Years in development, the new Radeon RX Vega line is finally arriving – AMD’s return to the higher-end of GPU performance after concentrating its efforts more on the mainstream and budget sectors of the market. There’s no shortage of cutting-edge tech here: the Vega processor utilises 12.5 billion transistors on a very large 486mm2 area of silicon, the chip paired with two 4GB stacks of cutting-edge HBM2 memory. It’s a bigger chip than Nvidia’s GTX 1080 Ti and it has more memory bandwidth – however, the top-end RX Vega 64 offers performance more in line with GTX 1080, while the cut-down RX Vega 56 reviewed here is clearly aimed to compete with the GTX 1070. It does this job rather well.
Before we go on, it’s worth stressing that the Vega architecture has a lot of features that may not be utilised much – if at all – today, but could make a big difference in future. For example, the double-speed FP16 support found in PlayStation 4 Pro was pulled ahead of time from the Vega spec sheet (referred to as ‘rapid-packed math’), while AMD has incorporated the most fully featured implementation of DirectX 12 features found in any GPU to date. Meanwhile, the high bandwidth cache controller is an attempt to allow the GPU fast, low latency access to memory beyond the 8GB of onboard RAM.
Elsewhere, the basic layout of Vega looks remarkably similar to the Fiji processor at the heart of the last-gen Fury X, but AMD is keen to point out that every aspect of the GPU has been retooled. There are still 64 compute units and 4096 shaders in the processor – pared back to 56 and 3854 in this cut-down model – but these are ‘new’ compute units with an extended instruction set and a more efficient design. AMD has also sought out faster clocks from the Vega silicon. The last-gen Fiji topped out at 1050MHz, Polaris was coaxed to 1450MHz while Vega is set to hit an impressive 1700MHz – although, again, expect a hit to that in the pared back Vega 56.
Hitting these clocks clearly hasn’t been easy though: even in its reduced state, Vega is undoubtedly power-hungry compared to the competition, as we’ll see later on in this review. That’s borne out by the design of the reference card we’re looking at here: it has two eight-pin power inputs, an even beefier configuration than the Titan Xp, despite its 210W board power rating. By comparison, the GTX 1070 Founders Edition only requires one eight-pin power input, making it easier to integrate with lower-end and medium level power supplies.
The Vega 56 card itself is a robust reference model from a company with a mixed history here – the new offering is an all-metal design, well-built, and features a vapour chamber heatsink (utilised on both Vega 56 and 64), a spec point Nvidia reserves only for GTX 1080, 1080 Ti and current-gen Titan products. External ports are a simple trio of DisplayPort 1.4s along with HDMI 2.0 – a good enough selection, but dual-link DVI should really be included on a card of this class. In operation, the Vega 56 has a temperature limit of 85 Celsius, but typically runs significantly under that. What’s more, there are power saver, balanced and turbo options. We’re reviewing in the default balanced mode, but turbo adds a couple of per cent to performance but increases power consumption.
|Radeon RX Vega 56||Radeon RX Vega 64||Radeon RX Vega 64 Watercooled|
|Peak FP32 Performance||10.5TF||12.66TF||13.7TF|
Before we diving into performance testing, AMD is making a very good point about a specific Radeon feature that proves transformative to gameplay – support for FreeSync. Variable refresh is a genuine game-changer, seeking to eliminate screen-tear and reducing the stutter you see when dropping frames with v-sync engaged. Once you have a VRR screen, it’s difficult to go back to a standard display. Now, of course, Nvidia has its own solution – G-Sync – and while the implementation is generally stronger and more uniform on those screens, they still command a significant price-premium. FreeSync, by contrast, is more likely to be a value-added extra with only a small cost increase. The bottom line is that the price for attaining a smooth variable refresh experience with a GTX 1070/Vega 56 level card is undoubtedly lower with AMD.
We’ll kick off our gaming analysis with a look at 1440p resolution – what we consider to be the best pixels vs performance target for running games at very high or ultra settings on this class of GPU hardware. Whether you’re using the standard 16:9 2560×1440 (as tested here), or else the ultrawide 3440×1440, you’re going to get an excellent experience from all of the cards tested here – high frame-rates and lots of eye-candy. It’s also fertile territory for FreeSync displays, a really good partner for the Vega 56.
For the arrival of RX Vega, we decided to beef up our GPU test suite with a couple of new entrants – Call of Duty: Infinite Warfare has impressed us with its immense level of optimisation and new rendering technologies. It’s also a title that generally runs better on AMD hardware (though Fury X doesn’t hold up so well here, curiously).
Also joining the fray is Ghost Recon Wildlands – the year’s best-selling game (in the UK at least). At ultra settings, this is one of the most GPU-heavy games we’ve tested, and it sucks up VRAM too, meaning that even at 1080p, the old Fury X really struggles here, artificially depressing its frame-rates the higher up the resolution chain you go.
