The Best Video Cards of 2017: All Price-Points Compared & Benchmarked

By Published November 26, 2017 at 5:27 pm
  •  

After a year of non-stop GPU and CPU launches, a GPU round-up is much needed to recap all the data for each price-point. We’ll be looking at strict head-to-head comparisons for each price category, including cards priced at $100-$140, $180-$250, $400-$500, and then the Ti in its own category, of course. As noted in the video, a graphics card round-up is particularly difficult this year: Chaos in the market has thrown-off easy price comparisons, making it difficult to determine the best choice between cards. Historically, we’ve been able to rely on MSRP to get a price (+/-$20, generally) for comparison between both AMD and nVidia; the partners hadn’t strayed too far from that recommendation, nor the retailers, until the joint mining & gaming booms of this year. Fortunately, much of that pandemonium has slowed down, and cards are slowly returning to prices where they sat about 6-8 months ago.

Another point of difficulty, as always, is that price-matched video cards will often outperform one another in different types of workloads. A good example would be Vega vs. Pascal architecture: Generally speaking – and part of this is drivers – Pascal ends up favored in DirectX 11 games, while Vega ends up favored in asynchronous compute workload games (DOOM with Vulkan, Sniper with Dx12). That’s not necessarily always going to be true, but for the heavyweight Vulkan/Dx12 titles, it seems to be. You’ll have to exercise some thought and consider the advantages of each architecture, then look at the types of games you expect to be playing. Another fortunate note is that, even if you choose “wrong” (you anticipated Vulkan adoption, but got Dx11), a lot of the cards are still within a couple percentage points of their direct-price competition. It’s hard to go too wrong, short of buying bad partner cooler designs, but that’s another story.

We won’t cover every single GPU released this year (GT 1030 and RX 550 not discussed), but will hit the main ones for the market. Finally, all discussion points will look at relative performance in head-to-head matchups, with the 100% identifier demarcating maximum performance. There is one exception to this rule in a Vega 56 Hybrid mod we did.

Note: Our GTX 1070 Ti review will have the most updated numbers for GPU benchmarks.

Best Video Card Matchups

AMD

NVIDIA

(Rough) Price Range

RX 560 4GB

GTX 1050 2GB

Desired: ~$100-$130
Actual: ~$110-$160

RX 570, RX 580

GTX 1060 3GB, 6GB

Desired: ~$180-$250
Actual: ~$190-$300

RX Vega 56

GTX 1070

Desired: ~$400-$450
Actual: ~$400-$550

RX Vega 56, RX Vega 64

GTX 1070 Ti, GTX 1080

Desired: ~$400-$500
Actual: ~$400-$800 (!)

No Show

GTX 1080 Ti

Desired: ~$700-$750
Actual: ~$720-$850+

AMD RX Vega 56 vs. GTX 1070

Buy the AMD RX Vega 56 here.

Buy an EVGA GTX 1070 SC here.

In our Vega 56 review, the TLDR version was that the card, if it could be found at its MSRP of $400 or reasonably close to GTX 1070 prices, made a lot of sense. We later revisited the Vega 56 card in our Hybrid mod, where we applied powerplay tables, 400W worth of power, a 360mm radiator, and pushed it to its absolute limits. The result was higher performance, as expected, but also completely blowing out the power efficiency. It was a bit impractical, but the point was that Vega 56 has a lot of room to play, and can effectively invalidate the Vega 64 card with even a light overclock.

vega56 vs 1070 performance

In terms of gaming performance, the GTX 1070 and Vega 56 cards are reasonably close in their matchups. The above chart of relative AVG FPS performance shows that, with the GTX 1070 SC leading primarily in Ghost Recon and For Honor, while Vega 56 leads in the rest. That’s what precipitated the launch of the GTX 1070 Ti. Vega 56 has a wider spread of games here where it wins and has a price which is theoretically lower.

gtx1070 vs vega power 3dmark

Here’s a chart of power performance. The GTX 1070 is much more efficient, but not in a way which makes Vega 56 significantly worse for room ambient or for the power bill, in most places. If these numbers do concern you, well, your choice has been made. If not, we’re still supporting Vega 56 as a buy, but only if it’s under GTX 1070 Ti prices. We also should note that the reference blower performs poorly, like always, despite having an excellently built PCB and VRM. You’ll want to either mod the card, which adds to cost, or wait for partner models.

rx vega 56 noise

The reference model is loud, as you can see in our RPM-to-noise response chart, and pushes high temperatures across the entire board. Vega is sensitive to high temperatures, just like Pascal, and so you’ll have direct clock benefit from a better cooling solution. To learn more about Vega, undervolting, overclocking, and more, check this article and then this one. Here's a link to an air cooler mod, if you're impatient for partner models.

Vega 56 vs. GTX 1070 Ti & GTX 1080

Buy the GTX 1070 Ti Gaming X here.

Buy the GTX 1080 here.

