RADEON REFRESH
AMD dials down the hyperbole with a marginal update to its graphics chip. BENNETT RING spots the difference...
While
we wait with bated breath for AMD’s new high-end GPU, the Vega series, which promises to rival Nvidia’s GTX 1080 Ti, we’ve been fed another annual refresh of their existing products. Last year AMD launched the RX4XX series to much fanfare, promising VR-ready performance at a price point that would make VR a whole lot less expensive. We covered the brand-new GPU in detail when it released in June of last year, based on the new Polaris architecture, and AMD has given it a fresh lick of paint to offer a slightly improved version of the product this year. Welcome to the Radeon RX5XX series, but don’t get too excited – not a whole lot has changed in the last twelve months.
With the new Vega architecture slated to target high-end gamers, the RX5XX range is aiming at the mid-range market. As such, prices range from $140 to $400, taking on the likes of Nvidia’s GeForce GTX 1060, but well below the $550 price of the GeForce GTX 1070. It was rather obvious AMD weren’t making too much of a fuss about these new series of cards when the first we heard of it via press releases by third-party product manufacturers, with no official information from AMD itself. When you’re launching a brand-new product range, it feels a little odd not to have a press release to accompany it, so it’s obvious that AMD aren’t hugely promoting this product refresh. So, just what have they done to the turn the RX4XX series into the RX5XX series?
THE BUDGET BATTLER RX550
There will eventually be four new products in the RX5XX series lineup, but we were only able to get our hands on two of these. At the very lowest end is the RX550, and this is arguably the most interesting card of the lot. Priced at around $140, it’s the most affordable of the range. This is based on AMD’s new Polaris 12 GPU, a very power efficient, entry-level performance card, that is built around 8 Compute Units ( CUs), yet features a 128-bit memory bus. It’s got half the CUs of the Polaris 11 design, and is built from 2.2 billion transistors. Like all of the new cards, these transistors are built using what AMD likes to call the ‘Latest Generation FinFET 14” process’. In other words, it’s the same 14nm process as last year, but 12 months of refinements have allowed them to reach higher yields, better power efficiency and higher clock speeds. It seems the P olaris 12 GPU is designed first and foremost for laptops, though it has shown up in discrete desktop cards; it’s just a shame we weren’t sent one, as it would be very interesting to see how it competes with Nvidia’s GeForce GTX 1050, which is very similarly priced. With only 2GB of onboard GDDR5 memory it’s not designed for anything higher than 1080p gaming, and we’ll endeavour to get our hands on one asap.
NEXT IN THE LINE
Next up we have the RX560, which is a rather different proposition to the two cards we did manage to get our hands on. When it came to the RX4XX series, AMD never shipped a Polaris 11 GPU with every single CU enabled, except in certain laptops. Enter the RX560, which has the full 16 CUs enabled, finally giving desktop users the full use of this GPU. Compared to its predecessor, the RX460, it’s not just the two extra CUs that give it an advantage. The Texture Units have been increased to 64 (up from 56 – part of the CU configuration), while the Boost Clock now peaks at 1275MHz instead of 1200MHz. In all other regards it’s pretty much identical though, with a 128-bit memory bus and 4GBV of 7Gbps GDDR5. However, due to the frequency increase, AMD has had to pump a little more power through the GPU, increasing the TDP up to 80W compared to 75W of its baby brother – not a large jump by any means… unlike the next two cards in the stack.
AMD’s improved manufacturing process has allowed them to eke out a little more performance
THE MAIN CONTENDERS.
The two cards we were sent are basically identical to last year’s cards, but with frequency increases. Both the RX570 and RX580 are the same as last year’s RX470 and RX480, but AMD’s improved manufacturing process has allowed them to eke out a little more performance. Both are based on the Polaris 10 architecture, but have been given more juice to increase their clock speeds. However, there is one difference AMD is pushing – a refinement in their memory technology.
In the past, if multi-monitors were in use, the card would use full power to the memory, but the new cards allow for a new mid-range power setting of 4Gbps. It also does so when playing back high-bitrate H.264 and HEVC videos, dropping to 4Gbps of bandwidth whereas the former cards were locked at 8Gbps when playing these high-res codecs. The result is a 30% saving in power to the memory modules, which isn’t insignificant in mobile variants of these chipsets.