1440p and its ultrawide variant offer the best quality levels and pixel-count for the Vega 56’s capabilities – and it’s ahead of the GTX 1070.
|2560×1440 (1440p)||RX Vega 56||R9 Fury X||GTX 1070||GTX 1080||GTX 1080 Ti|
|Assassin’s Creed Unity, Ultra High, FXAA||48.5||42.0||51.8||65.1||83.3|
|Ashes of the Singularity, Extreme, 0x MSAA, DX12||69.2||64.4||63.1||76.0||92.9|
|Call of Duty Infinite Warfare, Max, Post-AA||112.9||90.6||97.5||113.8||124.7|
|Crysis 3, Very High, SMAA T2x||66.0||66.1||66.9||83.4||108.2|
|The Division, Ultra, SMAA||65.3||55.7||57.8||71.3||90.9|
|Far Cry Primal, Ultra, SMAA||69.3||58.3||62.3||77.3||100.1|
|Ghost Recon Wildlands, Ultra, TAA||42.4||32.1||39.8||48.2||59.0|
|Hitman, Ultra, SMAA, DX12||93.5||82.9||83.8||103.4||127.1|
|Rise of the Tomb Raider, Very High, High Textures, SMAA, DX12||76.1||62.0||69.7||89.5||116.7|
|The Witcher 3, Ultra, Post AA, No HairWorks||72.9||61.4||68.0||84.1||109.5|
Next up, 4K results – and a question. Is it possible to get a good native ultra HD experience from Vega 56/GTX 1070-class hardware? The established wisdom is that GTX 1080-class hardware is where you start in terms of running top-tier titles but we’ve found that combining the GPU with an adaptive sync screen is a game-changer, as we discovered when gaming on a 1070-powered laptop with a 4K G-Sync display. Tweak settings to remain above 40fps and you can have a great experience. The 30-60fps No Man’s Land – where previously you to put up with v-sync stutter or tearing – is no longer an issue with adaptive sync tech.
In order to can get an idea of scalability between resolutions, our benchmarks remain maxed out here and don’t represent an ideal gaming experience – it’s all about judging relative performance between GPUs with these numbers and in that respect, the Vega 56 extends its lead over GTX 1070 here by a couple of percentage points on most of the titles tested. Even with the depth of field issue, AC Unity draws level and the only area of parity remains the puzzling performance of Crysis 3 on the new hardware. Factoring out those outliers, the overall lead is 12.8 per cent in favour of the AMD card.
Elsewhere, there are other results that once again show the stock Vega 56 operating at a kind of mid-point between GTX 1070 and GTX 1080 performance – Ashes of the Singularity seems to have regained its lead on AMD hardware after some driver revision work by Nvidia, but The Division, Far Cry Primal, Ghost Recon Wildlands and COD Infinite Warfare represent half of our tested games where the Vega 56 posts significant results that sit midway between 1070 and 1080 performance.
The Hitman result is another outlier – Vega 56 is still faster than GTX 1070, but the average frame-rate result is impacted a touch by occasional stutter, seen across all resolutions. It hits hardest on scene changes (but manifests occasionally elsewhere too) and is unlikely to affect gameplay.
You can game at 4K effectively with this class of card, but judicious settings tweaks are required and an adaptive sync display is a game-changer for the experience.
|3840×2160 (4K)||RX Vega 56||R9 Fury X||GTX 1070||GTX 1080||GTX 1080 Ti|
|Assassin’s Creed Unity, Ultra High, FXAA||25.8||23.3||25.9||33.0||45.4|
|Ashes of the Singularity, Extreme, 0x MSAA, DX12||54.4||48.8||48.7||60.2||76.8|
|Call of Duty Infinite Warfare, Max, Post-AA||70.4||60.0||58.5||75.3||96.6|
|Crysis 3, Very High, SMAA T2x||31.4||32.1||31.9||40.3||53.3|
|The Division, Ultra, SMAA||36.8||33.3||32.1||40.3||52.3|
|Far Cry Primal, Ultra, SMAA||38.6||35.1||33.8||42.3||55.2|
|Ghost Recon Wildlands, Ultra, TAA||26.3||18.6||23.3||29.6||37.3|
|Hitman, Ultra, SMAA, DX12||53.1||48.4||48.4||60.9||75.9|
|Rise of the Tomb Raider, Very High, High Textures, SMAA, DX12||39.6||34.0||36.1||46.2||60.5|
|The Witcher 3, Ultra, Post AA, No HairWorks||41.1||37.6||37.4||47.6||64.1|
If you want the highest frame-rates – but also the most variable gameplay experience – 1080p is the place to be. There are some remarkably high frame-rates here, but the averages don’t tell the story of the often janky experience you get here as you careen between CPU, GPU and even storage bottlenecks. It’s probably best seen here on the COD Infinite Warfare test, where 1080 and 1080 Ti hit the limit, with Vega 56 getting very close – the average grabbed down a touch by stutter you don’t seem to get on Nvidia cards.
