Nvidia responds to criticism of DLSS 5 and displays realistic Assassin’s Creed graphics on RTX 5090 cards

The executive director of Nvidia, Jensen Huang, used the GTC global event to confront recent questions about the Deep Learning Super Sampling technology in its fifth generation. Durante the technical conference, the executive directly addressed the concerns of users who point out an excess of artificiality in the graphics generated by artificial intelligence. The presentation focused on demonstrating that the system acts as an amplifier of technical capabilities, preserving the visual integrity originally conceived by the software development studios.
The introduction of the new system generated intense debates on technology forums and social networks after the release of preliminary demonstrations. Parte from the community expressed enthusiasm with the level of photorealism achieved, while another group criticized the dependence on algorithms for visual composition in high-budget productions, generating campaigns that mock the use of neural networks. In response, the manufacturer reiterated that the tool works as a complex reconstruction ecosystem, moving away from conventional post-processing filters.
The practical application of this innovation demands a robust hardware infrastructure, evidenced in the stress tests conducted by the company’s engineering team. Para deliver the promised fidelity, the system demands massive resources from the latest generation graphics cards, highlighting specific operational parameters for the proper functioning of the image reconstruction technology.
- Integral reconstruction of global illumination via continuous machine learning.
- Reduced command latency with frame generation enabled in real time.
- Native support for advanced graphics engines like Unreal Engine 5.
- Maintaining visual stability at dynamic resolutions well above the 4K standard.
Graphical artificial intelligence working mechanics
The neural reconstruction process does not operate in isolation at the end of the graphics card rendering belt. Artificial intelligence operates in all phases of creating the frame, processing geometric and lighting data simultaneously to compose the final scene sent to the user’s monitor.
This approach allows programmers to achieve levels of detail that would otherwise exhaust the resources of traditional rasterization methods in just a few seconds. The technology transfers the calculation load of complex pixels to the dedicated tensor cores present in the hardware, freeing the graphics processor for other simultaneous tasks.
By automating the recreation of textures and reflections, development teams gain scope to focus on the narrative and physics of virtual environments. The manufacturer argues that automation does not eliminate the human touch, but provides advanced digital support for technical artists to work with greater creative freedom.
Practical performance in large market productions
To support the technical specifications presented, the company showed unreleased gameplay excerpts from titles with great commercial appeal that are eagerly awaited by the public. The demonstrations focused on the performance of Assassin’s Creed Shadows and the remake of The Elder Scrolls IV: Oblivion, highlighting the fluidity of the animations and the accuracy of the ray tracing in complex environments.
The configuration used for the tests consisted of a system equipped with two GeForce RTX 5090 units operating in parallel. Essa extreme architecture ensured that the software ran with all graphics options at their maximum limit, sustaining high frame rates on high-frequency monitors and eliminating visual traces during sudden camera movements.
In the specific case of the Ubisoft franchise, neural tools were used to recreate the feudal Japão with a level of historical accuracy tied to a dynamic climate system. Shadows that interact with the day and night cycle demonstrated the algorithm’s ability to handle multiple light sources without compromising the frame refresh rate.
Artistic control and visual customization by studios
The fear of aesthetic standardization driven by artificial intelligence was one of the central points discussed by the company’s management during the question session. The official guarantee is that absolute control over the visual result remains the responsibility of the art directors of each project under development. The technology provides the raw processing base, but color calibration, shadow intensity, and texture sharpness are set manually by the studios’ creative teams. Esse level of autonomy prevents games from losing their unique visual identity, allowing fine adjustments ranging from ambient occlusion to prioritization between raw performance and static image quality, ensuring that the creator’s original vision reaches the end consumer intact.
- Sharpness adjustment adapted for different art directions and visual styles.
- Modulation of the intensity of volumetric effects processed by artificial intelligence.
- Integration of specific metadata to train the neural network for each game individually.
- Dynamic balancing between frame rate and native resolution by the player himself.
Need for advanced hardware and architecture optimization
The requirement for top-of-the-line components to fully execute the system raises questions about the accessibility of the technology in the short term. The current focus of demonstrations on GeForce RTX 5090 cards indicates that extreme photorealism remains restricted to the enthusiast segment of the personal computer market.
Hardware experts point out that the industry’s immediate challenge is to optimize these algorithms for less powerful and more popular architectures. The democratization of access will depend on the ability to scale the technology to mid-range computers and current generation desktop consoles.
The manufacturer signaled that the structure of the current cards was designed to begin this cycle of innovation, establishing a solid foundation of machine learning. Ongoing driver updates promise to refine the use of tensor cores over the coming months, improving power efficiency and data processing.
Monitoring user feedback serves as the primary metric for correcting residual visual artifacts that still occur in high-contrast scenes. The long-term technical goal is to make the transition between the native and reconstructed image completely imperceptible to the human eye, even in pixel zoom analyses.
Developers’ view on saving resources
Several international producers expressed support for the transition to neural network-based rendering presented at the event. Diretores technology report that raw processing savings make it possible to allocate memory for more complex systems, changing the way hardware budgets are distributed during game creation.
- Decreased time spent manually optimizing rendering code.
- Increased geometric density of objects without direct impact on latency.
- Enables extensive open-world maps without loading interruptions.
- Real-time global illumination processing with lower computational cost.
End of traditional rasterization and new graphic standards
The official projection presented at the conclusion of the conference indicates that conventional rendering methods are reaching their physical silicon limits. The investment in the neural ecosystem is treated by the manufacturer as the main and inevitable path for the continuity of graphic evolution in high-performance digital entertainment.
Deep integration between commercial graphics engines and tensor processing hardware dictates the technology industry’s new operating standard. The system’s ability to semantically understand the scene, differentiating the behavior of light when it hits water, glass or metal, consolidates the paradigm shift in the development of interactive software for the coming years.







