Nvidia counters criticism of DLSS 5 and displays realistic graphics with RTX 5090 cards in new event

The executive director of the semiconductor manufacturer, Jensen Huang, used the stage of the GTC global conference to provide technical clarifications on the fifth generation of Deep Learning Super Sampling technology. Durante the presentation aimed at developers and software engineers, the company’s management directly addressed recent complaints from users who pointed out an excess of artificiality in the images generated by artificial intelligence algorithms.
The introduction of the new upscaling system generated intense debates on technology forums and social networks shortly after the release of the first visual previews. A portion of the gaming 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 questioned the use of neural networks in rendering.
In response to the protests, the manufacturer reiterated that the tool works as a complex image reconstruction ecosystem, moving away from traditional post-processing filters. The technical demonstration focused on proving that the system acts as a multiplier of hardware capabilities, preserving the visual integrity originally conceived by the software development studios.
How neural reconstruction works in graphics architecture
The neural reconstruction process does not operate in isolation at the end of the graphics card’s rendering belt. Artificial intelligence operates at all stages of frame formation, processing geometric and lighting data simultaneously to compose the final scene that is sent to the user’s monitor.
This technical approach allows programmers to achieve levels of detail that would exhaust the resources of traditional rasterization methods in just a few seconds. The technology transfers the computational load from complex pixels to the dedicated tensor cores present in the hardware, freeing the main 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. Automation does not eliminate human labor, but provides advanced digital support so that technical artists can operate with greater creative freedom.
The system requires specific operational parameters for the image reconstruction technology to function properly, demanding massive resources from state-of-the-art video cards to deliver the fidelity promised in the original specifications.
Practical tests with established franchises in the market
To support the technical specifications presented, the company showed unreleased gameplay excerpts from titles with great commercial appeal and awaited by the public. The demos 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 ultra-high performance architecture allowed the software to run at the maximum limit of all graphics options, maintaining high frame rates on high frequency monitors and eliminating visual trails during sudden camera movements.
In the specific case of the Ubisoft franchise, neural tools were employed to recreate the feudal Japão with a level of historical accuracy tied to a dynamic weather system. Shadows interacting with the day and night cycle demonstrated the algorithm’s ability to manage multiple light sources without compromising the frame refresh rate.
Preservation of visual identity by development teams
The fear of aesthetic standardization caused by artificial intelligence was one of the central points discussed by the company’s management during the technical questions 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 rendering base, but color calibration, shadow density, and texture sharpness are manually adjusted by the studios’ creative teams.
This level of autonomy prevents games from losing their unique visual identities, allowing fine adjustments ranging from ambient occlusion to prioritizing between raw performance and still image quality. The tool was designed to ensure that the creator’s original vision reaches the end consumer intact, offering deep customization options in the graphics engine.
- Sharpness adjustment adapted to 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 individually for each game.
- Dynamic balancing between frame rate and native resolution by the player himself.
Processing requirements and hardware infrastructure
The requirement for very high-level components to run the system to its full potential 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 issue 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 midrange computers and current generation desktop consoles.
Technical reception and optimization of operational resources
Several international producers expressed support for the transition to neural network-based rendering presented at the event. Diretores of studios report that raw processing savings make it possible to allocate memory to more complex systems, changing the way hardware budgets are distributed during game creation.
The manufacturer signaled that the structure of the current boards was designed to begin this cycle of innovation and establish a solid foundation for machine learning. Atualizações continuous driver builds promise to refine the use of tensor cores in the coming months, improving energy efficiency and data processing.
- Holistic reconstruction of global illumination through continuous machine learning.
- Reduced command latency thanks to real-time enabled frame generation.
- Native support for advanced graphics engines such as Unreal Engine 5.
- Maintaining visual stability at dynamic resolutions well above the 4K standard.
Technological transition and the limit of physical silicon
The official projection presented at the end of the conference indicates that traditional rendering methods have reached the physical limits of silicon. The investment in the neural ecosystem by the manufacturer is seen 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 sets the new operating standard for the technology industry. 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. Monitoring user feedback serves as the primary metric for correcting visual artifacts that still occur in high-contrast scenes. The long-term technical objective is to make the transition between the native and the reconstructed image completely imperceptible to the human eye, even in pixel approximation analyses, ensuring absolute immersion without the need to exponentially increase the physical size of graphics processors.
Continuous updates and driver improvements
Constant monitoring of real-time performance allows the company to push software fixes that improve the neural network globally. Cada new supported game feeds the artificial intelligence database, creating a cycle of continuous improvement that retroactively benefits older titles using the same version of supersampling technology.