We use benchmarks to get an idea of relative performance under identical workloads, but it’s fair to say that we’re hitting so many different limits here that’s it’s not hugely instructional in terms of getting value from your GPU. However, there are a couple of takeaways – principally in how competitive the Vega 56 is overall here at lower resolutions compared to the Fury X, GTX 1070’s prior competitor in this space.
Looking back at our Fury X review, we saw a GPU competitive with Nvidia at 4K resolutions, but with gradually worsening relative performance the lower down the resolution chain you went – we even saw it on the pixel-rich ultra-wide 1440p tests we carried out. The driver improved over time and performance improved but even here you can see that Fury X performance worse here against the 1070 than it is at higher pixel-counts. With Vega, the fact is that AMD is still faster than GTX 1070, by and large, or at least maintaining parity. Whether it’s down to the hardware design or driver optimisations is difficult to say but the numbers speak for themselves.
Whether you’re gaming with Nvidia or AMD though, it’s difficult to come to any conclusion other than that you’re leaving a lot of GPU power on the table by deploying such a powerful GPU on such a low resolution overall – RX 570, RX 580 and GTX 1060 offer a far more consistent experience overall for full HD gamers.
1080p offers the highest frame-rates but also the most inconsistent experience. You don’t get the best out of any of these GPUs at this resolution.
|1920×1080 (1080p)||RX Vega 56||R9 Fury X||GTX 1070||GTX 1080||GTX 1080 Ti|
|Assassin’s Creed Unity, Ultra High, FXAA||71.4||66.8||81.0||99.3||124.8|
|Ashes of the Singularity, Extreme, 0x MSAA, DX12||80.0||75.5||71.2||85.5||98.8|
|Call of Duty Infinite Warfare, Max, Post-AA||122.9||105.7||118.8||124.6||124.9|
|Crysis 3, Very High, SMAA T2x||102.8||102.3||106.7||129.1||161.6|
|The Division, Ultra, SMAA||89.4||73.7||81.6||98.6||125.3|
|Far Cry Primal, Ultra, SMAA||95.7||75.9||90.4||107.7||134.4|
|Ghost Recon Wildlands, Ultra, TAA||53.2||35.3||52.3||61.3||72.7|
|Hitman, Ultra, SMAA, DX12||125.9||106.4||112.6||133.4||152.1|
|Rise of the Tomb Raider, Very High, High Textures, SMAA, DX12||110.5||86.5||107.7||133.9||173.3|
|The Witcher 3, Ultra, Post AA, No HairWorks||100.1||79.2||95.1||114.6||138.9|
If the new AMD architecture has a weakness, it is power consumption, where the Vega 56 posted a 25 per cent increase in power consumption in our Crysis 3 stress test, where the card is handing in exactly the same performance as the GTX 1070. However, we’re clearly looking at a substantial improvement in terms of both efficiency and performance compared to the old Fury X.
You’ll note that a fully maxed out overclock sees power consumption go through the roof. Here we’re running the Vega 56 card with its power slider maxed at +50 per cent, core frequency bumped by 11 per cent and with HBM2 RAM pushed to 950MHz, putting the card on par with the Vega 64’s prodigious 484GB/s of memory bandwidth. Generally, this overclock provides anything between a seven per cent to 14 per cent performance uplift, but the fans need be adjusted upwards to keep temperatures in check. With that in mind, pushing Vega 56 hard is a task probably better suited to third party cards that are coming in Q4 this year.
As expected, those games where performance sits at a mid-point between GTX 1070 and 1080 performance can match the higher end card’s stock frame-rates, or get to within margin of error. The Division overclocks to 72.5fps vs the 71.3fps on GTX 1080, for example. Hitman and Ashes of the Singularity match GTX 1080, as does Far Cry Primal. We tested overclocking at 1440p resolution, so results may be even more favourable at 4K.
Another route forward is simply to overclock the HBM2 memory to 950MHz and leave everything else alone. This requires little extra power (we gave it an extra five per cent but it may not even require that), it won’t impact thermal performance that much (if at all) and we noted that in titles like Hitman and The Witcher 3, overclocking memory alone got you approximately half of the performance increase of the maxed OC results. It’s worth experimenting with as it is essentially free performance with no downsides – the best kind of overclocking, in our opinion.
This scene from Crysis 3 is easily repeatable and incurs a heavy, sustained load on the GPU – good for testing overclock stability and for measuring power draw. We retain our Core i7 6700K here, but disable its overclock.
|RX Vega 56||RX Vega 56 OC||R9 Fury X||GTX 1080 Ti||GTX 1080||GTX 1070|
|Peak System Power Draw||330W||457W||385W||384W||303W||263W|