When we covered DirectX 12 performance a few months ago, we noted a distinct split between AMD and Nvidia. Specifically, AMD picked up substantially more performance from the new API compared with Nvidia, which typically lost DX12 benchmarks even if it won them under DirectX 11. (Note that this refers to Maxwell, which was still Nvidia’s flagship GPU family back then). The exception was Rise of the Tomb Raider, which ran faster on the GTX 980 Ti as compared to the Fury X in both DX11 and DX12.

Crystal Dynamics recently released a patch for RotTR and Overclockers3D put it through its paces. What they found suggested substantial gains for AMD’s Fury X in DX12, but relatively flat performance for Nvidia owners. At 1080p, the Fury X’s performance leapt 25%, from 64 FPS to 80 FPS — and while it still loses to the GTX 980 Ti at that resolution, by 4K, the Fury X is once again in the lead. Total DX11 vs. DX12 performance with the latest patch for both GPUs is shown below:


Image by Overclockers3D

Check Overclockers3D for additional details on how the patch impacted performance at each resolution. This patch also added multi-GPU support, a vertical-sync toggle, 3D Stereo, and multi-monitor support as well. Asynchronous compute support was also prominently added, at least for AMD cards.

A shifting DX12 landscape

It’s interesting to see performance shifting so dramatically in a game that’s already been out for months. While different AMD GPUs may see different levels of improvement from these changes, a 25% uplift for the Fury X at 1080p suggests a huge gain for other cards as well.

These types of improvements are going to be game and GPU-dependent and they don’t always scale evenly between different GPUs from the same developer. When we tested the RX 480 last week, we benchmarked the GPU in Ashes of the Singularity with async compute enabled and disabled to gauge the impact of the feature. Cards like the RX 480 and R9 380 showed only a minimal performance difference between having asynchronous compute enabled versus disabled, while the R9 Nano still picked up a substantial speed boost (44 FPS with async disabled versus 50 enabled).

Still, this update is good news for AMD owners, who should be able to expect at least moderate performance improvements across the board — and it puts AMD back in the pole position against the GTX 980 Ti in 1440p and 4K in DX12. While that’s less valuable now with Pascal on the market, the 1080 and 1070 are both still shipping in limited quantities. 28nm GPU performance will remain extremely relevant for most of the market for a good while yet.