AMD has finally announced “Big Navi”, with a tight 25-minute launch video. The RX 6800, 6800 XT and 6900 XT (nice) are poised to go directly up against Nvidia’s Ampere lineup, with many architectural improvements and a chunky 26.8 billion transistor chip. AMD note that many have asked them to compete at the highest end of gaming GPUs. If the performance claims hold up, that’s exactly what they’re now doing.
AMD Big Navi Announcement: The Architecture in Brief
The RX 6000 series is based on AMD’s new RDNA 2 architecture, which is significantly faster than RDNA. AMD attribute this to a wide variety of architectural improvements.
Clock Gating
Firstly, AMD mentioned pervasive fine-grain clock gating. This is likely a significant part of a claimed 54% performance/watt improvement. If clock signals aren’t sent to a tiny bit of the GPU not in use that very moment, that saves a little power. Over the card, a little adds up to a lot.
Infinity Cache
Secondly, AMD has introduced their Infinity Cache – a huge 128MB cache on the GPU itself, which AMD equate to the L3 cache on Zen CPUs. Infinity cache is a gamechanger. AMD say a mere 256-bit GDDR6 setup with Infinity Cache should perform more than twice as well as a 384-bit interface without it and use less power. The RX 6900 XT is a 256-bit GDDR6 card and, with some caveats, it looks to go toe to toe with an RTX 3090. This is shocking – the 384-bit GDDR6X memory setup on a 3090 has near 2x the bandwidth and it should be a walkover. Infinity Cache has made it a close fight.
Frequency
Thirdly, AMD has a massive 30% frequency boost on the same TSMC 7nm node. RDNA 2 uses custom libraries for the actual manufacture, optimised for higher frequency. It’s fair to say that AMD has made TSMC 7nm their own.
Smart Access Memory
Fourth (don’t worry, we’re nearly there), AMD has finally done what people have been wondering about since Infinity Fabric was introduced. AMD Smart Access Memory improves the connection between CPU and GPU, in an Infinity Fabric-like way. A Zen 3 CPU with SAM enabled can directly address 100% of the memory on an RDNA 2 GPU. Direct addressing allows for more direct data transfers. The performance gain in the games AMD have mentioned ranges from 5% to 11%.
Optimised Compute Units
Finally, the compute units themselves have been rearranged. According to Senior Director of RTG Laura Smith, the rearrangement aims to reduce how much data has to move around. Moving data takes energy, so it makes sense that doing less of that would save energy. Alongside the Infinity Cache, Smart Access Memory, and arguably even clock gating, more efficient data movement seems to be a theme with RDNA 2.
AMD Big Navi Announcement: Meet the RX 6800, 6800 XT & 6900 XT
AMD have announced three GPUs today; the RX 6800, RX 6800 XT and RX 6900 XT. Oddly enough, all three share the same memory configuration. All have 16GB of GDDR6 running at 16Gbps on a 256-bit memory bus – slim for a high-end card. In addition, all have 128MB of Infinity Cache.
With the memory setup the same for all three models, we can compare them on CU count, clock and power.
| Model | Compute Units | Game Clock | Max Boost Clock | Board Power |
|---|---|---|---|---|
| RX 6800 | 60 | 1815MHz | 2105MHz | 250W |
| RX 6800 XT | 72 | 2015MHz | 2250MHz | 300W |
| RX 6900 XT | 80 | 2015MHz | 2250MHz | 300W |
It’s surprising that AMD would cut a full quarter of the compute units, as well as reducing clocks, without cutting down the memory at all. Either GDDR6 is cheap, AMD feels 16GB is necessary for a high-end card, or maybe the 6900 XT could be limited by the memory subsystem despite the Infinity Cache.
AMD Big Navi Announcement: RX 6800, 6800 XT & 6900 XT Perfomance Claims
First up is the RX 6800 XT, which AMD is putting up against Nvidia’s RTX 3080. Now, Nvidia’s first-party benchmarks tend to use DLSS – meaning the game is run at a lower resolution and scaled up. DLSS relies on an AI algorithm trained by Nvidia, so support is limited. These are AMD first-party benchmarks, so no DLSS for you Nvidia. Native vs native at 4K, AMD showed the RX 6800 XT trading blows with the RTX 3080 across 10 games at 4K and winning at 1440p.
The RX 6900 XT went further, trading blows with the RTX 3090 at 4K – although this was with the benefit of AMD’s “rage mode” auto-overclock, as well as “smart access memory” working in synergy with a Ryzen 9 5900X. Finally, with an RTX 2080Ti subbing for the not-yet-retail RTX 3070, the RX 6800 non-XT won over the 2080TI at 4K and 1440p by a solid margin.
Overall if AMD’s numbers hold the 6900XT is competitive with an RTX 3090, the 6800XT is competitive with an RTX 3080, and the 6800 non-XT should beat an RTX 3070.
Pricing and Availability
The RX 6800 is priced at $579 – $80 more than the MSRP for an RTX 3070. Availability is due on the 18th of November.
The RX 6800 XT is priced at $649 – $50 below the MSRP for an RTX 3080 – and is also due on the 18th of November.
The RX 6900 XT is priced at $999 – a whopping $500 less than Nvidia’s “BFGPU”, the RTX 3090. AMD is holding the flagship back, and it won’t be available until the 8th of December.
