It seems that a year and a half ago the solution to the mystery of “What will the raytracing of the AMD Radeon be like?”. The patent, entitled “Method and acceleration system of ray tracing based on texture processing”, describes a hybrid approach between hardware and software to raytracing. According to AMD, its concept is better than solutions that work only by hardware. Although above we have added the link to the patent, in the DSO Gaming portal you can read selected fragments if you are interested in deepening. However, let’s share our vision of the matter.
In essence, AMD is going to introduce what it calls “fixed intersection ray intersection engine”, which is a dedicated hardware that deals with processing intersections of BVH (or “bounded volume hierarchy”: a tree outline of geometric objects). This fixed-function hardware (distinct from Nvidia RT cores and much simpler) operates in parallel to the texture filtering channel in the texture processor of the GPU. In this way, ray tracing can take advantage of existing memory buffers for textures instead of storing data unique to raytracing locally – something that increases the size of chips and their complexity. In addition, as there is no software that allocates resources and has to schedule tasks exclusive to this hardware, workload is saved.
We already know that the new Sony and Microsoft consoles will have a raytracing technology based on the Navi architecture. It is possible that these chips have received some of the most advanced advantages of the RDNA architecture, and that the components mentioned in the patent correspond to those of the consoles. The patent, registered a year and a half ago, should pay off soon, be it in 2020 with the new consoles, or with the second batch of Navi cards. Remember that the first RX 5700 graphics will arrive on July 7, but as you know, they will not use this hybrid solution to trace rays in real time …