Welcome to the start of our Radeon RX Vega coverage. Now we'd like to point out that we only received a Vega sample yesterday, on a Sunday. AMD ended up playing the role of the courier and hand delivered the sample to us because the initial Vega 64 shipment got lost along the way. Those samples should be arriving later this week though, so expect follow up coverage soon.
Due to the limited time we’ve had with Vega 56 we have put all our focus and energy on gaming benchmarks, with over 20 games lined up. Later this week, when we receive our delayed Vega 64 sample, we'll provide a more thorough review where we check for other GPU aspects such as overclocking, operating temperatures and power consumption.
On the upside, we were well aware Vega was on its way and set for release on August 14, so the second we wrapped up Threadripper testing, we changed gears and began GPU testing. All the data in this review is 100% fresh, we've recorded it over the last few days using the latest available drivers. As usual, the GPU testing was conducted on our Core i7-7700K test machine clocked at 4.9 GHz.
For now we have scores for the GeForce GTX 1080 Ti, 1080, 1070, and 1060, as well as the Radeon RX 580, and Fury X, covering performance at 1080p, 1440p and 4K resolutions.
At the head of AMD's new Radeon RX Vega family we have the liquid cooled edition of the RX Vega 64. This model features 64 compute units for a grand total of 4096 stream processors, or cores. It operates at a base frequency of 1406 MHz with a boost frequency of 1677 MHz. That’s a far cry from the 1.9 - 2.0 GHz Nvidia’s GPUs boost up to, but we knew the green team would still come away with a clock speed advantage.
Even at this frequency the RX Vega 64 LC Edition boasts a peak single precision throughput of 13.7 teraflops, that’s more than the 11.3 teraflops the GTX 1080 Ti spits out and roughly 60% more than the Fury X.
The air-cooled model known simply as RX Vega 64 has the same number of cores, they’re just clocked 11% lower for the base at 1247 MHz and 8% lower for the boost at 1546 MHz. As you might have guessed this has reduced the peak single precision throughput by 8%.
Then we have the cut down RX Vega 56 which we're testing today. It features 56 compute units for 3584 stream processors and that’s a 13% reduction in core count when compared to the 64 version.
All three models come fitted with 8GB's of second-generation high bandwidth memory (HBM 2) which provides 484 GB/s of bandwidth for the Vega 64 cards and 410 GB/s for Vega 56. The only concerning specification here is the board power rating which sees the liquid cooled model slapped with a 345 watt rating, 295 watts for the air-cooled model, and 210 watts for Vega 56.
The base Radeon RX Vega 64 "black" is coming in at $500, set to match the current MSRP of the GeForce GTX 1080, and here AMD says it will become "the new GPU king." The limited edition air cooled version will come at a $50 premium, so $550 all up. Meanwhile, the faster liquid cooled Vega 64 model will come in at $600, undercutting the GTX 1080 Ti by $100.
The Radeon RX Vega 56 will come in at $400, priced alongside the GTX 1070. In my opinion the RX Vega 56 looks to be the best value proposition of the lot and as the most affordable model we're kind of glad this is where we’re starting.
It's time we start looking at those blue bar graphs. Do note because we have tested with 25 games at 3 resolutions there are 75 (!) graphs in this review. We'll just focus on 1440p results for commentary due to time constraints, but the 1080p and 4K results will be shown alongside while discussing the 1440p numbers.
One last and very important note. Please be aware we no longer test with AMD or Nvidia reference graphics cards as we strongly recommend buyers avoid these models (due to the premium involved) and thankfully most consumers do anyway. So rather than testing with the GTX 1070 Founders Edition, we're using the MSI Gaming X model which is a little bit faster thanks to a better cooler and factory overclocking. You have the full list of graphics cards used in the table below as usual.
GPU Test System Specs
Far Cry Primal
First up we have Far Cry Primal and here is a quick look at the 1080p and 4K results. Vega 56 trails the GTX 1070 at 1080p and 4K which is odd given it pulled ahead at 1440p. The margins were slim either way but even so we was surprised to find the Radeon graphics card hitting the lead by a 5% margin at 1440p.
Tom Clancy's Ghost Recon Wildlands
Moving to to Tom Clancy's Ghost Recon Wildlands we see that AMD’s new mid-range to high-end contender trails the GTX 1070 though this time matches it at the 4K resolution. Switching back to 1440p we see that performance is much the same, technically Vega 56 was a frame slower but you know, margin of error and all, so yeah pretty much the same performance in Wildlands.
