The presentation of DLSS-5 The announcement at GTC 2026 has stirred up quite a buzz among gamers, developers, and hardware analysts. NVIDIA is talking about a historic leap in PC graphics, but for now, what we have is... Very eye-catching demos, many promises, and quite a few unknowns that have not yet been cleared.
After a start to the year marked by DLSS 4.5 and Multi Frame Generation 6xThe company is making another move with a significant shift in focus: DLSS 5 is no longer just a tool for improving performance, but a system designed for to visually reinterpret the games with an almost cinematic finish, relying on artificial intelligence models.
What exactly is NVIDIA DLSS 5 and how does it differ from previous versions?
DLSS 5 is defined as a real-time 3D neural rendering modelnot as a simple evolution of classic upscaling. We're no longer talking about a new variant of DLSS Super Resolution, but an additional layer that is applied at the end of the rendering chain to modify the appearance of the frame without altering the original geometry of the game.
Technology analyzes color data and motion vectors of each frame to understand what is happening in the scene: distinguish characters, hair, skin, fabrics, types of materials, ambient lighting conditions (front lighting, backlighting, cloudy skies, poorly lit interiors, etc.) and, from there, generate photorealistic lighting and materials anchored to the original 3D content.
In practice, this means that DLSS 5 can add very complex lighting effects without needing to spike the GPU workload as would happen with extremely aggressive ray tracing. Geometry, collisions, and gameplay not changeWhat changes is the way we see that same data on screen.
NVIDIA insists that the model maintains temporal coherence between framesThat is, the image should not flicker or break down when the camera moves, and the original artistic intention is preserved thanks to the fine controls that developers will have on the intensity and behavior of the effect.
How DLSS 5 Neural Rendering Works
At the heart of DLSS 5 is an AI model trained from start to finish to understand the semantics of the scenesimilar to projects like AI prototype for creating interactive worldsIt doesn't just apply a generic filter, but uses information from the graphics engine (color, motion, depth, materials, pre-lighting) to generate pixels with more realistic physical properties.
Among the improvements it promises are effects such as subsurface dispersion in the skin (that more natural look when light partially passes through the epidermis), more believable shines on textile or metallic surfaces, and a more convincing interaction between light and hairall while respecting the structure of the scene as defined by the game.
Technically, DLSS 5 runs in the last step of the pipelineFirst, technologies such as DLSS Super Resolution (upscaling), Ray Reconstruction (noise removal in ray tracing), and frame generation or multigeneration come into play; then, the neural model applies to that reconstructed image. its photorealistic reinterpretation.
The company claims that the system works in real time up to 4K resolutionand that it is optimized to be exploited primarily in the tensor kernels and neural shaders of the most recent architectures. However, the model shown at GTC was particularly heavyweight: the demo used Two GeForce RTX 5090s, one for gaming and one just for DLSS 5although the stated goal is for the commercial version to work with a single GPU.
To integrate the technology, the studios will use NVIDIA Streamline frameworkThis is the same environment that other DLSS variants rely on. From there, they can decide at what point in the rendering workflow it is applied, how it is combined with existing effects, and which parts of the scene should be excluded to avoid distorting its aesthetics.
Creative control, controversy, and doubts in the community
One of NVIDIA's key messages during the presentation was that The developers retain complete artistic control.DLSS 5 offers parameters to adjust intensity, color, gradation, contrast, brightness, and even mask specific areas of the scenario where it is not in our interest to activate the neural model.
This level of control is partly a response to the initial criticism that arose after the official videos and comparisons published by media outlets such as Digital FoundryIn some examples, especially in cinematic scenes from games like Starfield or Resident Evil RequiemThe results are spectacular, but at the same time they give the impression of soften or homogenize the visual style too much, bringing it closer to a type of image closely associated with AI creations that we have already seen on social media.
In certain frames, it can be seen that the facial features change slightly. The furs look more like something from an advertisement. that of an exhausted science fiction character, and some details of the setting are blurred or reinterpreted in a questionable way. Hence, there is a part of the community that fears that the AI will end eating away at part of the artistic personality of certain titles.
From the studio side, however, there have also been favorable voices. Representatives from companies such as Bethesda or Capcom They highlighted that DLSS 5 could help them achieve visual fidelity closer to cinema without sacrificing their own style, provided the adjustment tools are used correctly. The key will be in how to implement game by game and to what extent the original artistic decisions are respected.
It is important to keep in mind that, to date, what has been shown are demos and builds still in developmentMany of the games being tested with DLSS 5 weren't designed with this technology in mind from the start, so it's reasonable to expect better results when titles built with it are released. neural rendering in mind.
Relationship with ray tracing, DLSS 4.5 and the rest of the RTX ecosystem
DLSS 5 does not replace ray tracing or path tracingNor does it make technologies like DLSS Super Resolution, Frame Generation, or Multi Frame Generation 6x obsolete. What it does is add as an extra layer which builds on all of the above to go a step further in visual quality.
Ray tracing and path tracing remain the techniques responsible for calculate lighting with physical precisionmanaging reflections, shadows, and indirect lighting. DLSS 5, for its part, uses the result of that rendering (or a more traditional scheme) as a basis for Recreate an even more photorealistic look, as if the number of beams launched were virtually increased, but without the same performance cost.
