Let's get real. For years, NVIDIA has been well and truly trouncing the pants off AMD. Top end, middle end, bottom end. It didn't matter where you looked, AMD was getting flogged. Team Red tried to hang in there with their 200 and 300 series, but to really justify an AMD buy you needed a solid bargain.
Ever since the launch of Polaris, however, the mood has shifted. For the first time in a while, a large segment of gamers have a strong reason to consider AMD again. But while AMD hyped the Radeon RX 480 to the rafters, they've been a little more demure when it comes to the RX 470. And that's a shame, because at $300 it represents perhaps the strongest pitch AMD has had in years.
What Is It?
So for those a little confused by it all, the Radeon RX 470 is a new graphics card from AMD. I've been testing a third-party model from XFX, which you can buy today from PC Case Gear for $299, or $309 for the "Black Edition".
Why Should I Care?
You should care about the RX 470 if you're looking to build a new PC, but you only have around $300 to $350 to spend on a graphics card. The same goes for those who only have around $800 to $1000 to spend on a PC build.
You should care if you're looking to build a small form factor or mini-ITX PC -- something that can sit in the living room -- as a more powerful and more cost-effective version of a Steam box.
If you're looking to build a PC with more power than consoles, but you don't want to break the bank, the RX 470 is an excellent starting point -- especially if you're only ever going to be playing at 1080p, or you don't mind playing games at 1440p with some of the settings turned down.
Here's the specifications for our benchmarking machine:
CPU: Intel Core i7-4790K @ 4.0GHz RAM: 16GB Corsair Dominator DDR3-RAM @ 1600 Mhz GPU: MSI R9 390X Gaming 8GB Monitor: LG 27UD88 4K IPS Monitor PSU: Corsair HX850i 850W HDD: Samsung 850 EVO 1TB SSD Motherboard: ASUS Z97I-PLUS Mini-ITX (Intel Z97) Keyboard: Corsair K70 RGB Lux Mechanical Keyboard Mouse: ZOWIE FK2 Headphones: Audio Technica ATH-M50x
A huge thanks to AMD, NVIDIA, and particularly LG for sending in the parts that allowed me to test all the devices here. Special shoutout to LG for the monitor, which has made my life immensely easier over this past week.
3D Mark Fire Strike
A staple as far as synthetic benchmarks go, 3D Mark’s Fire Strike demo pushes computers and their graphics cards to the limit. They’re often the first benchmarks you’ll see whenever a new GPU is released. It’s not wholly indicative of what a card’s real-world performance will be like, but it does provide a reliable indicator of the performance between cards.
XFX aren't wizards; there was no way there were going to be able to make the RX 470 overtake the GTX 1060 or even the hefty R9 390X in synthetic tests like these.
But the key here is to keep everything in perspective. The XFX RX 470 starts from $299. The cheapest GTX 1060 is going for $369, although most cards start from just under $400.
That's a touch over 23% more expensive for roughly 11% more performance. (The Leadtek GTX 1060 has the same base/boost clock speeds as the Founders Edition GTX 1060 as well.) Keep that in mind, because that's going to be the thread that runs through all these graphs.
Total War: Warhammer
Creative Assembly are the kind of developers who make games that force you to upgrade. The Total War series has been a benchmarking staple for years thanks to the sheer amount of units being rendered at any one time.
The campaign map alone can bring most PCs to their knees, but for this I'm using the in-built scripted benchmark which was made available to the public not too long ago. It's a fairly short benchmark of a battle between the Dwarves and Greenskins, with plenty of action to stress the GPU.
All figures were attained on the Ultra preset and averaged over three runs for accuracy. The DirectX 12 renderer was also used.
Warhammer is an interesting case because it's an AMD-first game in many respects. The addition of DX12 support has swung the pendulum heavily in favour of AMD, provided the argument is one about value. The GTX 1070 and GTX 1080 are still the strongest performing cards, but it's the bottom end that is the most intriguing.
At 1080p, the RX 470 and GTX 1060 are basically neck and neck. Given that the RX 480 and the GTX 1060 are competing at the same price point, that's a staggering result. The RX 470 begins to trail off once the resolution is bumped up to 1440p, although not substantially slow (an average of 48.4fps to the GTX 1060's 50.5fps).
