Nvidia has made the new version of DLSS 4, which has been tested in beta for a long time, stable. The update, announced at CES 2025, indicates a significant leap in artificial intelligence-supported image enhancement technology.

The new version of DLSS 4 has been unveiled

DLSS 4 has moved from the neural network-based architecture used in previous versions to a new Transformer-based model. With this transition, image clarity increases, latency decreases and significant improvements are achieved in ray tracing performance.

The basis of DLSS 4 is the DLSS Transformer model, which completely renews Nvidia’s Super Resolution algorithm. This system can analyze the position and importance of each pixel within the scene more precisely.

Thus, visual defects such as loss of sharpness and edge blur seen in previous techniques are largely eliminated. The Transformer structure goes beyond traditional methods and produces frames much closer to real resolution.

The new system directly affects not only visual quality but also performance. With DLSS 4, the number of parameters used is doubled, while a performance increase of 30% to 50% is achieved in ray tracing applications.

This improvement causes latency to decrease from 3.25 milliseconds to 1 millisecond on high-end graphics cards such as the RTX 5090. This decrease makes the gaming experience more fluid and responsive.

Not only the new generation graphics cards, but all RTX models starting from the RTX 20 series will benefit from the advantages of the Transformer model. A wide range of users, including the first RTX cards with Turing architecture, will experience a noticeable improvement in image quality with the DLSS 4 update.