Next up we have the hugely Overwatch and this is a title that was historically favored the green team and we find that yet again that is the case here with the GTX 1070 comfortably beating the new Vega 56 GPU at 1080p and 4K. Looking at the 1440p numbers we see that the 1070 is 9% faster on average which is a decent win for Nvidia.
Battlefield 1, Watch Dogs 2, Titanfall 2, Resident Evil 7
Battlefield 1 [DX11]
Whether you’re mind controlling messenger pigeons or moonlighting as a tank driver, either hobby is best enjoyed with Vega 56. We saw a nice little bump in performance at 1080p and a solid 13% gain at 4K. For our resolution of choice Vega 56 offers 11% more performance over the GTX 1070 which is certainly a noteworthy gain and a great result for the red team in Battlefield 1.
AMD lacks the mad driver hacks they need to take the lead in Watch Dogs 2 and as a result Vega 56 falls short at both 1080p and 4K. we was actually very surprised to find the GTX 1070 providing 15% more frames at 4K. That said even at 1440p things were getting away from the new Radeon GPU as it was 7% slower with an average of 58 fps, so a bit of a disappointing result here.
Typically Titanfall 2 has been a game that’s favored the red team in the past so we was quite surprised to see the GTX 1070 matching Vega 56 in this title. Even at 4K the Radeon GPU wasn’t able to run away with it. We see pretty much identical performance at 1440p so while not a bad result for AMD, we had expected them to pick up a handy win here.
Resident Evil 7: Biohazard
Another game where we expected AMD to do well was Resident Evil 7: Biohazard and this time they do indeed come away with the win. That said Vega 56 wasn’t much faster than the Fury X in this title, at least at 1080p, the Fury X does fall away massively at 4K. That said before it’s 4GB HBM buffer limits performance the Fury X does look very mighty at 1440p and is just a few frames slower than the Vega 56 GPU which seemed unlikely. In any case a win here for AMD.
PlayerUnknown's Battlegrounds, Mirror's Edge Catalyst, Mass Effect Andromeda
You can’t do a 20+ gaming benchmark series these days without including PlayerUnknown's Battlegrounds, could you just imagine the comments section, we mean guy can’t let GTAV go. Anyway a bit of a bloodbath here, AMD clearly has some more driver development work to do. The GTX 1070 storms ahead at 1080p delivering 27% more frames. Vega kind of comes back at 4K but by this point the frame rates are so low it doesn’t matter. Meanwhile at 1440p the 1070 was 24% faster making this a very convincing win for Nvidia.
Mirror's Edge Catalyst
When it comes to hardcore parkour Vega 56 gets around okay, matching the GTX 1070 at 1080p, though it did slip away at 4K. Looking at the 1440p results the 1070 was a few frames faster so another small win for the green team.
Mass Effect Andromeda
Moving on to Mass Effect Andromeda we find Vega 56 trailing the GTX 1070 at both 1080p and 1440p which is a little disappointing. Looking at the 1440p numbers there isn’t much in it but still Vega 56 comes in a few frames slower. Typically AMD is very competitive in this title so another worrying sign for Vega.
Dishonored 2, Witcher 3, Dawn of War III, DiRT 4
Unlike Mass Effect Andromeda, we've found that in the past AMD’s not super competitive in Dishonored 2. We certainly see that here we comparing Vega 56 and the GTX 1070 at 1080p and 4K. Looking to the 1440p results the GTX 1070 was 14% faster and that’s quite a margin, more crucially whereas the 1070 average well over 60 fps, Vega fell short of the desired frame rate.
The Witcher 3: Wild Hunt
Looking back to The Witcher 3: Wild Hunt we see that Vega is competitive in this title but at the same time it’s not really beating the much older GTX 1070 either. It edged ahead at 1080p and 4K but the results won’t really wow anyone. Meanwhile at 1440p Veag 56 was a few frames faster but again doesn’t really offer anything new at the $400 price point.
Warhammer 40,000: Dawn of War III
The Dawn of War III results are very much a mixed bag and we see that here when looking at 1080p and then 4K. The Radeon GPU was slightly slower at 1080p while it pulled into the lead at 4K. Quite shockingly though it puts fourth by far its best result at 1440p where it matched the GTX 1080. This is a somewhat baffling results but having re-tested twice we was faced with the same confusing results. Not sure what’s going on here but we're keen to look into it more shortly.
