NVIDIA is embarking on a groundbreaking transformation with its DLSS technology, transitioning from conventional CNN-based methods to sophisticated transformer models. This shift promises a fourfold increase in computational power, significantly enhancing image quality.
NVIDIA’s DLSS has long been a frontrunner in upscaling technology, due not only to performance boosts but also to a robust ecosystem that appeals to developers and gamers alike. Now, DLSS is on the verge of a major evolution. By adopting a transformer model, it aims to introduce a spectrum of improvements, from heightened image clarity to advanced ray reconstruction.
After six months in beta, NVIDIA’s DLSS Transformer model is poised for integration into various titles. The vision transformer at its core assesses every pixel in a frame, recognizing their significance, and repeats this process across multiple frames to produce intricately detailed visuals.
This innovative approach doubles the parameters used in previous CNN models, resulting in superior graphics, reduced ghosting, and smoother edges. Excitingly, the new transformer model will be compatible with all RTX GPUs, spanning from the Turing architecture to Blackwell. While exact upgrades remain under wraps, NVIDIA suggests substantial enhancements over CNNs in the upscaling process, hinting at a noticeable difference.
With the DLSS Transformer model now out of beta, an official rollout is anticipated in the coming months, promising to revolutionize gaming visuals.
