AMD Radeon RX 5700/ 5700 XT revealed: full Navi specs and analysis
Team Red says they're faster than RTX 2060 and RTX 2070.
AMD has taken the wraps off its first brace of Radeon Navi products, revealing the RX 5700 and the more powerful RX 5700 XT. Both are based on a new processor known as Navi 10, using the new RDNA micro architecture - with the products designed to outgun Nvidia's RTX 2060 and RTX 2070. The two cards cost $379 and $449 respectively, undercutting their Nvidia counterparts but lacking features including hardware accelerated ray tracing. Both launch on July 7th, alongside AMD's new Ryzen 3000 line of CPUs.
Full specs for the products reveal that the RX 5700 XT uses a fully enabled Navi 10 chip, featuring a total of 40 compute units and 2560 shaders. Peak compute tops out at 9.75TF - significantly lower than the existing Radeon RX Vega 64. However, AMD has massively revamped its compute unit architecture, with a big improvement in performance. The end result is that despite a 24 CU deficit, AMD says that the 5700 XT has a 14 per cent performance lead over RX Vega 64 using 23 per cent less power.
The standard RX 5700 loses some of the finish of the XT's shroud, but cuts elsewhere look surprisingly modest. The 40 CU total in the higher end model drops to 36 in the RX 5700, while typical gaming clocks are dropped by 130MHz. What's interesting here is that both units deliver eight gigabytes of GDDR6 memory, with the same 14gbps modules delivering 448GB/s of memory bandwidth - so there's no reduction in memory throughput here on the cheaper card.
Performance-wise, AMD says that the RX 5700 delivers an average 10 per cent lead over the RTX 2060, while the RX 5700 XT normalises at around a six per cent advantage over the RTX 2070. However, there are outlier titles including Forza Horizon 4 and Battlefield 5 where AMD reckons it delivers massive performance advantages of over 20 per cent at 1440p resolution.
The notion of Navi's 9.75TF outperforming Vega's 12.6TF may well be the reason why the next-gen consoles - both based on AMD's latest GPU architecture - won't be marketed based on teraflop numbers alone in the way that Xbox One X was. AMD has pushed IPC hard with Navi, with a new compute unit design with a 2x increase in instruction rate, twice the scalar units and schedulers. There's reduced execution latency, improved cache efficiency and a new cache hierarchy. Today's teraflop with Navi clearly delivers a lot more real-life gaming performance than last gen's Vega.
Radeon RX 5700 XT | Radeon RX 5700 | |
---|---|---|
Compute Units | 40 | 36 |
Stream Processors | 2560 | 2304 |
Base/Game/Max Boost | 1605MHz/1755MHz/1905MHz | 1465MHz/1625MHz/1725MHz |
Memory | 8GB GDDR6 | 8GB GDDR6 |
Memory Bandwidth | 256-bit/ 448GB/s | 256-bit/ 448GB/s |
ROPs/Texture Units | 64/160 | 64/144 |
TDP | 225W | 180W |
There are other fascinating elements to the Navi package - including the arrival of a new display controller that carries through delta colour compression right through to compatible monitors - including a new Asus display that'll hit with a 43-inch HDR 1000 panel capable of 120Hz gaming at 4K resolution using a single cable. Interestingly, AMD is retaining HDMI 2.0 output, which suggests a 4K60 limit on that output.
Other features Navi delivers are more software-based. Anti-lag sees the driver attempting a deeper level of synchronisation between CPU and GPU in order to reduce 'button to response' latency. Impressive results are mooted here, with an average 14ms reduction in lag - which could be seriously good news for esports players. Curiously, the tech only works right now with DX9 and DX11 titles.
Also promising is the arrival of Contrast Adaptive Sharpening, developed by FXAA author Timoty Lottes. The idea is to concentrate sharpening on low contrast areas, heightening detail in an artefact-free manner, with no halo edge ringing. This can be combined with an optional upscale - the idea being to maximise visual results on, say, a 4K display while rendering at a lower native resolution (1800p was mentioned as a recommended resolution to try). AMD is positioning this as an alternative to Nvidia's DLSS, with just a one to two per cent hit to performance - in contrast to Team Green's more expensive solution.
I do think the notion of a direct comparison is a little off, as the two technologies are doing very different things and each will have their own plus and minus points. CAS has no temporal component, for example, it's screen-space only. Meanwhile, DLSS has some clear artefacting issues, but temporal stability is excellent and it is adding additional detail to the scene as opposed to sharpening a native resolution image. I get the idea that CAS is aimed more towards addressing the in-surface softness introduced by the temporal anti-aliasing and temporal super-sampling used in most modern game engines, where I suspect it will perform beautifully. I'm not sure how effective an upscale on top of that can be, but the brains behind this tech are top class and we'll be sure to check this out more closely when we can.
On a broader level, the arrival of hard facts and figures on Navi and RDNA are welcome. AMD cites a 2.3x increase in performance from silicon area vs the 14nm Vega 64, and the fact that a part with 40 compute units can comprehensively best a last-gen 64 CU part is possibly the biggest leap we've seen from AMD since Graphics Core Next launched way back in 2011. If GCN's apparent 4096 shader/64 CU limitation is still in place, a prospective Navi 20 with maximum CUs should still deliver a potent punch. Also interesting is that we have a firm die size measurement for Navi 10 - 251mm2. That's up from the 36 CU RX 480/580/590 based on Polaris 10, which came in at 232mm2. Transistor count has almost doubled between the two parts, despite Navi 10 only delivering four extra CUs.
There are next-gen console implications too. The processors in Sony and Microsoft boxes have typically been in the 320-360mm2 area, meaning that with eight Zen 2 cores on top of Navi 10, we're already getting very close to a console's typical silicon budget - and of course, it's highly likely that four or more CUs may be disabled to increase production yields.
Overall, Navi is looking interesting - though perhaps the performance win against Nvidia is more useful rather than game-changing, especially at what I'd consider to be higher than expected price-points. And of course, the Navi cards lack hardware ray tracing acceleration and desirable features found in Nvidia's Turing, such as variable rate shading. AMD says the market isn't ready for RT yet, but DXR software is appearing and the bottom line is that it'll run on Nvidia cards but not AMD. Meanwhile, both Microsoft and Sony are including RT as a feature in their next-gen consoles: clearly ray tracing is important, it is happening, open standards are there and the fact that MS has confirmed it has a hardware accelerated implementation in turn suggests strongly that the technology exists within AMD right now, even if it's not quite ready for show time in current Navi parts.
The AMD RX 5700 and RX 5700 XT are due for release next month and we'll bring you detailed reviews as quickly as we can.
We attended AMD's tech day for press and analysts in Los Angeles in order to bring you this coverage. AMD covered travel and accommodation costs.