It's also worth pointing out that NVIDIA gets a better return under DX11 in Warhammer. But as time passes, DX12 will become the default -- not just for Warhammer, but more and more games in the future. It'll be interesting to see what, if any, changes NVIDIA can do to reassert themselves for games like these, where the developers are able to leverage techniques (such as asynchronous compute) that currently benefit Team Red a lot more than Team Green.
The Talos Principle
Croteam’s first-person puzzler has an excellent in-game benchmark that’s representative of real-world gameplay. It also scales remarkably well, although that's more down to the fact that the game is getting a little long in the tooth. Nonetheless, it does look bloody nice at 4K.
The settings that was used for all resolutions: Ultra settings for CPU, GPU, GPU Memory, and Level Caching; Fullscreen, 2.4x 3D Rendering MPIX, and 4x MSAA. All figures have been averaged out over three runs.
Let's just have a nice chuckle at the 225fps for a second.
But beyond that, this is the kind of performance you can expect from a lot of older games -- a buttery smooth frame rate, even at 4K. The only card to really set itself from the pack at 4K is GIGABYTE's GTX 1080, with a substantial gap between the GTX 1070/R9 390X from the rest of the pack.
Again, it's worth retaining some perspective. The RX 470 is clearly the slowest performer out of the whole pack, but you have to ask yourself -- how much extra performance are you getting, and for how much?
The cheapest RX 470: $299. The GTX 1060 and RX 480 start from roughly $359-369, with the GTX 1060's more regular price closer to $400.
Ahh, DOOM. I didn't think much of the multiplayer for id's latest demon stomper, but by God was the singleplayer fun. And as an added bonus, the game's optimisation is superb.
Average frame rates are shown below, with the Ultra preset, FXAA (1x) and chromatic abberation enabled. All runs were conducted by running through the game’s third mission, Foundry, on the Ultra Violence difficulty for two minutes. No enemies were killed to ensure consistency with the amount of bodies being rendered and encountered, although slight differences occurred from run to run (such as the enemies opt to beat the snot out of each other).
I also tested the game using the OpenGL and Vulkan renderers, both under the same conditions as above.
There's quite a bit to unpack here, but the part where you can comfortably get 83fps at 1080p from a $300 card with the settings on maximum is a damn good place to start. And if you're using the Vulkan renderer, there's enough headroom to play at 1440p as well.
What's really interesting here is how the GTX 1060 performs amidst all this. It handily beats the RX 470 and RX 480, with even a small margin over the more powerful R9 390X at all resolutions. But under Vulkan, that advantages vanishes completely.
The GTX 1060 is basically only a smidgen ahead of the RX 470, which is a tough spot for a card that costs at least 23% more. Hell, if MWAVE runs out of stock and you have to buy a regular version of the GTX 1060 at closer to $400, the price-to-performance comparison gets even worse.
And as a small note, while the RX 470 did manage to hold above 30fps at 4K here I would strongly advise against it. The card only has 4GB of onboard memory, after all.
Ashes of the Singularity
Ashes is a little similar to Warhammer, in that AMD cards tend to punch well above their weight. NVIDIA cards don't have the same sort of DX12 issues here that they do in Creative Assembly's game, however, with Ashes having been more or less the first staple for DX12 testing.
The extreme preset and 4X MSAA anti-aliasing was used for all resolutions, while the benchmark was set to GPU focused each time. Scores shown below are averaged out across three runs.
The Ashes benchmark is a heavy beast, so it's not surprising that all of the cards failed to get anywhere close to triple figures. It's also worth giving a shoutout to the Ashes developers for just how well their games scales over higher resolutions.
And it's hard not to look at this graph as a win for AMD, especially at 1440p. The two cards are neck and neck, and the RX 480 doesn't really do that much better than the RX 470 in comparison.
We're still dealing with the preview code of Codies' latest F1 racer. But it's already proved to be a well-optimised game with an excellent in-built benchmark, much like previous years.
All tests were run in heavy rain (instead of dry conditions) using the Ultra High preset, with 4x anti-aliasing and SMAA enabled. Codemasters' local PR also confirmed that the game is currently running on DX11, with DX12 support to be added post-release.
