Lately, there have been many rumors going around on the web about AMD’s much anticipated Big Navi GPU. Most of these talk about the launch date of Navi 2x being sometime in October or November, while the rest detail the shader count and Compute Units. However, none of them say a word about the alleged ray-tracing capabilities of AMD’s next-gen graphics architecture. Well, fret not, that’s what we’ll talk about in this post.
A number of AMD patents have been shared on Twitter explaining how RDNA 2 (Navi 2x) will implement ray-tracing using dedicated ray-triangle intersection hardware and BVH structures. You can go through them here.
If you don’t have the stomach for all the technical details, let me simplify it for you. In short, it seems that Navi 2x will indeed feature dedicated hardware for ray-tracing. These seem to be paired with each SIMD (two per Compute Unit), very similar to how NVIDIA includes one RTCore per SM.
The implementation also appears to be nearly identical: The entire process is accelerated using BVH structures with dedicated hardware present to identify ray-triangle intersections and all the accompanying data. However, there’s no mention of an AI-based upscaling technique such as DLSS. I suspect AMD will be using FidelityFX CAS or a derivative of it.
One primary difference that I can see is that while NVIDIA uses the RTCores for both BVH acceleration and ray-triangle tests, AMD appears to be using just dedicated hardware just for the latter. It’s possible I missed something or the patents don’t mention everything as BVH acceleration needs to be done using hardware.
Furthermore, other than ray-tracing techniques, there are also patents discussing different forms of variable rate shading, use of workgroups for pixel shading as well as use of primitives for early culling and deferred vertex shading. The latter is similar to NVIDIA’s mesh and hull shaders. You can read more about that here: