Is DLSS 5 a real time diffusion model on top of a 3D rendering engine?

NVIDIA CEO Jensen Huang has officially detailed DLSS 5, the next evolution of the company's Deep Learning Super Sampling technology. The core breakthrough is its architecture: a real-time AI diffusion model is applied as a layer on top of a traditional, deterministic 3D graphics rendering pipeline. Huang described this as applying a "probabilistic model" to a "deterministic one." Instead of just intelligently upscaling pixels from a lower resolution, DLSS 5's diffusion model can generate entirely new, photorealistic visual information—like finer textures, more realistic lighting interactions, and complex material details—that the base game engine did not explicitly compute. This moves beyond simple resolution enhancement into the realm of AI-assisted frame generation.

This hybrid approach aims to solve the fundamental tension in real-time graphics between performance (frames per second) and visual fidelity. The deterministic renderer provides a stable, predictable base frame, while the probabilistic diffusion model "hallucinates" high-quality details in real-time, filling in the gaps with AI-generated content that is visually coherent and convincing. The result is a potential paradigm shift where games can run at high frame rates while exhibiting a level of detail previously only possible in offline, pre-rendered scenes or at much lower performance. It represents a move from AI as an upscaling tool to AI as an integral, generative part of the rendering process itself.