DiRT 4 threw up some more highly confusing numbers so again please take these with a grain of salt until we have more time to confirm them or someone else shows the same or similar performance. Rather than disregard these results we re-tested Vega 56 as well as the GeForce 10 series cards and kept finding the exact same results. All the quality settings were the same and visually Vega looked like it was rendering everything the same.
For whatever reason Vega 56 was able to beat the GTX 1080 at 1440p and even crush the GTX 1080 Ti's minimum result. This goes against everything seen so far, so again take these numbers with a grain of salt until we can confirm them with at least one other source.
Quake Champions, Prey, For Honor, Deus Ex: Mankind Divided
Quake Champions is a game we decided to include last minute and this is the first time we benchmarked this title. Here we see some very competitive results from AMD at 1080p though things became a bit too overwhelming at 4K. Backing down to 1440p we find mixed results, AMD hard the better 1% low while Nvidia peaked a little higher. Overall performance was much of a muchness so another draw.
Prey is yet another game where AMD typically does very well and yet we again find Vega 56 struggling against the GTX 1070. It’s the minimum frame rate, those 1% lows, that look particularly bad at 1080p though by the time we get to 4K the margins are much the same. Again we see, this time at 1440p, that the minimum fps result is an issue for Vega despite the average looking quite good.
Vega 56 and the GTX 1070 again go head to head and deliver pretty much the same results, this time in For Honor. That said while it was a dead heat at 1080p the 1070 pulled ahead at 4K. At 1440p we start to see the 1070 edge ahead but really we are just talking a 1 - 2 fps difference at this realistic resolution for these mid-range to high-end GPUs.
Deus Ex: Mankind Divided [DX12]
Deus Ex: Mankind Divided is another title that we’ve tested using DX12 and this hands the advantage to Vega. AMD won quite comfortably at 1080p and remained ahead for the 4K testing as well. Meanwhile at 1440p the Vega 56 card was 12% faster than the GTX 1070. Though shockingly just 8% faster than the Fury X.
Hellblade, Crysis 3, The Division, Hitman
Hellblade: Senua's Sacrifice
Here’s another new game we've added to the list, Hellblade: Senua's Sacrifice. Unfortunately as this is a new title and it does use the Unreal 4 engine, AMD has some drier work to do here as well. Vega 56 was 10% slower at 1440p when compared to the GTX 1070 and again just a few percent faster than the old Fury X.
Another older game we've brought back for this one is Crysis 3 and despite it’s age AMD also have some driver work to do here as well. Shockingly Vega 56 was actually slower than the Fury X at 1080p, though it did pull ahead at 1080p. That said it still trailed the Fury X at 1440p as it turned in a GeForce GTX 1060 like performance.
Tom Clancy's The Division [DX12]
Okay so now we’re getting into the low level API testing and first up we have The Division using DX12. Here Vega 56 looks very impressive racing ahead at 1080p and maintaining that strong lead even at 4K. This is also seen at 1440 where Vega 56 delivered an impressive 94fps on average making it 18% faster than the GTX 1070.
The Hitman results are interesting as Vega 56 looks great at 1080p but when we jump to 4K things settle down and now AMD is only able to match the GTX 1070. This is also true for the 1440p resolution as both Vega 56 and GTX 1070 average the same 117 fps.
Doom, Total War: Warhammer, Rise of the Tomb Raider
The only Vulkan title to be included is of course Doom and here we see at 1080p, Vega 56 is able to hit the 200 fps frame cap. Even at 4K frame rates still remain above 60 fps at at all times and AMD takes the win. Looking at the 1440p results we see Vega is able to deliver 10% more performance which isn’t bad, though given the RX 580 is almost 20% faster than the GTX 1060, we was expecting more to be honest.
Total War: Warhammer [DX12]
Total War: Warhammer was also tested using the DirectX 12 API and here Vega 56 again beat the GTX 1070 by a decent margin. In fact, Vega was 17% faster at 1440p though just 9% faster when comparing the minimum frame rate.
Rise of the Tomb Raider [DX12]
Finally we have Rise of the Tomb Raider and again testing was conducted using DX12. That said keep in mind that this is an Nvidia sponsored title and as such has always favored the green team, even with using the low level API. At 1440p performance is much the same and both Vega 56 and the GTX 1070 average 64 fps.
