The Evolution of Ray Reconstruction in DLSS 4.5
The path to higher fidelity gaming has long been hindered by the inherent noise of ray-traced rendering. When a graphics card calculates light and shadows, it cannot sample every pixel, resulting in a noisy image that requires denoising. Traditionally, this was handled by manually optimized algorithms that often sacrificed detail for performance. According to heise online, the AI model introduced in DLSS 4.5 replaces these manual algorithms by analyzing spatial and temporal data from the game engine to reconstruct missing pixels with greater reliability.
At Computex 2026, the company showcased how this second-generation transformer model builds on previous iterations. While early versions of DLSS 4.5 focused on general upscaling, the new Ray Reconstruction model specifically targets the “technical showcase games” that rely heavily on path tracing. As Wccftech reported, this update addresses a significant gap where players previously had to choose between the benefits of improved upscaling and the necessity of ray reconstruction.
Performance Gains and Technical Improvements
The technical leap in this release is rooted in a larger training dataset and more efficient hardware utilization. The new model delivers 35 percent more compute capability while processing 20 percent more parameters, all while maintaining performance levels comparable to the August 2023 version.
Key technical enhancements include:
- Enhanced Spatial Awareness: The model now processes game-engine pixel sampling and motion data more intelligently, resulting in improved temporal stability.
- Finer Developer Control: Developers gain greater authority over temporal accumulation, allowing for precise tuning of how the model responds to dynamic scenes.
- Reduced Artifacting: By anchoring spatial awareness more deeply, the model minimizes residual artifacts that often plague path-traced content.
As noted by NVIDIA, the ecosystem now supports over 1000 RTX-enabled games and apps. This massive adoption rate is driven by the consistent integration of Tensor Cores, which facilitate the AI-accelerated workflows that define the modern RTX experience.
Real-World Impact in High-Fidelity Titles
The tangible benefits of the updated model are most visible in scenes that previously challenged traditional denoisers. In “Indiana Jones and the Great Circle,” testers observed that the new model resolves snow and particle effects with significantly fewer streaks. Similarly, in “PRAGMATA” by CAPCOM, lighting responsiveness to laser effects is sharpened, and the system leaves fewer artifacts when light sources are toggled off.
The update is not merely about raw lighting; it is about visual clarity in complex textures. For instance, the famous CRT television scene in “Alan Wake 2” now displays individual lines of static noise properly, rather than blurring them into a generic gray. These improvements are part of a broader effort to bring ground-truth accuracy to real-time rendering.
Deployment and Availability for RTX Users
Starting in August, GeForce RTX users will be able to inject the new DLSS 4.5 Ray Reconstruction into 27 specific titles. This list includes industry benchmarks and highly anticipated releases such as “Avatar: Frontiers of Pandora,” “Hogwarts Legacy,” “Half-Life 2 RTX,” and “DOOM: The Dark Ages.”
Users will manage these settings through the NVIDIA app, where the model can be manually adjusted alongside upscaling and frame generation components. The company confirmed that this update will be compatible with all existing GeForce RTX graphics cards, ensuring that the improvements are not restricted to the latest hardware releases.
While speculation regarding a potential “DLSS 5” circulated prior to the Computex announcement, the company’s focus remains firmly on refining the transformer-based architecture within the 4.5 framework. For now, users can expect a more stable, sharper, and noise-free experience as these updates roll out across the supported library this summer.
As TechPowerUp noted, the integration of these sophisticated AI models continues to bridge the gap between real-time performance and movie-quality visual effects, a goal first established with the launch of the original RTX 2080 in August 2018.
