: the 1070 in general is more than twice as fast as a GTX 770, and in some cases (Shadow of Mordor, for instance) it’s nearly three times as fast!
Overclocking of the GTX 1070 also pays dividends, though not quite as large as those on GTX 980 before it. We were able to increase the core to +200MHz over stock, and with higher power and thermal limits (and a higher fan speed), performance improved by 15 percent. Memory overclocking also went well with our sample, hitting 9.4 GT/s—a frankly staggering number for GDDR5, considering Titan X and 980 Ti previously topped out around 8.0-8.4 GT/s. But if you’re looking at ‘typical’ overclocks, a GTX 980 Ti with a good overclock is basically going to match a 1070 OC.
Other features are likely to come into play over the coming year, however. One of the coolest is the simultaneous multi-projection (SMP) feature that allows the GPU to calculate up to 16 projections in a single pass. Why would you need multiple projections in a game? For multi-screen surround setups, it can get rid of the distortion you see with ‘curved’ side displays. More pertinent for most people is the potential improvements in VR performance, since each eye requires its own projection. Nvidia claims that with SMP in a VR title, Pascal (GTX 1080) is up to twice as fast and three times as efficient as the Titan X; extending that to the GTX 1070, it should be around 60-70 percent faster than the Titan X in such cases.
So what’s not to like? Besides the price, where the least expensive cards are still $50/£35 more than the GTX 970 at launch, the biggest problem is going to be availability. Everywhere we’ve looked, on launch day the GTX 1070 is either sold out or way overpriced, and judging by the GTX 1080 it might be a few weeks or even a month or more before things settle down. And once cards are readily available at Nvidia’s stated MSRP ($380/£339), people will be looking at the upcoming fall and winter and wondering what will come next. Just don’t get too stressed out about finding the ‘perfect’ time to upgrade to a newer, faster card, because something better is always right around the corner.
The best high-end graphics card
- Fastest current GPU
- Excellent efficiency
- Great new features
If you want the fastest graphics card on the planet, it’s a no-brainer: the GeForce GTX 1080 wins, hands down. In our testing, it’s over 30 percent faster than the GTX Titan X and GTX 980 Ti, and it uses almost 30 percent less power. Again, that’s all thanks to the move from 28nm transistors to 16nm FinFET, allowing Nvidia to cram more parts into a smaller area, with less power leakage thanks to FinFET.
Now toss in the architectural improvements of Pascal, and long-term we could see even greater improvements in performance compared to Maxwell. Pascal has better delta color compression, resulting in higher effective memory bandwidth by about 20 percent (according to Nvidia). GDDR5X clocked at 10 GT/s means the 320GB/s of actual memory bandwidth in practice works as well as the Titan X’s 384GB/s. Graphics preemption can be used to improve load balancing and time-sensitive work (e.g., asynchronous time warp in VR). And Simultaneous multi-projection (doing up to 16 projections in a single pass) means in VR workloads the GTX 1080 could actually end up being twice as fast as the Titan X.
What’s not to like? The biggest drawback is easy: the price. No, we don’t feel $600/£529 is necessarily too much to ask for this level of performance—after all, the GTX 980 Ti was doing well at $650/£579 since it launched last year—but it is a lot of money. Then there’s the whole Founders Edition business, a reference card carrying reference clocks and using Nvidia’s blower cooling setup, all for $100/£70 extra. Even at the Founders Edition price, however, the GTX 1080 is an amazingly powerful card, with the result being that inventory keeps selling out, and the only cards that are frequently in stock are currently selling at well above MSRP. Unless you have money to burn, we recommend waiting for the custom cooled cards from Nvidia’s partners to begin showing up at the target $599/£529 price range.
Overclocking of the Founders Edition is possible, though our experience to date is that gains are slightly less than what we’ve seen on previous Nvidia cards. The GTX 980 Ti, 980, and 970 often hit 25 percent core overclocks compared to stock; we managed to add just 200MHz (about 12 percent) to the base clock, with a 750MHz (+17.5 percent) increase in VRAM speed. Other tweaks to the power target and voltage led to roughly 15 percent better performance when overclocked, but similar efforts on a 980 Ti could close the performance gap to around 20 percent.
Finally, this isn’t the fastest Pascal chip we’re likely to see from Nvidia. GP104 is for gaming and consumer graphics, but the GP100 boasts far higher specs in many areas. It has 3584 CUDA cores compared to the 2560 in the GTX 1080—and there are still 256 CUDA cores disabled in the Tesla P100—with 16GB of HBM2 memory providing 720GB/s. That’s at least 20 percent higher core performance with more than twice the memory bandwidth; it doesn’t take a rocket scientist to figure out that Nvidia will likely release an even faster card in the future, though it might be another 12 months. How much would a consumer version of GP100 even cost, though? It might end up in the same ballpark as the Broadwell-E i7-6950X ($1723/~£1400); that or Nvidia will do a GP102 or GP110 without HBM2 that they can price closer to previous Titan cards.
But for now, if you’re after the best—aka fastest—graphics card, GTX 1080 holds the crown. And if you’re really looking to set a performance record, you could put two of the cards in SLI. Just know that such a setup will need a lot of CPU performance to keep it happy, as even a 4.2GHz i7-5820K runs into CPU limits at 1080p Ultra with a single GTX 1080. At least Nvidia is no longer recommending 3-way or 4-way SLI as useful solutions, though it’s still possible to go that route if you want.
