NVIDIA showcased DLSS 5 neural rendering at SIGGRAPH 2026 in Los Angeles, continuing development despite initial negative gamer reaction. The company demonstrated three models—Model A, Model B, and Model C—each offering different structural intensity, global illumination, and texture detail levels with varying parameter counts. Developers can mix and match these models per scene or character, and they run efficiently on a single GPU.

A key improvement is preservation of artistic intent, addressing criticism from the original demo. Every object, scene, and setting can be customized to maintain visual identity, object semantics, character poses, lighting, motion, and art direction. Adjustable parameters allow per-scene and per-character tuning while DLSS 5 applies material photorealism within these constraints.

NVIDIA solved temporal streaming challenges by using game engine motion vectors for instant frame-by-frame texture generation, eliminating shimmer or corruption. For speed, 4K frames at 60 FPS render in under 16 ms, with most processing time consumed by the game engine rather than DLSS 5. The model operates on a single GPU with VRAM efficiency, though 4K requirements remain unclear ahead of the expected fall launch.