It seems that AMD’s RDNA 2 ray-tracing implementation is likely to be very similar to NVIDIA’s RTX. NVIDIA Turing GPUs have a separate pipeline for ray-tracing which includes BVH acceleration and ray-triangle intersection tests. This is mainly performed by the RTCores while the SM and CUDA cores continue with the standard shading and rasterization. You can read more about how this works here:
From what we’ve heard about the PS5, the consoles will use a different approach, likely due to limited resources available at higher resolutions. The Unreal 5 demo used Epic’s Lumen GI for lighting. Unlike RTX which uses a uniform algorithm for the entire scene, Lumen has a hybrid technique similar to the Crytek Noir demo.
Lumen uses ray tracing for indirect lighting, but not the triangle ray-tracing used by NVIDIA’s Turing GPU. Lumen traces rays against a scene representation consisting of signed distance fields, voxels, and height fields. These don’t require any special ray tracing hardware for BVH acceleration or ray-triangle intersection testing.
The finer details like small object shadows, AO and other related shadowing are calculated by tracing the rays in screen space. This is similar to the ReShade implementation. Medium-scale light transfer that requires less precision is carries out using SDF while voxels are used for large scale light tracing.
Coming to how the RDNA 2 based Navi 2x GPUs will resolve ray-tracing, the most likely scenario is BVH structure acceleration along with ray-triangle testing. We’ll most likely see it employed in parallel with proprietary Radeon upscaling (FidelityFX?) and/or variable rate shading to offset the performance impact. Recent RDNA 2 ray-tracing data mined by “_rogame” further cements this theory.