Eurogamer.net

AMD Radeon RX 580/ RX 570 review

Just how much faster are the new second-gen Polaris cards?

AMD call it Polaris refined, but to all intents and purposes, RX 570 and RX 580 are indeed upclocked versions of their predecessors. There's the same complement of 32 and 36 compute units respectively, power efficiency is diminished in the quest for higher frequencies and by extension, the emphasis this time is on increased performance. And to be frank, that's fine with us - just don't go into a 570/580 purchase thinking you're getting the new AMD Vega technology, because you're not.

That said, AMD gets a lot of heat for what are often palmed off as 'rebrands', but at the end of the day, we all buy new gaming graphics cards because we're looking for a boost to gaming frame-rates, or the ability to run at higher resolutions with richer visual effects in play. Looking over the benchmark data, it seems that AMD's primary aim with the RX 580 is to more closely challenge the GTX 1060 as the performance winner in the mainstream GPU sector, while pushing RX 570 up to and beyond GTX 970 frame-rates, giving both GTX 1050 Ti and the 3GB version of the GTX 1060 a run for their money..

For the purposes of this review, AMD sent us three samples - at the lower end, we have the Asus Strix RX 570, its 32 compute units reaching a maximum of 1300MHz. Alongside it, we have two RX 480s - factory-overclocked Sapphire and MSI samples. The Sapphire sample in particular is fascinating owing to a mammoth, weighty heat sink arrangement and boost clocks that hit a remarkable 1450MHz with no overclocking required. These are frequencies higher than any Polaris card we've seen before, and 57MHz faster than the MSI's maximum 'out of the box' clocks.

What's immediately clear from the brace of RX 580s is that the push for higher performance comes at a cost. The Sapphire card pulls in power from six-pin and eight-pin inputs, while the MSI model settles on a single eight-pin instead - both beyond the single six-pin input of the original reference model RX 480. Pushing to higher clocks, put simply, requires more juice. The Sapphire card's substantial thermal solution ensures quiet operation, but the fans do kick in more noticeably on MSI's model. A good case will muffle the noise, but it's rare that we hear much of anything noise-wise from this firm's Twin Frozr cooling assembly. Meanwhile, the Asus Strix line - known for its quiet operation - performs as expected on the RX 570, though clocks for the reduced cluster of CUs are lower at 1300MHz, making the job a little easier to handle.

Rich's video review of the RX 580. It's good, but is it good enough?

The move from AMD from reference cards to a complete third-party cooler line-up is a good idea - principally because overclocking the standard RX 480 resulted in a lot of noise, while the lacklustre thermal management caused the card to vary significantly from its max boost clocks, even without an OC in place. In short, it got hot and throttled potential performance. While not perfect, all three cards we tested here spent much more time locked to their maximum frequencies.

The question is, has AMD's performance-orientated gambit paid off? How do the RX 570 and RX 580 stack up against the Nvidia competition? And more to the point, with many outlets slashing prices hard on the RX 480, to what extent do we see improvement over the debut Polaris graphics card?

As always with a mainstream card, we kick off with 1080p benchmarks: consider these as like-for-like comparisons based on game engine throughput. It's all about judging relative performance under identical loads, as opposed to presenting a guide to how well these titles will play on your system. We've got eight games tested here, and by moving to The Division's DX12 renderer, we now have an even split between DX11 titles and those using the latest Microsoft API.

The first thing to note is that despite the Sapphire RX 580 featuring higher clocks than its MSI equivalent, only Assassin's Creed Unity sees any benefit above 1-2fps. You can push the MSI card to match the Sapphire, but the real question from these results is whether it is worth it with such a marginal return.

[UPDATE: As the RX 570 sits between the GTX 1050 Ti and the GTX 1060 3GB in its pricing, we've re-benched the former and added it to the 1080p table below.].

Every frame we've captured for our benchmarks is available to you via our frame-time/frame-rate analysis videos. The delta between all of these cards is pretty low.

1920x1080 (1080p) RX 580 8GB Sapphire RX 580 8GB MSI RX 570 4GB Asus RX 480 8GB GTX 1060 6GB GTX 1060 3GB GTX 970 4GB GTX 1050 Ti 4GB
Assassin's Creed Unity, Ultra High, FXAA 57.2 54.7 48.4 53.5 59.6 56.4 52.0 35.7
Ashes of the Singularity, Extreme, no MSAA, DX12 56.6 55.9 51.1 51.7 52.0 48.8 46.2 31.1
Crysis 3, Very High, SMAA T2x 80.5 79.0 71.2 72.8 79.3 75.8 71.6 46.4
The Division, Ultra, SMAA, DX12 65.5 66.1 59.2 60.0 59.5 55.5 49.0 32.8
Far Cry Primal, Ultra, SMAA 65.8 65.6 58.1 61.2 66.9 63.7 49.8 40.8
Hitman, Ultra, SMAA, DX12 92.9 90.8 80.6 83.2 84.0 70.6 68.0 42.2
Rise of the Tomb Raider DX12, Very High, High Textures, SMAA 80.2 77.7 67.6 70.8 77.8 73.9 70.5 44.6
The Witcher 3, Ultra, No Hairworks 72.1 70.8 62.2 65.5 68.9 65.8 60.9 40.8

It's the RX 570 vs RX 480 results that prove interesting - the 570 is slower, but not that noticeably during gaming, and in many of the benchmarks we're effectively on level pegging, and overclocking can indeed make up the difference. Compared to the slower 4GB iteration of the RX 480, the new 570 should be just as fast, if not fater. The RX 480 effectively offered DX11 performance on level-pegging with the classic GTX 970 (which we've re-benched for this piece on latest drivers) and better DX12 frame-rates, and that remains the case here with RX 570, with three out of the four DX12 titles seeing AMD push ahead. Only Rise of the Tomb Raider sees the 970 hang on to a performance lead under the latest API.

We should stress that all the Nvidia cards here are tested at reference clocks, while the two RX 580s each have fairly meaty factory overclocks in place - the Sapphire model in particular really pushes the Polaris core to its limits. In a sense, it's not a particularly fair fight, but the 'slower' MSI model offers equivalence to a standard GTX 1060, with performance leadership on DX12 renderers like Ashes of the Singularity, The Division and Hitman. The Sapphire model adds 1-2fps to the MSI performance overall, but drops a touch on The Division - perhaps owing to the MSI card offering a memory bandwidth overclock where the Sapphire card runs its GDDR5 at the stock 8gbps.

Factoring in the 3GB GTX 1060 into the equation, it's clear that the RX 570 still falls a little short here in DX11 workloads, but pulls ahead with DX12. Overall though, it's handing in great performance for a second tier product, just as its predecessor did back in the day. The custom RX 470s were criminally overlooked and offered great performance for the money - and we expect nothing less from its upclocked successor.

The real question for the PC gaming enthusiast is to what extent the RX 580 offers an improvement over the RX 480. The answer there varies between titles - across the batch, we have a seven to 13 per cent uplift, averaging somewhere in the middle. That doesn't sound so bad but there's a catch: those figures come from comparing the highly overclocked Sapphire RX 580 to a stock, reference design RX 480. The differential drops with the MSI card and will doubtless drop still further compared to an RX 580 running at its reference 1340MHz boost clock. On top of that, there'll be a big grey area comparing factory OC 480s to 580s to the point where we do have to wonder whether there's much point with the 500 series, beyond marketing.

What we will say is that the RX 580 may improve over time with driver revisions. So, to illustrate, here's a shot from our Assassin's Creed Unity benchmark - you'll note that for some parts of the sequence, the RX 480 is actually faster, contributing to a lower than expected bump to 580 performance - something we shouldn't really be seeing. This was captured from the same system running the same driver, and certainly looks like a bug as opposed to a true reflection of the new card's throughput.

These cards all work out pretty well for 1440p gameplay, but you may want to dial down settings to level out the lowest performing areas.

2560x1440 (1440p) RX 580 8GB Sapphire RX 580 8GB MSI RX 570 4GB Asus RX 480 8GB GTX 1060 6GB GTX 1060 3GB GTX 970 4GB
Assassin's Creed Unity, Ultra High, FXAA 38.1 36.5 30.9 35.5 38.3 35.8 33.6
Ashes of the Singularity, Extreme, no MSAA, DX12 49.7 49.7 44.0 45.0 45.6 43.5 39.1
Crysis 3, Very High, SMAA T2x 49.1 48.2 43.8 44.1 48.6 46.6 44.0
The Division, Ultra, SMAA, DX12 47.4 48.7 42.5 43.6 41.4 39.0 35.1
Far Cry Primal, Ultra, SMAA 47.7 47.4 41.5 43.8 45.6 42.2 39.4
Hitman, Ultra, SMAA, DX12 68.0 66.3 58.9 60.3 62.6 47.2 48.3
Rise of the Tomb Raider DX12, Very High, High Textures, SMAA 53.8 52.6 46.8 48.2 50.4 47.2 45.2
The Witcher 3, Ultra, No Hairworks 51.8 51.0 44.7 46.9 48.7 46.4 43.1

