Rumor | This is how Xbox Series X and PS5 ray-tracing works: leaked patent shows RDNA2’s texture processor-based ray tracing capabilities
Tipster @rogame recently spotted a patent application that seemingly proves how RDNA2’s—and the Xbox Series X’s—ray-tracing hardware works. The patent, filed way back in 2017, details a texture processor-based method to ray-tracing via interconnected arrays of shaders, texture processors, and cache.
The texture processor itself—the ray-tracing hardware on the Xbox Series X—consists of a texture address unit, texture cache processor, filter pipeline, and ray intersection engine. While Microsoft’s tell-all reveal of the Xbox Series X specs gave us a tremendous amount of insight, the one area where information is still not forthcoming is exactly how is ray-tracing hardware will function.
Connecting RDNA2 to the patent, @rogame spotted a leaked quote which stated that “for TA/TD/TCP, nearly all GFX10 ASICS only require 10 or 12 instances (Arden would require 14, Mero would require 8). The “instances” here seemingly refer to the number of functional texture processor units per shader array: Navi 10 (the RX 5700 series) has 10 CUs per shader array, each of which is accompanied by 10 texture processor instances. “Arden,” the code name for the Xbox Series X’s GPU has 14 CUs per shader array, together with 14 instances of texture processor hardware, which appears to scale 1:1 with the number of CUs per shader array.
The leaked quote also refers to “Mero,” an upcoming RDNA2 APU with a 32CU shader configuration. Mero showed up in rumors as January. But as a 32CU APU that’s mentioned alongside Arden, it’s likely that we’re looking at the (relatively weaker) hardware powering the PS5, but which also support hardware accelerated ray-tracing.