Modern real-time rendering uses rasterization + ray tracing + denoising together, not pure ray tracing

The current state of real-time rendering is not a choice between rasterization and ray tracing but a combination: rasterization handles primary visibility (the fast path that established hardware excels at), ray tracing adds physically based secondary effects (shadows, reflections, global illumination at ~1 ray/pixel), and denoising reconstructs a clean image from the noisy ray-traced result. Each component compensates for the others’ weaknesses. Ray tracing without denoising is too noisy to display at 1 spp. Denoising without ray tracing has nothing to denoise. Rasterization without ray tracing approximates effects poorly. Together, the three-stage pipeline produces images that look significantly better than pure rasterization, even if not yet as good as offline path tracing.

Examples

A game frame: (1) G-buffer filled by rasterization for primary visibility; (2) 1 spp ray-traced shadows + reflections; (3) DLSS/NRD denoiser reconstructs clean shadows and reflections. Final image composites all three.

Assessment

Draw (or describe) the three stages of a modern real-time ray-traced frame and explain the role of each stage. Which stage produces the final output?