NVIDIA is making a major leap with its DLSS (Deep Learning Super Sampling) technology, shifting from its traditional CNN (Convolutional Neural Network) framework to a Transformer model, marking a significant enhancement in upscaling performance. For years, DLSS has been a standout in the gaming world, offering remarkable performance improvements and providing an ecosystem that has benefited both developers and gamers alike.
Now, NVIDIA is taking DLSS to the next level with a massive upgrade.
This upgrade comes in the form of the Transformer model, which will likely bring impressive advancements in several key areas, from image quality to ray reconstruction. After six months of extensive work, NVIDIA has officially released the DLSS Transformer model out of beta, meaning gamers can expect it to be integrated into more titles soon.
So, what exactly does the Transformer model bring to the table? The Transformer uses a vision transformer architecture that evaluates all the pixels in a given frame, allowing it to understand the significance of each individual pixel. This method is applied across multiple frames, helping to create more detailed pixels for a superior visual experience. Notably, the Transformer model uses double the parameters of the CNN approach, resulting in better visual quality, reduced ghosting, and smoother edges.
What’s even more exciting is that this new DLSS Transformer model is compatible with all RTX GPUs, ranging from the Turing architecture to the latest Blackwell series. This means that users across a broad spectrum of GPU generations will see improvements in performance and quality. While the full extent of the upgrades is yet to be fully realized, NVIDIA claims that the Transformer model brings a huge uplift compared to CNN-based models. Early reports suggest that gamers will notice significant improvements in visual fidelity and performance.
As the DLSS Transformer model exits beta, we can expect official updates to integrate this cutting-edge technology into games within the coming months, ushering in a new era of upscaling performance.