If you play on PC and have a GeForce RTX graphics card , you've almost certainly seen the DLSS acronym in the graphics settings menu of more than one game. Until now, it was largely associated with increasing FPS and upscaling resolution, but with DLSS 5, NVIDIA has decided to take things to the next level and change the way the final image on screen is generated. It's not just about "making it run more smoothly": we're talking about rewriting lighting and materials in real time with AI.
This new version has generated a lot of buzz among gamers and developers : on the one hand, it's being touted as the biggest graphical leap since real-time ray tracing; on the other, some see it as an AI filter that obscures the original artistic intent of games. In this article, we'll take a closer look at what DLSS 5 is, how it works, how you can use it on Windows, and why it might be worth enabling … or disabling, depending on your situation.
What exactly is DLSS 5 and how does it differ from other DLSS versions?
DLSS stands for Deep Learning Super Sampling , and it encompasses a family of neural network models that NVIDIA has been integrating into its RTX GPUs since 2018. Up to DLSS 4.5, the main focus was on two things: image reconstruction (upscaling + anti-aliasing) and frame generation to increase FPS . DLSS 5 changes the focus: it ceases to be just a tool for gaining performance and becomes a real-time neural rendering model.
When NVIDIA talks about a "neural rendering model," it means that DLSS 5 doesn't just take a pre-rendered image and refine it. It takes the raw data generated by the game engine in each frame (color, motion vectors, material information, depth, etc.) and, based on that, injects photorealistic lighting and materials into the scene. All of this is done with very strict time consistency to prevent flickering or erratic artifacts from one frame to the next.
In other words, the game's 3D world is still built with classic geometry and shaders, but the final stage is no longer just simple post-processing. DLSS 5 acts as an intelligent final layer that reinterprets how that world should look based on what it has learned during training, bringing the result closer to what we usually see in cinematic visual effects, while keeping the gameplay intact.
NVIDIA itself defines it as its biggest leap in computer graphics since the arrival of real-time ray tracing with the first RTX cards. Jensen Huang, the company's CEO, even described it as the "GPT moment" of graphics: a blend of classic, hand-drawn rendering with highly controlled generative AI.
How DLSS 5 works on the inside: AI, vectors, and neural rendering
To understand why DLSS 5 is more than just a flashy filter, it's helpful to review how these models are trained and run. NVIDIA uses autoencoder neural networks : networks that learn to convert a complex input (a frame with a lot of data per pixel) into a very rich intermediate representation, and from there reconstruct an improved output.
During training, DLSS AI is fed with enormous amounts of reference frames generated using extremely high-quality offline techniques (far exceeding what can be calculated in 16 ms, which is the budget time per frame at 60 FPS). Each pixel is not just an RGB value: it is tagged with semantic information such as depth, normal, material it belongs to, direction of movement, light sources, etc.
During real-time inference on your PC, the process would be something like this in conceptual terms: each pixel with its color and associated data becomes a feature vector, the model generates an intermediate representation of the entire frame and, from there, produces an enriched version of the image, going so far as to expand a base pixel into a much denser grid of rewritten pixels with more sophisticated lighting and materials.
Unlike generative AI that draws images from text without constraints, DLSS 5 must be deterministic and stable over time . The game engine dictates the rules: the AI is "anchored" to the geometry and motion vectors of the original 3D content. This anchoring prevents the image from changing unpredictably from frame to frame and ensures that your inputs (mouse, controller) continue to respond reliably.
All this processing primarily runs on the Tensor Cores of RTX GPUs , using reduced-precision instructions like FP8 to handle large models in real time. In newer architectures (Blackwell and successors) , neural shaders , designed specifically for this type of hybrid graphics and AI workload, also come into play.
Visual advantages: lighting, materials, and photorealistic realism
The most striking benefit of DLSS 5 is in image quality. If everything is set up correctly, you can achieve a much richer and more believable scene without the game having to rely on unwieldy geometry or absurd amounts of rays in a classic path tracer.
Among the specific improvements highlighted by NVIDIA and initial analyses are:
- Advanced cinematic lightingDLSS 5 is capable of recreating complex effects such as rim lighting, backlighting, or subsurface scattering, making the skin look more natural in different lighting conditions.
- Material depth and richer PBRThe roughness, gloss, and responsiveness of PBR materials are refined, as well as small but critical elements such as eyes, hair or tissues They gain definition and coherence.
- Temporal consistencyThe model is trained to produce a stable image from frame to frame, reducing flickering, ghosting, and other artifacts that often appear when a lot of post-processing magic is involved.
- Real-time work up to 4KAll this visual spectacle runs in real time, promising to maintain smooth gameplay, even when combining DLSS 5 with other features such as classic upscaling or frame generation.
In practice, when you see demos like Resident Evil Requiem, Starfield, or Hogwarts Legacy with DLSS 5 enabled, you'll notice a clear improvement in global illumination, contact shadows, and micro-details on faces, clothing, and metallic surfaces. All of this is achieved without the game having to explicitly simulate every light bounce with thousands of rays, something that would be impossible at a 16ms frame rate, even with top-of-the-line graphics cards, for years to come.
The real goal of DLSS 5: to bridge the gap between real-time rendering and cinema.
DLSS 5 was created precisely to address this bottleneck: the gap to photorealism cannot be closed with brute force in real time alone, so the only way forward is for an AI to learn patterns of how light and materials should behave and apply them efficiently on the fly.
That's why NVIDIA is selling DLSS 5 as the spearhead of neural rendering : a major first step towards a future where much of the image won't come directly from hand-written shaders, but from trained models that have seen millions of examples of lighting, materials, and scenes. And all while maintaining artistic control, something crucial for studios.
Relationship with ray tracing and path tracing: allies, not substitutes
A frequently asked question is whether DLSS 5 is meant to replace traditional ray or path tracing . The short answer is no: they are different technologies with different objectives, although they work very well together.
Ray tracing, or path tracing, focuses on calculating, in a more physically accurate way, how light behaves: direct and indirect illumination, hard and soft shadows, diffuse and specular reflections, caustics, etc. Each ray sent has a cost, and reducing noise requires many samples per pixel, which is incredibly expensive in real time.
DLSS 5, on the other hand, doesn't cast more rays . What it does is take the color and motion information that the game already has (which can come from traditional raster rendering, hybrid ray tracing, or limited path tracing) and reconstruct photorealistic lighting and materials as if we had fired many more rays than are actually calculated.
In practice, this allows you to reduce the number of physical rays the engine needs to achieve a convincing result, offloading some of the work to AI. And if you also enable features like Ray Reconstruction , the combination can deliver scenes with less noise and more detail without putting undue strain on the GPU.
Compatibility with DLSS Super Resolution, Frame Generation and Ray Reconstruction
DLSS 5 isn't coming alone: it's designed to integrate seamlessly into the NVIDIA ecosystem of technologies already used in many current games. Everything is organized in a multi-step pipeline where each module plays its part.
In simplified terms, the usual order would be:
- DLSS Super Resolution (upscaling)The game renders at a lower internal resolution, and this stage reconstructs the missing pixels using AI, providing advanced anti-aliasing and extra detail. It runs before DLSS 5.
- Ray ReconstructionIt acts as an intelligent "noise destroyer" when ray tracing is present, replacing simpler temporal filters. It also predates DLSS 5.
- Frame Generation / Multiframe Generation: generates complete intermediate frames using AI from real frames, completing several synthetic frames for each one rendered traditionally.
- DLSS 5 (neuronal rendering)It is placed at the end of the pipeline, after frame generation, and is where the infusion of lighting and photorealistic materials on the already reconstructed and stabilized content.
This means that the final quality you get with DLSS 5 will also depend on how the DLSS Super Resolution modes are configured (Quality, Balanced, Performance, Ultra Performance) and the selected model (for example, newer "Transformers" models tend to offer better detail).
How DLSS 5 maintains the original artistic intent of the game
One of the most frequent criticisms on social media and professional forums is that DLSS 5 "redraws" the game and can betray the art direction decided by the artists. NVIDIA is well aware of this concern and has therefore provided developers with granular controls for the technology.
Within the integration through the NVIDIA Streamline framework, studios can adjust parameters such as:
- Intensity of the effect: how noticeable is the model's intervention in terms of lighting and materials.
- Color correction and mixing: adjustments for color grading, color mixing, saturation, contrast, and brightness.
- Masking and application areas: which objects or areas of the game are affected by DLSS 5 and which are left virtually untouched.
Thanks to this control, a studio can decide, for example, that DLSS 5 should only be applied strongly to characters and certain materials , but much more subtly to backgrounds or elements where the original aesthetic is very defined. Furthermore, by using motion and color vector information from each frame , the model tends to respect the scene's structure and avoid abrupt changes.
Controversies and criticisms: photorealism or “AI slop”?
It's not all praise. Following the first demos and official videos of DLSS 5, criticism has emerged from various quarters . Part of the community fears that this technology will become a kind of "universal filter" that will make many games end up looking too similar , especially in terms of faces and skin tones.
The term "AI slop faces" has become popular on social media to describe AI-generated or retouched faces that share a certain plastic look, with very soft lighting and almost perfect skin, similar to what we see in automatic selfie editing tools. Some developers and gamers fear that if DLSS 5 is overused, games will lose visual personality in favor of a hyper-realistic but somewhat homogenous look.
Added to this is the general debate about the use of AI in creative environments: illustrators, 3D artists, lighting specialists or compositing experts fear that companies will use tools like DLSS 5 to cut production times and costs , reducing the need for manual work and further squeezing studio margins.
On the other hand, there are also those who argue that DLSS 5 is simply another tool in the developer's toolbox . There's no obligation to activate it or push it to its limits. Studios can use it to enhance specific scenes or to offer an alternative graphics mode (for example, a "cinematic" preset versus a more "original" one). Ultimately, the responsibility lies in how it's used, not in the technology itself.
Hardware requirements: which graphics cards work with DLSS 5
In the public demonstrations of DLSS 5 during GTC 2026, NVIDIA used a GeForce RTX 5090 as the reference GPU. In one of the most demanding demos, they even showcased a system with two RTX 5090s : one dedicated to neural rendering (DLSS 5) and the other to conventional game rendering. This has raised many questions about what home hardware will be needed to run it.
