This week, AMD launched their Radeon RX 500 series of graphics cards. The lack of any fanfare surrounding this launch is down to the fact that these cards are rebrands – they are not based on the upcoming Vega architecture. The RX 500 series is a respin on the Polaris 10 and Polaris 11 architectures, which launch on a revised 14nm FinFET process. This brings with it lower heat levels, more power efficiency, and a little more performance than the RX 400 series brought to the table. In today’s lineup, there is the RX 580 and the RX 570, but AMD is also planning to launch the RX 560 and the brand new RX 550 in the near future.
AMD Radeon RX 580 and RX 570 comparison
|RX 580 8GB||RX 480 8GB||RX 570 4GB||RX 470 4GB|
|Base clock speed||1257MHz||1120MHz||1168MHz||926MHz|
|Boost clock speed||1340MHz||1266MHz||1244MHz||1206MHz|
|Texture units (TMU)||144||144||128||128|
|Memory bus size||256-bit||256-bit||256-bit||256-bit|
|Memory clock speed||8.0GHz||8.0GHz||7.0GHz||6.6GHz|
From the starting line, we have the successors to the Radeon RX 480 and the RX 470. These are respins of the Polaris architecture under the new codename “Polaris 20”. It is the latest version of Graphics Core Next 4.0 on an improved 14nm FinFET process, refined to increase clock speeds and keep power consumption in check. You’ll see reviews online that have the RX 580 peaking at around 200W of power draw at maximum clocks, but that’s actually much better than what most RX 480 cards could achieve at the same power level, because they were often thermally limited.
Otherwise everything is pretty much the same with these cards when comparing generations. The chips are almost identical, the display outputs are identical, and even the memory chips used are pretty much identical – AMD did not elect to move the line to GDDR5X VRAM, nor did they decide to add in extra special sauce. These cards are intended as an upgrade to older generations of GPUs, which we’ll talk about in a moment. I believe that AMD’s strategy for this rebrand won’t work in the way they were hoping, but there are merits to choosing cards from the RX 500 family instead of RX 400.
What is interesting is how the updated FinFET process increases clock speeds and raw performance much higher than before. AMD previously had to massage expectations with the RX 400 family by telling the media that the boost clocks were typical clock speeds that you’d see with lighter loads, but with RX 500 we’ll probably discover that this is not the case anymore. Previous attempts at overclocking the RX 480 to around 1450MHz yielded in a power draw around the 250W range, while the RX 580 at those same speeds might be drawing up to 60W less. TechpowerUp’s review of the Sapphire Nitro+ RX 580 revealed that performance increased by 9%, but this necessitated a 30W increase in the average power draw.
Some reviews are going to point this out and call it a failure, but a near 10% increase in performance for a pittance in terms of power draw is basically a generational leap on its own. AMD is bumping up clocks, bumping up performance, giving you better numbers on average in terms of frame times, and they’re charging the same price as the old model. Unlike the Hawaii-Grenada switch, which was a near-identical rebrand, this is different silicon.
There are also reports of some RX 480 and RX 470 owners being able to flash their chips into RX 580 and RX 570 cards, with higher and more stable overclocks than before. The reason for this is probably that AMD slipped Polaris 20 into production alongside Polaris 10 in order to fulfill last minute orders going out to their partners for Polaris 10 product, and thus any recently made RX 480 is basically an RX 580 in disguise. It is also possible, however, that the new BIOSes are more mature and robust, and it may benefit owners of reference RX 480s, for example, to update their BIOS to these newer versions.
AMD Radeon RX 560 and RX 550 comparison
|RX 560 4GB||RX 460 4GB||RX 550 2GB||R7 250E 2GB|
|Base clock speed||1175MHz||1090MHz||1183MHz||800MHz|
|Boost clock speed||1275MHz||1200MHz||N/A||N/A|
|Memory bus size||128-bit||128-bit||128-bit||128-bit|
|Memory clock speed||7.0GHz||7.0GHz||7.0GHz||6.6GHz|
Moving down to the entry-level part of the update we have the new Polaris 12 family. Interestingly, the RX 560 replaces the RX 460, but it is not an identical chip swap. The RX 460 available in most regions in the world was based on Baffin Pro with 896 shader cores, but there was a better version based on Baffin XT, which was a full implementation of Polaris 11 hardware. That chip was typically only found in the Chinese market, confusingly labelled similarly to the regular RX 460, and it was also found inside Apple’s Macbook Pro 15 laptop labelled as the Radeon R9 M370X. So, not only is it clocked higher than the old RX 460, it also boasts more execution units. I’d estimate the performance gain to be anywhere between 15-20%, which is a sizeable jump for this kind of product.
At the bottom of the rung is the new RX 550. This is not a respin of any previous chip, but rather a brand new one that’s never been seen before. The RX 550 is based on the “Lexa Pro” architecture, which means that there’s a “Lexa XT” chip out there, presumably sporting 768 shader cores and a bit more memory bandwidth. Compared to the R7 250E, this doesn’t look like much of an upgrade, until you factor in that the R7 250E is based on the Cape Verde architecture, which launched in December 2013. The RX 550 is way, way faster than the chip it replaces in every direction, and it even gives budget gamers new features like FreeSync compatibility and the ability to run games with the Vulkan API. This jump is bigger than expected for shoppers at this price point, and it’ll put even more pressure on NVIDIA because they do not have an equivalent card to compete with.