Let’s look at the RX570 first. It has the same 32 CUs of the RX470, which means it has identical texture units and ROPs (128 and 32 respectively). However, both the base clock and Boost clock have been given an increase. The base speed of the RX470 has increased from 926MHz to a whopping 1168MHz, which is a 26% increase. While that sounds freaking fantastic, it’s worth bearing in mind that base clock speeds are pretty much irrelevant for gamers, as our GPUs tend to run at Boost speed while gaming. In that regard, the Boost speed improvements are nowhere near as
impressive. While the RX470 had a maximum Boost speed of 1206MHz, the 570 increases this to just 1244MHz, a mere 3% increase. Memory speed has increased slightly as well, up from 6.6Gbps to 7Gbps, a decent 6% increase. We should point out that our Aorus sample increases this to 1295MHz in OC mode, and to 1280MHz in gaming mode, the latter of which we tested with. We wanted to test the chipset, not the card, but we can expect most other manufacturers to offer decent factory overclocks as a result of the refined manufacturing process.
That extra frequency doesn’t come from nowhere though, with AMD having to pump more power through the cards. As a result the new RX570 draws 150W of power, compared to the 120W of its predecessor. In all other regards, the cards are identical, with 4GB of memory, a 256-bit memory bus, and 5.7 billion transistors. Considering it’s now possible to pick up an RX470 card for $269 compared to the $289 for the RX570, the price difference is minimal, so you may as well suck up that $20 for the slightly improved performance.
THE BIG MAN ON THE BLOCK
Finally we come to the top-end of AMD’s new mainstream cards, the RX580, which we were able to benchmark against the prior generation. At $399 for the Aorus RX580 8GB we were sent, it’s still not a huge price hike over existing 8GB RX480 cards, which go for basically the same price.
In return for this zero-price difference, you get a card that is basically identical to the RX480 but with some frequency boosts. The base clock jumps from 1120MHz in the RX480 to 1257MHz in the RX580, a generous 12% leap, but what we really care about is the Boost speed increase. In this case the difference isn’t quite as large, but still decent nonetheless. The RX480 has a Boost frequency of 1266MHz, which has been increased to 1340MHz in the RX580. Our review sample, the Aorus RX580GB bumps this up even more, to 1439MHz in OC mode, and 1425 in gaming mode – again, we used the gaming mode for testing to see how the new cards will function, not heavily overclocked versions of them. While our card came with 8GB of memory, 4GB versions will also be available, starting at a retail price of around $335. Given how memory hungry the Polaris 10 architecture is, we’d suggest going for the 8GB version.
Once again, that performance increase doesn’t come for free – AMD has had to pay the power piper to do so. The RX480 had a power consumption of 150W, but this has increased to 185W in the new RX 580. The memory speeds between both cards are identical, with the 8Gbps version running at 8Gbps over a 256-bit bus.
TESTING TIMES
Now, onto the testing. Unfortunately we didn’t have an RX480 here, so instead tested the new cards against a range of comparable cards.
Using our new benchmark suite, we were
NVIDIA really has stolen AMD’s thunder in the midrange, and even a frequency boost hasn’t been enough to cover the gap
less than impressed at the results. As our RX570 was only a 4GB version, we benched it against both a 4GB and 8GB RX470. As you can see, it lost out to the 8GB RX470 in every test, though the older card is $30 more expensive due to that extra 4GB of memory. And when you compare the results against the much more affordable Gigabyte GeForce GTX 1060, things go from bad to worse.
The Nvidia card took out the top spot in every test bar 3DMark, and often by quite a healthy margin. Considering it’s 25% cheaper than the Aorus RX580 8GB card, there’s simply no reason to even consider the Aorus RX580 8GB.
BRING ON VEGAS
It appears that Nvidia really has stolen AMD’s thunder in the mid-range with its GeForce GTX 1060, and even a frequency boost from AMD hasn’t been enough to cover the gap. As it stands, there’s simply no way we could recommend the new RX5XX series over the GeForce GTX 1060. Here’s hoping that AMD’s Vega lives up to all the hype, which we’ll know soon enough with Computex just around the corner. By the way, we’ll have full coverage of both the RX550 and RX560 in PC PowerPlay’s upcoming Tech Special, where we have tested every major GPU on the market – keep your eyes peeled!