APIs like a ray tracing API can be implemented in driver software (on-CPU) or in device firmware (on-GPU, but programmatic, not in silicon). Technically this also applies for the old fixed-function pipeline, of course, but it's worth considering the difference.
A raytracing API also isn't forced into the pipeline for every rasterized polygon like old features - hardware T&L, geometry shaders, etc - were. It's something you use on-demand in a compute or fragment shader.
A raytracing API also isn't forced into the pipeline for every rasterized polygon like old features - hardware T&L, geometry shaders, etc - were. It's something you use on-demand in a compute or fragment shader.