I don't see that in the hotchip I see something else.
REFERENCIA:
Textures Unit SX= 2 Texture filters Units + 8 Texture Mapping
Raytrace Acel SX = 2 Textures filters Units + 8 textures Mapping calculate RT ( I dont thing how)
RDNA2 RT acel mayby same Textures Filters Units + 8 Texture mapping + RT accel
RT accel = Clock per CU controller. Variable Clock per CU (like Mat). RT accel can not do calc itseft because is very small (is not RT hardware Nvidia) is only a clock controller but very important, when CU calc RT you dont need FP32 maybe FP16 is enough. Due to this CU can go up and accel clock to calculate RT. Is very intelligent, but less efficent than Nvidia. But good for consoles is cheap and good control for another things.
Maybe MS could not access the variable clock technology because it is a technology that has come out of collaboration with sony so they had to spend silicon to be able to perform the same calculations due to the fixed clocks.
However, until the whitter paper comes out, we won't know anything.
RT core is so big on nvidia that it led me to think that the RT accelerator is just a CLock control unit for each CU depending on the task to be performed.
nintendosmash.com