Where Does RX Vega 56 Stand?
Short notes on Temperatures and Power Consumption
We haven’t had time to do a full battery of temperature tests but after monitoring temps for 20 minutes Vega 56 peaked at 75 degrees and the fan spun up to 40% which actually wasn’t that bad. For those wondering overclocking appears very limited at the moment but we're not 100% sure if this is a reference card issue or a software problem, it might actually be a software related issue. Anyway for a reference card it actually works quite well but as always we're keen to check out the board partner models.
Time to see just how power hungry Vega really is, well at least the cut down 56 model anyway. Here we see that the results aren’t that bad, granted Vega 56 does push total system consumption 13% higher when compared to the GTX 1070, it’s actually a small improvement over the RX 580 which we wasn’t expecting to see. So Vega 56 isn’t going to make your power supply sweat or smoke or whatever it is they do when under pressure.
Well that was interesting. For the most part it appears RX Vega 56 is either able to match the GTX 1070 or at times be the slightly faster GPU. Let’s take a look at the average results across the 25 games tested before examining a full performance breakdown.
The 1080p graph is up first, and yes, this looks exactly like what we were expecting to see. Overall Vega 56 delivered the same experience as the GTX 1070, which is, ahh let’s discuss that shortly.
At 1440p we find almost the same. This time Vega 56 is slightly faster, we’re talking a small 2.5% margin here. Overall, Vega 56 is 18% faster than the Fury X which is a decent step since it reduced total system consumption by 8%.
Finally at 4K we find similar performance between Vega 56 and the GTX 1070. Vega did edge ahead here ever so slightly, but again we’re looking at pretty much the same experience.
Let’s move on to see how they compared on a per game basis.
This certainly gives us a much better picture of what’s going on and how well Vega 56 really stacks up. First, there is that suspicious win in Dirt 4 but even if you remove that result from the equation, Vega is still 1% faster. Not exactly a pantsing but at least it’s not slower.
What’s worth noting is that Vega found success in most of the more modern low-level API titles with solid wins in The Division, Total War: Warhammer, Deus Ex: Mankind Divided and Doom. It also did well in Dawn of War III, Battlefield 1 and Resident Evil 7.
Where Vega 56 struggled was Crysis and PlayerUnknown’s Battlegrounds, and both games clearly lack driver optimization, so what we're seeing here should be possible to correct. The frame rates in Crysis 3 were all over the place, but we don’t think the fix will be all that involved. Battlegrounds is obviously a big one and AMD will want to address that very promptly.
Overall performance looks solid and RX Vega 56 is certainly a GTX 1070 contender, even if it has rocked up to the fight banging its chest over a year late. It's certainly good that we finally have some competition at these higher price points, but we feel like if you’re going to come to the party an entire product cycle late, you kind of have to hit it out of the park. That’s not what AMD has done here.
There’s still much to discuss, more testing that needs to be done, and the whole Vega 64 review still pending. Though this review gives us a pretty good idea of what's on offer today, as usual a bit of time will need to pass before we can put our finger on what will change for consumers wanting to buy a new graphics card in the next few weeks or by the end of the year.
For example, we’ve seen previously the RX 480 lagging around 15% behind the GTX 1060 upon release. Months later that gap was closed to nothing and AMD's offering became considerably more attractive. RX Vega 56 is currently matching the more mature (and entirely awesome for the past 14 months) GTX 1070, but we can easily imagine it becoming some 10% faster before too long.
But as a consumer, should you bank of that and take the Vega plunge? Or instead go with a known quantity in GeForce GTX 1070. It’s hard to discuss pricing these days since the GTX 1070 and most other high-end GPUs are selling well over the MSRP, and even then availability can be quite poor. We hate to think how poor availability of Vega is going to be over the next few weeks, if not months. So in a way you might be forced to wait and see how well Vega matures anyway.
This is what we've accomplished in the ~36 hours that we've had Vega 56 on hand. Expect more testing shortly along with a good look at Vega 64 later this week.
Pros: Solid performance across the board. 8GB of HBM 2 memory should prove useful in the long term. Matches the current performance value king, the GTX 1070, which is no small feat. Modern DX12 architecture. Freesync gaming, yea!
Cons: Vega seems to be less efficient than Nvidia's Pascal. We were expecting a bit more from the late challenger.