To be clear, we recommend all the cards tested here for 1080p gameplay at high frame-rates (though GTX 970's DX12 performance really isn't great) but if you're not adverse to trading quality for performance, these GPUs can provide good performance at 1440p too. And in terms of straight AMD vs Nvidia comparisons, it's fair to say that the Radeon cards hold up better at this higher resolution, with a small but significant shift in the balance of power.

Comparing the MSI version of the RX 580 to the full-fat GTX 1060 6GB, the new Radeon pushes ahead on six out of the eight titles, effectively tying on Crysis 3 and losing only on Assassin's Creed Unity. The very slight performance boost offered by the Sapphire model opens up the gap a little further. It's a fascinating result here, but do remember that we're comparing factory OC models to reference Nvidia designs. AMD DX12 results again show big boosts over Nvidia, but another mitigating factor worth remembering is that the green team's DX11 performance is often faster than its DX12 results, and outside of the Windows Store, virtually all DX12 titles also have DX11 modes, where Nvidia still holds up.

At 1440p, the RX 570 continues the trend it set at full HD - it's effectively offering GTX 970 performance or better. While the stalwart 970 overclocks like a demon (often taking it up to GTX 1060 levels), what's clear is that DX12 support is lacking. The RX 570 aces the 970 comprehensively on three out of the four titles in our test suite using the new API, and even pushes ahead here on Rise of the Tomb Raider, something it couldn't achieve at 1080p. On a related note, if you're wondering why the 3GB version of the GTX 1060 performs so badly in Hitman, that's the result of using ultra textures when there's no room in VRAM to accommodate them all - performance is significantly depressed as a consequence.

In concluding the benchmarks, we'd like to stress one thing - these tests can only represent a small window into actual gameplay on a certain PC configuration, and we find more nuances with Nvidia vs AMD comparisons in our game-specific performance tests. So for example, the RX 480 is consistently faster than GTX 1060 in COD Infinite Warfare, but the positions reverse dramatically if you're pairing the cards with a less capable processor - for example, a dual-core Intel chip like the i3 7100 or Pentium G4560. Deus Ex Mankind Divided has a demanding benchmark where Radeon cards pull ahead significantly, but in gameplay, the performance differential narrows to nothingness in many intense scenes. Obviously though, in GPU-bound scenarios, the RX 580's clocks can only mean one thing - improved performance compared to the RX 480, though not a game-changing leap.

RX 580 Sapphire RX 580 MSI RX 570 Asus RX 480 GTX 1060 6GB GTX 970
Peak System Power Draw 327W/ 338W OC 292W/ 325W OC 272W/ 285W OC 271W 230W 295W

Looking at peak power consumption from the wall in a testing like-for-like scenario, it's clear to see how AMD has achieved its performance objectives here - significantly more power is thrown at the Polaris 10 processor, 21W more with the MSI card and a substantial 56W more with the super-clocked Sapphire. Meanwhile, the speed bump with the Asus RX 570 brings it into line with the reference RX 480. In overclocking each of the Radeons, we ramped up the power as far as it would go in MSI Afterburner, upped the core clocks until we lost stability and did the same with the memory.

The Sapphire Nitro RX 580 really is living life on the bleeding edge. We could only add a paltry 10MHz (!) to its core - upping that to 20MHz resulted in a crash. This is the least amount of overclocking headroom on the core we've ever encountered in all of our GPU testing. However, the GDDR5 modules could be overclocked to 9gbps. This combination effectively added 1-2fps to the tally. The chances are we didn't need to up its power setting to anything like the +50 per cent level we dialled in. The MSI card has more headroom compared to its standard settings, but ended up at the same 1460MHz core/9gbps memory overclock as the Sapphire model. Essentially, the Nitro card extracts virtually everything the Polaris processor has to offer, while the MSI model requires rudimentary manual adjustment to achieve identical results. Power consumption obviously rises still further with an overclock in place, making the GTX 1060 look remarkably frugal.