Here’s a look at how the GTX 1070 Ti changed the game. We’re starting with non-overclocked numbers, but we’ll get to overclocking – it matters:

vega 56 vs 1070ti vs 1080 performance

Without an overclock, running the reference clocks – because that’s what nVidia enforced on manufacturers – we’re already running the 1070 Ti at anywhere from 89% to 98.4% of the GTX 1080. RX Vega 56 is listed for perspective, operating at 74% to 88% of the GTX 1080; remember, this is not percent performance of the 1070 Ti, but of the 1080. We are scaling versus a baseline of a GTX 1080, just to be clear.

The GTX 1070 Ti is a GTX 1080 with one less SM, just like the GTX 1060 3GB is a GTX 1060 6GB with one less SM. The difference amounts to one of clocks more than shaders, as we’ll show in this overclocking chart.

A lot of folks jumped on the bandwagon that the 1070 Ti is a pointless card, but ever since launch, we’ve disagreed with some of that rhetoric: The GTX 1070 Ti, even before an overclock, largely invalidates the GTX 1080s – not the other way around. At $50 cheaper on average, the 1070 Ti provides 90-98% of the performance when stock, and mostly achieves parity when overclocked. Unless you can find a GTX 1080 on sale for under $500, or unless GTX 1070 Tis end up around $480-$490 exclusively, we think the 1070 Ti makes more sense than the 1080. You can save a few bucks, and as we showed in our recent man vs. machine overclocking, you can get back some performance with 15 minutes of overclocking work.

vega 56 1070ti 1080 oc performance

gtx1070 vs vega power 3dmark

Power consumption isn’t even that different, as you’d expect, and that’s because it’s the same silicon. So, to recap, everything depends on price: At price parity or favor toward Vega 56, we’re recommending Vega 56 over the GTX 1070. At a price advantage of $40-$50 over the GTX 1080, we’re recommending a 1070 Ti and maybe a short overclock. Under $500, we’re recommending a 1080.

GTX 1060 3GB vs. GTX 1060 6GB

Buy the GTX 1060 3GB here.

Buy the GTX 1060 6GB here.

Before getting into mid-range hardware, let’s first address the differences between the 1060 3GB and 6GB cards: As a recap, we concluded our GTX 1060 3GB card review by declaring that it should have been named “GTX 1050 Ti,” as the card doesn’t just halve the VRAM, it also removes one SM from the GPU, reducing core count by 10%. Base clock is the same between them, but the core count and memory capacity both lower. As for what that means for FPS, well, it really depends on the games you’re playing.

gtx 1060 6gb vs 3gb performance

Here’s the relative performance for AVG FPS across several games. The GTX 1060 6GB card is represented at 100% and runs at the same clock as the GTX 1060 3GB card. For the most part. For the most part, the GTX 1060 3GB card always maintains at least 88% of the performance of the 6GB card, with its average performance closer to 92-93% of the GTX 1060 6GB card. Scalability gets a bit rough when blasting VRAM, but we’re more likely to become bound by ROPs or cores and clocks before becoming bound by VRAM. There are a few games where frametime consistency dips at higher texture qualities, represented by our 1% and 0.1% low converted FPS values.

gtx 1060 6 vs 3gb 1pct low performance

The 1% lows are mostly consistent with the average, as shown in this chart, and we tend to sit between 88% and 95% of total performance potential.

gtx 1060 6 vs 3gb 01pct low performance

The 0.1% lows, shown in this chart, are also largely consistent – except more VRAM-abusive games, like Mordor, drag us down to 70% of total performance potential. Fortunately, although occasionally noticeable, this isn’t experience-ruining.

Generally speaking, if the two cards cost the same, the 6GB version is going to be a better buy; they generally don’t cost the same, though, and if the price gap nears $40-$50, it may be better to buy the 3GB version, unless you are running specific applications that are VRAM-intensive. For anyone attempting to make the argument that higher capacity will last longer in the long-run, keep in mind again that we’re more likely to become ROPS- or core-bound in the pipeline, prior to becoming memory-bound.

Now that we’ve gotten that out of the way, we can turn toward RX 580 and GTX 1060 comparisons.

RX 580 vs. GTX 1060 vs. RX 570

Buy the RX 580 here.

Buy the GTX 1060 6GB here.

Buy the RX 570 here.