In the typical NVIDIA pipeline, the flow would be something like this: the game renders the frame at a lower internal resolution, DLSS Super Resolution rescale and rebuild That image, Ray Reconstruction cleans up any ray tracing noise, Frame Generation or Multi Frame Generation generates intermediate frames, and finally DLSS 5 applies its neuronal interpretation of the result.
This structure makes it clear that DLSS 5 depends directly on the quality of the rest of the chainIf the upscaling isn't good, or the ray tracing has been configured with values that are too aggressive or too low, the neural model won't be able to work miracles. NVIDIA presents it as a one more piece within a broader transition towards “neural rendering”which includes initiatives such as RTX Neural Shading, neural texture compression, and RTX Mega Geometry.
From a marketing perspective, Jensen Huang himself has compared this move to the arrival of programmable shaders 25 years ago, stating that DLSS 5 would be “the GPT moment of graphics”Combining classic rendering with generative AI aims to achieve a remarkable leap in visual realism without sacrificing artist control. The tone is ambitious, but it remains to be seen whether this will hold true until gamers can test the technology on their own systems. if expectations match reality.
Compatible hardware, performance and power consumption: what is known and what is not
In the hardware arena, NVIDIA has been quite cautious following incidents such as the Game Ready driver with critical flaws. Today Only compatibility with GeForce RTX 5090 has been explicitly confirmedThe GPU used in the GTC 2026 demos. The company argues that the model is still being optimized and that is why it cannot provide a definitive list.
Even so, everything points to DLSS 5 It will need FP8 operations and modern tensor coresThis aligns perfectly with the RTX 50 series and, presumably, with a good portion of the RTX 40 series. At best, there could be partial support in less powerful models of these generations, although with performance or memory limitationsand while AMD drives the era of the AI-powered PC.
RTX 30 and earlier graphics cards lack Native support for FP8So, in principle, they would be excluded from the plan unless NVIDIA opted for a specific model based on INT8, something that has not yet been confirmed. For European users considering upgrading their equipment, this positions DLSS 5 as a clear incentive for the next generation high and upper mid-range.
Another aspect that remains up in the air is the impact on performance and memory consumptionThe demo with two RTX 5090s hinted at a considerable resource cost, with a potential usage of up to 32 GB of VRAM in that experimental configuration. The version intended for the average gamer will be lighter, but even so, it's reasonable to expect that More than 8 GB of graphics memory will be needed to take full advantage of it at high resolutions.
NVIDIA has promised that will publish specific performance and impact data on VRAM As the launch approaches, many users and specialized media outlets, including those in Spain and the rest of Europe, will be able to assess whether it is worthwhile to activate this additional layer or whether, on the contrary, it is more advantageous to prioritize frames per second over extreme visual quality.
Compatible games and support from major publishers
Beyond the technical aspects, one of the areas where NVIDIA has flexed its muscles is in the support from publishers and studiosThe company claims it is already working with names like Bethesda, Capcom, Hotta Studio, NetEase, NCSOFT, S-GAME, Tencent, Ubisoft or Warner Bros. Games, among others.
The list of titles announced with DLSS 5 support is surprisingly extensive for a technology that hasn't yet reached the market. Among the games already confirmed are: Starfield, Resident Evil Requiem, Assassin's Creed Shadows, Hogwarts Legacy, Delta Force, The Elder Scrolls IV: Oblivion Remastered, Naraka: Bladepoint, AION 2, Black State, Cinder City, Justice, Sea of Remnants, NTE: Neverness to Everness, Phantom Blade Zero, Where Winds Meet and more projects to be announced.
In many cases we talk about productions with a strong presence on European PC and which often make headlines in the Spanish gaming press, so it's expected that, once DLSS 5 is available, performance and image quality tests will arrive quickly. For PC gamers in Spain, DLSS 5 could become one more criterion when evaluating PC versions versus console versionsespecially in titles where the setting and lighting play a leading role.
From the studios' perspective, some public statements point to a certain enthusiasm, but also to a cautious approach. Bethesda has emphasized that DLSS 5 will allow them to The artistic style and detail shine without being tied to the classic limitations of real-time rendering.Meanwhile, Capcom emphasizes its potential to enhance the atmosphere and emotional impact in sagas such as resident evil.
Even so, actual adoption will depend on how costly it is to integrate DLSS 5 into each engine, the technical support offered by NVIDIA and whether the studios consider that the visual gains compensate for the time and resources of implementation, especially in multiplatform projects where parity between versions must be maintained.
With everything announced so far, DLSS 5 is shaping up to be An ambitious experiment by NVIDIA to bring artificial intelligence to the heart of game renderingBeyond upscaling and simple frame generation, the promise of bringing the look of video games closer to photorealistic cinema is appealing, but it comes with reasonable doubts about compatibility, resource consumption, and respect for the visual identity of each title. We'll have to wait until autumn for the first tests on real-world systems, including in the Spanish and European markets, to see if this "GPT moment for graphics" translates into an improvement that players will want enabled by default or just another advanced option, useful in some cases and dispensable in others.
