Nvidia has unveiled DLSS 4.5 at its 2026 International CES media event as a significant upgrade to its Deep Learning Super Sampling technology, targeting better gaming performance and higher image quality. The company positions DLSS 4.5 as a major update over DLSS 4, which debuted in 2024 with the GeForce RTX 50-series “Blackwell” graphics cards. The new release is designed to make games with performance intensive path tracing effects more accessible, with the fastest RTX 50-series GPU models being able to play path traced games at 4K Ultra HD with 240 Hz.
The core of the DLSS 4.5 update revolves around two main additions, the 2nd Generation Super Resolution Transformer and Dynamic Multi Frame Generation. DLSS 4 was defined in part by its high quality upscaler, which moved away from convolutional neural network based Artificial Intelligence models to a newer transformer based model that delivered a noticeable jump in image fidelity. At CES, Nvidia said it has now updated that algorithm with an even newer transformer based model that has been trained on a much larger dataset, with the focus on further raising sharpness and detail retention while maintaining performance.
According to Nvidia, this 2nd Gen Transformer was developed using five times the compute power that went into the first-gen DLSS 4 Transformer upscaler, and it makes use of the FP8 data format that is supported by the Tensor cores of GeForce RTX 40-series “Ada” and 50-series “Blackwell” GPUs. DLSS 4.5 continues to support older 30-series “Ampere” and 20-series “Turing” GPUs, although these generations do not have FP8 acceleration, so the performance uplift over native would not be as significant as with the newer hardware. With this mix of algorithmic and hardware enhancements, DLSS 4.5 aims to push higher resolutions and refresh rates, including demanding 4K path traced workloads, to a broader segment of PC gamers.