The RX 570 is the real star of the show in terms of overclocking. The 7gbps GDDR5 memory runs perfectly happily at 8gbps, while the Asus Strix's 1300MHz core (already running 66MHz over reference spec) easily made the leap to 1400MHz. We re-benched three titles and found the 570 to outperform the reference RX 480. At 1080p, Far Cry Primal runs about 2fps faster, the Division bested the 480 by a factor of 5fps, while Rise of the Tomb Raider inched ahead by 2fps. The RX 580 might have pushed Polaris 10 further, but the RX 480 is still a good card, and to be able to beat that level of performance from the second tier 500-series offering is a great showing. Perhaps it's not surprising bearing in mind how well MSI's RX 470 ran for us back in the day, but maybe this relaunch will give the product more of the attention it deserves.

AMD Radeon RX 570/ RX 580 - the Digital Foundry verdict

With a performance uplift of seven to 13 per cent comparing a pushed-to-the-limit factory OC RX 580 to a reference RX 480 (meaning even less of a differential with factory OC 480s), the top-end RX card just got better - but not dramatically so - and the increase in performance doesn't seem to warrant rolling a whole new 500 series of GPUs. RX 580 is more of a replacement as opposed to a true evolution of the RX 480. The timing also feels a little strange, coming just 10 months after the Polaris architecture's debut. The RX 480 was and is a really decent performer, and we would have preferred a more capable successor a little further down the line. For its part, AMD is positioning this as an upgrade for R9 380 and R7 370 owners, where the performance uplift is substantial, but the same can be said for the existing RX product line too.

On-paper benchmarks show the RX 580 cutting into many of the GTX 1060's performance wins while extending the leads it has with DX12. However, it's worth remembering that our Nvidia hardware is running at reference clocks when the market is packed with third party cards, many with their own factory OCs which will muddy the results somewhat, whether you're comparing RX 580 with its Nvidia rival, or indeed the outgoing RX 480. The problem with delivering only a relatively small improvement in performance is that some existing factory-overclocked 480s could get uncomfortably close to RX 580 results, based on how well our standard reference 480 holds up.

Establishing a new 500-series line with limited performance gains also doesn't reflect well on the true innovation to come - we know that AMD has new GPUs based on the Vega architecture coming in the second half of the year, where we'll find genuine architectural improvements that do more than just facilitate higher frequencies (though Vega does that too). These will be higher-end cards but at some point, we will see that technology migrate down to the mainstream. However, in the meantime, the RX 570 and RX 580 effectively represents a holding pattern of sorts as opposed to genuine innovation. That's actually a little disappointing, bearing in mind how impactful RX 480 was in this market sector last year.

Meet the creator of Jenga Do you really know the rules? Meet the creator of Jenga

In terms of recommendations to builders of potential mainstream gaming PCs, the overall outlook remains unchanged in the great Polaris vs Pascal shoot-out: the RX 580 has generally better DX12 performance, but it is closer now in the DX11 titles where Nvidia enjoyed dominance. The price the user pays is in efficiency, where the GTX 1060 is now way ahead, and Nvidia's DX11 driver continues to be less CPU intensive - something for those with less capable CPUs to consider in a world where DX12 support still only accounts for a minority of titles. However, in most cases, those with decent i5-class hardware will get similar results between both RX 580 and GTX 1060 in most DX11 games, with performance differentials varying on a game by game basis. RX 480 vs GTX 1060 was always much of a muchness with little in the way of dramatic wins for either product - by extension, RX 580 redresses the balance in some areas where the 480 was a little weak, but not dramatically so: it doesn't deliver any kind of killing blow to the GTX 1060.

The RX 570 is a lovely product, though it really needs to be a touch cheaper to really find its niche. The cut-down Polaris offering may lose four compute units and some frequency up against its bigger brother, but this impacts its real-life performance less than you might think - it's simply a great GPU for 1080p60 gameplay, and well worth considering, especially considering the nice boost you get from overclocking. The only argument against it remains the fact that the RX 470 did much the same job using the exact same technology, again making the refresh seem just a little superfluous.

Comments (57)

Comments for this article are now closed, but please feel free to continue chatting on the forum!