With the specs out of the way, let’s chew on AMD’s marketing slides for a bit.
As a reminder, every Radeon RX 400 and RX 500 series GPU on the market today supports AMD’s new recording feature called Radeon ReLive, available in their drivers. ReLive is AMD’s capture tool based on the Video Coding Engine (VCE) hardware present in the GCN architecture. Polaris 10, Polaris 20, Baffin XT and Pro, and Lexa graphics cards all ship with VCE 3.0 hardware, allowing recording of gameplay up to UltraHD 4K at 30 frames per second. ReLive can also handle your streaming requirements as well, broadcasting the game footage along with input from a digital camera to your platform of choice. It’s well worth your time to just try it out for a bit, even if you decide to switch back to OBS anyway.
All of the graphics cards discussed today support Radeon Chill, a feature for Radeon RX 400 and RX 500 series graphics cards that allows you to run games at reasonably high framerates, but with much lower power draw than normal. Radeon Chill is a variable frame rate cap together with on-the-fly tweaks to the buffer flip behaviour of modern games. The framerate automatically increases or decreases based on the amount of input from your peripherals. This ensures that there’s slightly less lag between an action that you make using your peripherals and the time it’s taken to display that on your monitor.
If you’re on a non-FreeSync monitor, this sort of gives you the same kind of immediate response that comes with using a variable refresh rate monitor. Whenever a new frame is displayed on your monitor, that frame data also includes the most recent input data from your peripherals. In practice the benefit you get in gaming isn’t always immediately apparent. Real-time strategy, fighting, racing and shooter fans will get more use out of this than most other gamers.
The gameplay experience you get out of this feature is also different to implementing a frame rate cap or using frame rate target control (FRTC). With a frame cap, any frames rendered outside of the limit will get dropped from the queue, which means that there is potentially missing animation in the image projected on the display. FRTC puts backward pressure on the queue by only rendering a new frame at the set interval, which means that there’s some lag introduced into the experience just like you would see with VSync enabled.
Radeon Chill is theoretically the best of both worlds, but how the game engine works will affect its suitability, which is why AMD has worked with game developers to enable Radeon Chill for tested titles.
Because the RX 500 series is a rebrand, AMD doesn’t see people from the RX 400 series upgrading to it. Instead, they expect to attract users of the now-EOL Radeon R9 380X and the NVIDIA GeForce GTX 970. The RX 580 promises to be better than these cards for many games, but unless you’re switching over specifically for the FreeSync and better Vulkan/DirectX 12 support in the case of the GTX 970, you might be justified in saving your money and playing at slightly lower settings while you wait for Vega or NVIDIA’s GTX 2000 series.
When it comes to the RX 570, the argument is much the same, but it’s a much stronger one against the cards it replaces. It is sometimes twice as fast as the outgoing Radeon R7 370 or the Geforce GTX 960, and it’ll pack up to 8GB of VRAM to boot. The RX 570 should also slot into the same systems already running these cards without fuss, because its power requirements are very similar, but the performance on tap is far higher. Against the R7 370, the upgrade is rather compelling; you get access to the latest 3D APIs that support asynchronous compute, and you are able to fully enjoy the use of a FreeSync monitor.
Things are quite predictable now. AMD is targeting graphics cards of old that held the same price points as the cards they’re introducing today. Against the R7 360, it’s a no-brainer, but keep in mind that the RX 460 will be discounted in the very near future, and sales of those cards could work on better on your wallet. It can still be overclocked to improve performance by around 10%, and it won’t be too far behind the new RX 560.
Finally, there’s the gem of the lineup, the brand new Radeon RX 550. As I’ve said before, AMD hasn’t had an update to their super-budget lineup for three years now, and the R7 250 is a bit long in the tooth. Well, the RX 550 is sometimes twice as fast as the old contender, and it is well suited to deliver better performance than Intel’s integrated graphics. For budget builds this card will be a great choice, and if you just need a hardware upgrade to make your life easier in running some effects in Adobe Photoshop or Premiere Pro, or even doing batch video conversions into the H.265 video codec, this will be a sweet deal for you.
It’s not just gamers who benefit either. Those of you who have a home theatre PC in your living room will benefit handsomely from the new technologies on offer, especially if you have a 4K display with HDR capability. The RX 550 supports HDMI 2.0 and Displayport 1.4 outputs, and if my hunch is right about the power usage, it’s around 30W at maximum load as well. That means it’ll enjoy a lot of slim, low-profile designs for HTPC and workstation use, and there’ll definitely be a passive design or two from AMD’s partners as well.
Lastly, as a reminder, AMD wants their customers to know that all of these products, down to the RX 550, will support the full range of FreeSync variable refresh rate monitors. While no-one is going to buy 144Hz displays for such cheap cards, the options aren’t too bad either. AOC’s G2260VWQ is a 1080p TN display with a variable refresh window of 48-75Hz. Viewsonic’s VX2457 is a 24-inch version of the same thing. The Acer KG2212Q and Samsung CF22390 are both brand new monitors that have not been released yet, and both use HDMI 2.o connectors with FreeSync capability and a refresh window of 48-75Hz. The CF22390 is a curved display, but I’m not sure how much this will benefit anyone at this small size, even if it is an IPS-like VA panel.
That’s everything from AMD’s stable! The RX 580 and RX 570 go on sale this week at participating retailers, and I’ll have a price guide up for those cards quite soon.