NVIDIA’s GeForce RTX 30 series graphics cards are primarily known for their excellent ray-tracing performance, but they have one primary drawback compared to the AMD Radeon RX 6000 series: The notably lower memory buffer. This was reflected quite clearly in Doom Ethernal at QHD and 4K with ray-tracing (source: GameGPU). The GeForce RTX 3060 12GB model was a tad bit faster than its higher-end sibling, the RTX 3070 which comes with a smaller 8GB buffer (lows: 89 vs 89 FPS). Similarly, the Radeon RX 6800 (without DLSS) and the older RTX 2080 Ti (generally slower than the 3070 in RT titles) are also ahead due to their wider VRAM buffer.
Without DLSS (with RT), the GeForce RTX 3070 is even slower than the Radeon RX 6800, netting an average of 60 FPS and lows of 53, while the 6800 manages 94 FPS (average) and 73 FPS (lows), a sizable advantage of 20 FPS or 28% at 1440p. The RTX 3060 sits between the two with an average of 80 FPS and lows of 70 FPS without DLSS. Finally, the Radeon RX 6900 XT is the second-fastest card behind the RTX 3080 with an average of 111 FPS, compared to 129 FPS on the latter.
The results are more or less the same at 1080p, with the RX 6800 scoring 129 FPS on average (and 101 lows), and the RTX 3060 and 3070 lagging by 21 and 14 FPS, respectively (without DLSS). The RTX 2080 Ti is a smidge faster with an average of 97 FPS and lows of 76. The RX 6900 XT and RTX 3080 once again grab the second and first spot, respectively.
4K is taxing on all GPUs, but the 8GB cards are brought down on their knees without DLSS, with the RTX 3070 good for just 17 FPS (average) and the 3060 Ti pegged at 15 FPS. In comparison, the GeForce RTX 3060 manages an average of 45 FPS and lows of 36 FPS while the RX 6800 settles just north of the 50 FPS mark, behind the RTX 2080 Ti and the RX 6900 XT at 56 and 62 FPS, respectively.
The VRAM consumption graph of these GPUs is rather surprising. At 4K, the RTX 3080 uses 9.8GB out of 10GB, while the RTX 3070 consumes the entire 8GB available to it. Meanwhile, the RX 6800 and 6700 XT both use just over 9GB of memory at 4K while being limited to over 8GB at lower resolutions. The RTX 2080 Ti, on the other hand, which has a video memory of 11GB doesn’t cross the 9.6GB mark at 4K while being limited to under 8.5GB at QHD and FHD.
Finally, the lowest-end card on the chart, the RTX 3060 consumes the highest amount of memory with a max usage of 10.3GB at 4K, 9GB at 1440p, and 8.66GB at 1080p. It appears that the cards with higher bus widths such as the RTX 2080 Ti and 3080 are able to keep their usage under the maximum available limit (at least for NVIDIA cards) while the RTX 3060 and the RX 6800/6900 XT use more memory due to the slimmer external bus.
Note: The game doesn’t seem to assign the entire memory buffer to the texture pool (with an upper cap of 10-ish GB), but even then, GPUs with wider buses seem to be using lesser memory than their counterparts with the same buffer size. For example, GPU memory usage in increasing order with memory bus width: RTX 3080|320-bit<RTX 2080 Ti|352-bit>RTX 3060|256-bit. This is strange because even though the RTX 3080 has a slightly slimmer bus than the RTX 2080 Ti, it has an overall higher external bandwidth due to the use of GDDR6X memory. As such, take this as mere speculation. Will update the post when I get more info.