The company clarified that the dual-GPU configuration was an extreme setup for the demo , designed to showcase the model's unrestricted potential, and that the goal is for DLSS 5 to run on a single consumer graphics card. Even so, they made it clear that they are still optimizing the model , which is why there isn't yet a definitive list of compatible GPUs.
Based on the available information, the following can be stated:
- GeForce RTX 50These will be the cards with full and priority support, starting with the RTX 5090 and the rest of the Blackwell range.
- GeForce RTX 40Because they have powerful Tensor cores and FP8 support, they are very likely to be compatible, although some more modest models might have limitations. performance or graphics memory limitations.
- GeForce RTX 30 and earlierThey don't have native support for FP8, so in principle would be left outUnless NVIDIA decides to release a variant of the model adapted to INT8 or other specifications, something that it has not yet confirmed.
NVIDIA also noted that the model we saw at GTC 2026 was very large and resource-intensive , with VRAM consumption reaching up to 32 GB when combining the two RTX 5090s. The version that will reach gamers will be scaled down and optimized, but it's still reasonable to expect that DLSS 5 will require more than 8 GB of video memory to run smoothly in demanding titles.
Impact on performance and resource consumption
For now, NVIDIA hasn't released definitive figures on how many FPS are lost or gained by enabling DLSS 5 compared to a configuration with only DLSS Super Resolution and Frame Generation. What they have made clear is that this is a resource-intensive model, with a significant computational cost , although it's optimized for real-time performance.
It's important to understand that DLSS 5 isn't primarily designed to boost performance; its purpose is to enhance visual fidelity . This doesn't mean it will inevitably destroy FPS: by combining it with upscaling and frame generation, it's possible to maintain very high refresh rates even with a significant increase in apparent visual complexity.
What is almost certain is that DLSS 5 will have a significant VRAM footprint , as the model needs space for its weights, intermediate buffers, and input/output data. On graphics cards with limited memory, enabling it along with ultra textures and maxed-out ray tracing could compromise stability or force performance cuts.
When is DLSS 5 coming and which games will you be able to use it in on Windows?
NVIDIA has set the arrival of DLSS 5 for autumn . There's no exact date, but the window is expected to run from September to December, depending on the progress of optimization and testing with partner studios.
The official list of games announced to be compatible with DLSS 5 is already quite impressive and mixes new releases with titles already on the market that will be updated:
- AION 2
- Assassin's Creed: Shadows
- Black State
- CINDER CITY
- Delta Force
- Hogwarts legacy
- Justice
- NARAKA: BLADE POINT
- NTE: Neverness to Everness
- Phantom Blade Zero
- Resident Evil Requiem
- Sea of Remnants
- Starfield
- The Elder Scrolls IV: Oblivion Remastered
- Where Winds Meet
To this list, we must add a rather clear "and more" in NVIDIA's communication, suggesting that other ongoing projects already have development support but cannot yet be released. Among the studios that have confirmed their collaboration are Bethesda, Capcom, Ubisoft, Tencent, Warner Bros. Games, NetEase, NCSOFT, S-GAME, and Hotta Studio —all major players in the industry.
How to activate DLSS 5 on Windows and get the most out of it
In practical terms, using DLSS 5 on Windows will be very similar to what you already do today with DLSS 2/3/4. You'll need to meet three basic conditions: have a compatible GPU , have up-to-date drivers , and run a game that includes explicit support for DLSS 5.
Once that's done, the typical steps to follow in most games will be:
- open the menu Graphics or Video Options.
- Find the section of DLSS or NVIDIA Technologies.
- Activate the functions of DLSS Super Resolution in the mode you prefer (Quality, Balanced, Performance…).
- Enable, if you wish, frame generation to increase FPS, provided you don't notice latency problems or artifacts.
- Activate option DLSS 5, Neural Rendering or similar that the game offers, adjusting the intensity of the effect if it allows you to.
Each developer will decide which sliders and presets to expose to the end user . In some cases, you might only have an on/off switch with one or two general modes (for example, "Standard" and "Cinematic"); in others, you might be able to adjust finer parameters such as overall intensity or an associated color profile.
If you want to get the most out of it without any surprises, the best thing to do after activating it is to explore various areas of the game (low-light interiors, sunny exteriors, nighttime scenes, etc.) and compare them to the effect with it deactivated. This way, you can see if you like the resulting look or if you prefer a more faithful appearance to the base image.
Also keep in mind that you can always prioritize: for example, you can use DLSS Super Resolution + Frame Generation to gain FPS and leave DLSS 5 off if you don't like its aesthetics, or vice versa, activate DLSS 5 with a more conservative upscaling mode to maintain the sharpness of the original render.
To conclude, it's worth taking away one key idea: DLSS 5 isn't mandatory , but it does open the door to a type of graphics on PC that, just a few years ago, seemed exclusive to big-budget offline film productions. Depending on your hardware, your tolerance for AI, and your aesthetic preferences, it could become your new default... or something you'd rather leave on hold until it matures a bit more.