F1 2016's benchmark isn't scripted in so much as it is a bunch of AI racers doing a single lap around Melbourne. That means conditions can change slightly from test to test, at least in terms of what happens on the track.
Vettel managed to get himself rammed off the road a couple of times during testing, and that's the only explanation I have for the RX 480 coming in last at 1080p here. It's easier to render the side of a wall for 10 or so seconds than a handful of cars speeding down the main straight, although precisely how much impact that might have had -- especially since tests were averaged out over 3 runs -- is unknown.
Nevertheless, the GTX 1060, RX 470 and RX 480 all returned largely identical results. The smaller amount of RAM in the RX 470 began to show at 1440p, although you could comfortably get performance to hover around the 60fps mark by dropping the graphics preset to something lower.
The GTX 1070 and GTX 1080 tend to enjoy a substantial advantage when it comes to DX11 games, and that's best reflected here. Both cards have a handy lead over the rest of the pack at all resolutions.
But again, the price-to-performance argument rears its ugly head. The cheapest GTX 1080 still costs over $1000 (with the cheapest GIGABYTE models starting from $1049) more than three times the cost of a RX 470 and more than double the price of a GTX 1060 or a RX 470 -- but you don't get anywhere three times the performance.
Benchmarking VR is a tricky process, and there's no simple way to go about it. Measuring the average frame rate doesn't accurately reflect whether you're having a smooth experience, and the adaptive technologies at play makes testing problematic.
One very rudimentary tool is the SteamVR Performance Test, which anyone can download for free. It doesn't really return a great deal of detail, and the graph caps out at 11 (which both the GTX 1070 and GTX 1080 reach). Nonetheless, it does at least give us another reference point.
There's not a great deal to be said here. The RX 470 isn't pitched as a VR capable card, and it doesn't perform like one. You want at least 6GB of VRAM if you're going to play with VR, although the gap between the GTX 1060 and the RX 480 is intriguing.
3D Mark's Time Spy
The Time Spy demo is the DX12-specific test created by Futuremark, the makers of 3D Mark. The new synthetic test was only released to the public last month.
It's a pretty beastly test, with an intensity somewhere between Fire Strike Extreme and Fire Strike Ultra.
The figures largely mirror the results from the earlier Fire Strike scores. The RX 470 and RX 480 are at the bottom of the pack, although the latter is basically on-par with the GTX 1060. The GTX 1060 is around 10% faster than the RX 470 too, but that's not much of a gap given the price differential.
The theme with all of this is perspective. If you go off the raw numbers, NVIDIA pretty much wins out every time -- it just depends on how much. But then you have to consider how much those cards actually cost -- and how much gamers tend to actually spend on their PC.
It's nice to have a PC that costs thousands of dollars. The last one I built cost me almost $5000. But while it was a beast, that's not what most people do. They spend somewhere between $500 and $1500. They don't have the hardware to play games at 4K. They certainly aren't getting invested in VR just yet.
Those are the kind of gamers that AMD is aiming for. It's where the majority of PC gamers are. Just look at the Steam survey results -- the most popular GPUs are the middle end GTX 970, 960, and if you discount the Intel integrated GPU in laptops, the truly aging GTX 750 Ti.
The problem for Australians, however, has always been price. What tastes good to gamers in the United States or Europe turns decidedly sour when the Australia tax comes into play. And the RX 470 isn't immune to that. But it's also the same for every GPU that's been released this year, and as a result a large amount of Australians have a real choice to make.
If you're just getting into PC gaming for the first time, you're not going to splurge thousands. You want to spend hundreds. And the more that price becomes a factor, the better and better the XFX RX 470 looks.
The GTX 1060's pricing looks almost exorbitant in some scenarios, and that can be a major factor when building on a shoestring budget. $70 or $80 can easily be put into a second hard drive, more RAM, a nicer case, or even a better CPU.
Of course, there are plenty of gamers -- myself included -- who want more. And if you're looking to go to crazytown fps at 1440p and beyond, the RX 470 will do nothing for you. But that's not the majority of the market. And Australians within that market -- whether it be someone building a PC for the first time, or someone looking for a sub-$400 GPU upgrade -- now have a real, sincere reason to consider AMD.