Nvidia presented an updated approach to DLSS 5 during its SIGGRAPH 2026 presentation, which the company published.
The technology combines conventional upscaling with an additional generative-processing layer. Developers are promised three intensity levels, object-level controls and real-time model switching.
Core frame elements are intended to remain intact during standard upscaling before the generative layer is applied. Nvidia has not yet said whether that layer can be disabled completely.
Each frame is processed independently to avoid adding control latency. After early demonstrations used two RTX 5090 cards, Nvidia said the model can run on one GPU and consume less video memory.
The company also said the system can process 4K images in real time, although full hardware requirements have not been disclosed.
DLSS 5 is expected in the autumn, provisionally in the third quarter of 2026. It remains unclear whether support will be limited to Blackwell-generation cards or extended to older RTX hardware.
