Overview
- Nvidia showed developer controls for DLSS 5 at SIGGRAPH, including two global sliders called Structure and Tone plus per‑object masking to vary how much the system alters high‑frequency detail and overall lighting.
- The company also revealed three trained models labeled A, B and C that developers can switch between in real time to pick different levels of detail and visual effect.
- Nvidia says DLSS 5 works by taking a conventionally rendered frame and applying a learned generative stage anchored by internal buffers and motion vectors so the model processes frames causally and avoids temporal artifacts.
- The firm presented claims that the feature is optional and can run on a single GPU with VRAM efficiency, but it did not publish concrete performance numbers or minimum GPU requirements, leaving practical costs unclear.
- Despite Tuesday’s demonstrations, players and developers remain skeptical about whether the beautification stage can be fully bypassed and what controls Nvidia might add before the expected Q3/fall 2026 launch.