As a refresher, the RX 580 uses the same Polaris GPU as the RX 480, and was functionally a refresh. The card significantly increased power consumption over the RX 480 and the GTX 1060, with improvement only to idle power consumption. This was a launch where AMD and its board partners opted to blast the Polaris clock, effectively pre-overclocking the cards, and increase the voltage going to the core. The result was a highly competitive product with the GTX 1060, when looking at gaming performance, but higher power draw as a result. Whether that’s relevant, we’ll leave up to you.

rx 580 vs gtx 1060 relative performance

(See previous power chart for RX 580 vs 1060 power draw)

As for the average gaming performance, the GTX 1060 and RX 580 trade back-and-forth. The award for “best” gets passed between them largely depending on game, with nVidia generally pulling ahead in DirectX 11 titles and AMD pulling ahead in Vulkan or DirectX 12, somewhat expectedly. The RX 580 has room to play with overclocking and undervolting, making it a good option for people who don’t mind being under the hood, so to speak. The trouble is availability and price, as ever. If you can find these cards at price parity, we just recommend picking based on per-game performance. Figure out which has the performance most representative of titles you intend to play, consider power consumption for the nVidia argument, consider FreeSync for the AMD argument, and pick primarily based on price and availability.

This discussion does overlook the RX 570, though, which we found to be the most promising of the launches. The RX 570 cards have been difficult to get near MSRP since launch, and that problem was exacerbated with the mining trend. If you can find one below RX 580 prices, ideally around $180, then we still like the RX 570. Much like Vega 56 and Vega 64, the difference is one of shaders, and most games favor the clock difference over the shader difference; you can easily compensate in most titles by overclocking an RX 570, and it is a much more worthwhile purchase than the GTX 1050 Ti at its $150 to $170 price point. That said, the GTX 1050 Ti is one of the few cards that, for a few months this year, was left untouched by miners, making it an odd, unexpected anchor for gaming PCs. We never fully recommended the 1050 Ti, as the RX 570 was much better at the high-end, and the 1050 or RX 560 fulfilled the low-end. It wasn’t until the GPU shortage that the 1050 Ti developed more value. That’s mostly solved for by now, though.

RX 560 vs. GTX 1050

Buy the RX 560 4GB here.

Buy the GTX 1050 2GB here.

rx 560 vs gtx 1050 relative performance

Moving on to low-end GPUs, we now look at the RX 560 and GTX 1050 cards. At present, the RX 560 ranges from $110 to $140. To save everyone some time, the $140 units are a rip-off – don’t bother. The only challenger to this level of worthlessness is the GTX 1050’s $150 units. Not the Ti, but the straight 1050.

Ignoring the overpriced units, the reasonably priced GTX 1050s cost between $112 and $120, with the reasonably priced RX 560s around $110 after rebate, if you count those toward the price, or $124 flat. These cards should both be close to $100, but the GPU market behavior has also affected the low-end, unfortunately.

We’ll make do with what we have, though. Looking at average performance, the GTX 1050 maintains a lead in the same games as the previous pairings: Ghost Recon and For Honor, representing our Dx11 tests, with FireStrike representing one half of the synthetics. The lead tends to be in the range of 8 percentage points. The RX 560, meanwhile, maintains a lead in TimeSpy, a significant lead in DOOM, and another significant lead in Sniper, with a small lead in Ashes.

When the RX 460 launched, it was a bad purchase compared to the GTX 1050. The RX 560 changed that, and showed performance advantages in Dx12 and Vulkan titles. The RX 560 doesn’t have huge deficits versus the 1050 in Dx11 – they’re workable, and even recoverable with an overclock – but maintains large leads in Dx12 and Vulkan APIs, presently. Not shown in these charts, but shown in our review charts, the RX 560 also maintains more consistent frametime performance in some games, partly thanks to its 4GB framebuffer. As with the other cards, it comes down to price. At price parity, our pick is the RX 560. We’d like to see the RX 560 closer to $110, and we’d like to see the 1050 closer to $100, but there’s not much we can do about that.

High-End – GTX 1080 Ti vs. Itself

The GTX 1080 Ti remains uncontested at the absolute high-end, and so we’ll point you toward our already-compiled list of the best GTX 1080 Ti designs that we’ve tested this year.

As for cards we’re recommending against, there are primarily two devices that we strongly encourage against right now: The Vega 64 card, as Vega 56 + an OC instantly invalidates it, and the Titan Xp, which can be beaten by out-of-box 1080 Tis in gaming workloads (though does have an extra 1GB of VRAM, apparently useful for some machine learning applications – but not for us).

The rest are mostly picked. If you’ve got questions about any of them, we’d point you toward our GTX 1070 Ti review, as it will contain the most recent data; otherwise, of course, individual reviews for each card are still online.

Editorial: Steve Burke
Video: Andrew Coleman

Last modified on November 27, 2017 at 5:27 pm
Steve Burke

Steve started GamersNexus back when it was just a cool name, and now it's grown into an expansive website with an overwhelming amount of features. He recalls his first difficult decision with GN's direction: "I didn't know whether or not I wanted 'Gamers' to have a possessive apostrophe -- I mean, grammatically it should, but I didn't like it in the name. It was ugly. I also had people who were typing apostrophes into the address bar - sigh. It made sense to just leave it as 'Gamers.'"

First world problems, Steve. First world problems.

We moderate comments on a ~24~48 hour cycle. There will be some delay after submitting a comment.

Advertisement:

  VigLink badge