AMD Radeon RX Vega 64 performance preview
Initial benchmarks from the top-end liquid-cooled AMD graphics card.
The embargo has lifted on AMD's Radeon RX Vega line, and our core emphasis today is on the lower-priced entry-level RX Vega 56 - pared back in terms of processing power and memory bandwidth, but still faster than Nvidia's GTX 1070. But what about the top-end RX Vega 64? Unfortunately, AMD delivered both products mid-way through last week, giving us a very limited window in which to test. We concentrated on getting RX Vega 56 review completed, but we can present some initial findings for the top-end offering.
And when we say top-end, we mean it. AMD has sampled press with the liquid-cooled version of the card, which has a significant clock-speed advantage over the air-cooled model (which we haven't seen) and is only available as a bundle deal in its 'Aqua Pack', which gives you discounts off a high-end FreeSync screen and a Ryzen 7 motherboard/processor combo. It sounds rather like a limited edition, and we would rather have reviewed the air-cooled version, which is more likely to end up in users' hands.
As things stand then, we can assume that the figures here represent Vega 64 at its very best, where it goes toe-to-toe against Nvidia's GTX 1080. The outliers we found in Vega 56 testing - issues with Assassin's Creed Unity and a curious lack of a performance uplift in Crysis 3 - remain in effect here. Disregarding these results, our 4K benchmarks here show the Vega 64 pushing ahead of GTX 1080 in Infinite Warfare with a small five per cent lead, but otherwise the results are very, very similar between the two cards.
Bearing in mind the price differential between Vega 56 and the top-tier, liquid-cooled Vega 64, the increase in frame-rates is smaller than you might imagine. Factoring out Assassin's Creed Unity (which Vega clearly has issues with) the performance bump ranges from 11 to 17 per cent, with an average bump of 13 per cent across the titles tested. It's very easy to overclock the Vega 56's HBM2 memory as well, which will close the gap in games that can use the memory bandwidth.
| 3840x2160 (4K) | RX Vega 64 Liquid | RX Vega 56 | R9 Fury X | GTX 1070 | GTX 1080 | GTX 1080 Ti |
|---|---|---|---|---|---|---|
| Assassin's Creed Unity, Ultra High, FXAA | 27.4 | 25.8 | 23.3 | 25.9 | 33.0 | 45.4 |
| Ashes of the Singularity, Extreme, 0x MSAA, DX12 | 60.5 | 54.4 | 48.8 | 48.7 | 60.2 | 76.8 |
| Call of Duty Infinite Warfare, Max, Post-AA | 79.2 | 70.4 | 60.0 | 58.5 | 75.3 | 96.6 |
| Crysis 3, Very High, SMAA T2x | 35.0 | 31.4 | 32.1 | 31.9 | 40.3 | 53.3 |
| The Division, Ultra, SMAA | 41.0 | 36.8 | 33.3 | 32.1 | 40.3 | 52.3 |
| Far Cry Primal, Ultra, SMAA | 44.4 | 38.6 | 35.1 | 33.8 | 42.3 | 55.2 |
| Ghost Recon Wildlands, Ultra, TAA | 30.8 | 26.3 | 18.6 | 23.3 | 29.6 | 37.3 |
| Hitman, Ultra, SMAA, DX12 | 60.3 | 53.1 | 48.4 | 48.4 | 60.9 | 75.9 |
| Rise of the Tomb Raider, Very High, High Textures, SMAA, DX12 | 44.5 | 39.6 | 34.0 | 36.1 | 46.2 | 60.5 |
| The Witcher 3, Ultra, Post AA, No HairWorks | 47.0 | 41.1 | 37.6 | 37.4 | 47.6 | 64.1 |
We'll be posting a full review soon - though really, testing the air-cooled version is a priority as that is the product you're far more likely to be buying. And this in turn poses the question of just how much faster the liquid edition is than the standard model, and just how close Vega 56 gets to the more common Vega 64 variant. The top-tier Vega actually comes with a secondary BIOS that mimics the power profile of the air-cooled version - this seems to reduce performance by around three per cent based on very, very early tests, but how closely this maps to the air-cooled Vega remains to be seen.
There are some further observations we can contribute at this early stage. Firstly, the liquid-cooled Vega 64 is similar to its last-gen R9 Fury X counterpart in that the robust cooling set-up here has allowed AMD to push the Vega core very close to its limits. We could only overclock the core by four to five per cent before the system became unstable, and there's not much overhead on the HBM2 memory, either.
We also encountered a troubling issue during testing. Initially, we tested the card with a Corsair RM1000i power supply, which should meet AMD's recommendation for a 1000W PSU. However, our system would randomly reset during gameplay. Initially we thought the review sample was bad, so AMD swapped in another. However, this too began to manifest the same issue - until we swapped in a 1200W Thermaltake Toughpower PSU, and this resolved the problem. The Corsair unit has shown no signs of being faulty in the past, and it's successfully powered GTX 1080 Ti in SLI - so the instability on the water-cooled Vega is puzzling. For what it's worth, the Vega 56 - even overclocked to the limit and drawing similar levels of power to the Vega 64 - worked perfectly fine on the Corsair PSU.
First impressions then? Having looked at the lower-end Vega 56 and the premium Vega 64, the cheaper model is clearly showing the best value overall. The watercooled card can meet the challenge of the GTX 1080 but it's more expensive and requires a lot of power and a closed loop cooler to meet the capabilities of Nvidia's more efficient reference card. But we should stress that these are initial numbers - we'll have a full review with across the board metrics as soon as we can.