Games are rarely bottlenecked by your CPU, but dozens of games every year will push your graphics card to its limits. It’s the component you’ll want to upgrade most frequently (though if you buy the right card, it should last you at least two years). For gaming systems, it’s also likely the most expensive part in your build. On a practical budget, it’s critical to find the graphics card with the best ratio of price to performance. That’s why we’ve previously looked at cards in the $300/£250 range, though we’ve stretched that with the GTX 1070 this round.
At $450/£399), Nvidia’s GTX 1070 is a killer card, outperforming older cards that initially cost twice as much, and prices will only go down from here. It’s overclockable, quiet, and efficient; more importantly, it’s able to run every game we’ve tested at more than 60 frames per second at 1080p Ultra, and most of the games are still breaking 60 fps at 1440p Ultra. You can argue about price and whether or not you really need Ultra quality settings, but right now, the GTX 1070 is the best card for the price.
While the GTX 1070 is the card we’d recommend to most—but not all—PC gamers, there are many viable alternatives. Maybe you don’t care for Nvidia or their overwhelming market share, or maybe you’ve got cash to burn, and need a card that can run games at 4K resolution. Perhaps you’re trying to build a dirt-cheap gaming PC and you can’t with an even cheaper graphics card. Our graphics card guide includes options covering the entire market, from budget to mainstream to high-end gaming PCs.
Testing graphics cards
Our graphics card recommendations are based on our own benchmarks and testing, as well as research into the reviews and testing done by other sites. Along with Maximum PC, we have benchmark data for a range of Nvidia and AMD graphics cards, including the GTX 900 series, Nvidia 700 series, AMD R9 300 series, and the R9 Fury/Nano cards.
What makes the best graphics card? For PC gamers, it’s a balance of price and performance. The graphics card must be able to run demanding games at high framerates and settings, with 1920×1080 being the most common resolution. However, we also test at 2560×1440 and 4K, which are becoming increasingly popular choices, particularly at the high-end. The best graphics card shouldn’t cost more than other cards with comparable performance, and the card should be fast enough to still perform respectably two years later, even if it can’t run everything at max settings.
Graphics performance isn’t the only consideration. The quality of game drivers and other features supported by the card are important. The card’s noise level, power draw and temperature matter, too. Thankfully, nearly all of the cards run fairly quiet, even under load, and temperatures are within the acceptable range as well—though Nvidia still has an advantage when it comes to power for now.
From a high level, we tested each card on a high-end PC (4.2GHz overclocked i7-5930K), at 1080p, 1440p, and 4K with Ultra settings at the lower resolutions and Ultra or High at 4K. We included results from fifteen games, some newer and some slightly older: Ashes of the Singularity (DX12), Batman: Arkham Origins, The Division, Doom (2016), Fallout 4, Far Cry Primal, GTAV, Hitman (DX12), Hitman: Absolution, Metro: Last Light, Middle-Earth: Shadow of Mordor, Rise of the Tomb Raider, The Talos Principle, Tomb Raider 2013, and The Witcher 3. Here’s how they stack up in terms of average and minimum frame rates across these games:
We’ve currently selected the Nvidia GTX 1070 and GTX 1080 as the two best graphics card for most gamers after benchmarking all of the latest GPUs and comparing the numbers—including power requirements, noise, and performance. For those who can’t afford to spend that much money, the middle and bottom of the chart still hold a lot of compelling options, and it’s only going to get better as new parts arrive. Here’s a different look at all of the cards we tested, this time rating them in terms of value—FPS for money spent:
Those charts show the rather interesting breakdown of the market right now. AMD is competing on value if not performance, and their R9 380/390 cards pack a serious punch. Only the GTX 950 comes close to breaking into the top three slots (and it does so on our UK pricing scale), but once we leave the R9 300 series behind, Nvidia completely dominates the middle of the chart. Note that the prices used for these charts reflect current street prices, and we used Founders Edition pricing on the GTX 1080 and 1070—they’ll look a lot better in a month or so most likely. AMD’s Fiji meanwhile ends up looking like mostly a proof of concept, or a test drive for HBM; they’re still fast cards, but at current prices they’re not really worth picking up, and the cost of HBM means they’re unlikely to fall much lower in pricing.
Looking forward, the computer graphics world is a fast-changing field, and with the 16nm and 14nm FinFET process now coming online for GPUs, things are going to be very interesting. Nvidia has aimed squarely at high-end gamers and enthusiasts with the GTX 1070 and 1080, while AMD is gunning for the mainstream market with their RX 480 targeting a starting price of just $199. Will that give you R9 390 performance for less money while using two thirds the power? That seems to be the plan, but we’ll know more come July. Hopefully AMD can do a bit better at keeping retailers in stock at launch, but that’s always difficult for highly anticipated hardware.
While we love getting new hardware in for testing and review, the days of revolutionary upgrades mostly seem to be in the past. Even with the move from 28nm to 16nm FinFET, we’re still not doubling performance, at least not at the same price point. If you find your current system isn’t keeping up with the gaming times, however, 30-60 percent better performance will certainly help. Those who already own an R9 300 or GTX 900 series card should be safe for the time being, while R9 200 and GTX 700 series owners may be feeling the upgrade itch. Anything older than that and the latest generation hardware can easily double performance—or more. After five years stuck at 28nm manufacturing, the GPU companies have finally received their long-awaited process shrink. Thank goodness.
A note on affiliates: some of our stories, like this one, include affiliate links to stores like Amazon. These online stores share a small amount of revenue with us if you buy something through one of these links, which helps support our work evaluating PC components.
Source link
