NVIDIA DLSS 5 Neural Rendering Tested: Photorealism and Performance

Having spent several hours in direct, deep performance and image quality testing, the unvarnished truth about NVIDIA's DLSS 5 neural rendering for GeForce RTX 50 series GPUs is a mix of impressive visual fidelity and steep computational demands. Today marks the official availability of DLSS 5, provided you have a GeForce RTX 50-series graphics card and wish to play NBA 2K27. While NVIDIA has promised DLSS 5 for many other titles in the future, including Assassin's Creed Shadows, Hogwarts Legacy, Resident Evil Requiem, Starfield, and Oblivion Remastered, NBA 2K27 serves as the first showcase. Consequently, this evaluation focuses exclusively on that title.

DLSS 5 Neural Rendering (NR) utilizes a specialized AI model to apply generative post-processing to video games in real time. The specific model trained and deployed by NVIDIA comes in three variants, all designed to produce better photorealism. Users do not select which model to use in-game; rather, the developer selects its preferred version along with many other configurable controls for the technology. This report does not delve into the technical mechanics of how DLSS 5 works, but rather focuses on what it is, what it is not, how it performs, and how it looks in practice.

Understanding the Technology: A Post-Processing Filter

DLSS 5 Neural Rendering can be integrated into a game engine by developers using specialized masks to tell the model how to treat various objects in the scene, or to ignore them altogether. However, it is crucial to understand that DLSS NR is fundamentally a post-processing filter. It takes data from the game engine, feeds it through an AI model, and the output is what the user sees. The model in question is a pixel-space per-frame diffuser, essentially an image generator similar to Stable Diffusion or FLUX. The primary difference is that the "prompt" is the game engine output rather than text.

In this sense, DLSS 5 NR is somewhat analogous to the universal post-processing mod ReShade, the successor to SweetFX. ReShade allows users to insert shader programs into game renderers, read depth buffers, and alter tonemapping and color grading. Several aspects of how DLSS 5 Neural Rendering affects a scene, particularly the color tone and grading and the extra shadow detail, are common ReShade effects. One can achieve approximately 70% of the subjective graphical fidelity of DLSS 5 NR using existing ReShade shaders, with a lower performance impact.

However, the remaining 30% is where the AI-powered image processing distinguishes itself. DLSS 5 is capable of altering the image to give the appearance of drastically more detailed light simulation, including advanced subsurface scattering, realistic-looking light transmission through foliage and hair, and impressive contact shadowing and shadow falloff. These effects are difficult, if not impossible, to convincingly fake with conventional post-processing shaders. It is important to note that this is a post-processing effect; it is not actually adding these elements to the scene. There is no additional simulation going on; this is not a rendering technique like path tracing. The AI model is essentially painting in details on each frame based on its training data. This raises a theoretical risk of converging every game that uses it to a similar sort of photorealism, though with only one game officially released using it so far, and developers retaining control over model selection and effect intensity, this remains a speculative concern.

Visual Stability and Consistency

Detractors of DLSS 5 and AI technologies in general have spread concerns about artifacts and instability. After using the technology unguided and unsupervised for several hours, these myths can be dispelled. There are no telltale signs of AI generation related to DLSS 5. There is no wobbling or warping of faces or objects, no corruption of text or design elements like the UI and HUD, and no strange artifacts such as 'holes' in the rendering or unexplainable lights. In NBA 2K27, the technology has been essentially flawless in terms of quality.

This extends to character consistency. When looking at a player like Cade Cunningham, the model is notably consistent from frame to frame, across camera angles, and even under rapid camera motion. NVIDIA claims that the model is deterministic, a claim that seems to hold up in the real world with this game. In stills, close-ups, and zoom shots, DLSS 5 can look incredible. Some of the moment-to-moment scenes in NBA 2K27, such as when the virtual camera zooms close on players or coaches between quarters, get remarkably close to photographic quality. This implementation of DLSS 5 appears to be relatively modest compared to NVIDIA's earlier demos with Resident Evil 9 or examples of people hacking the leaked version into other games.

However, in normal gameplay in NBA 2K27, the effect can become subtle. Because the camera is generally quite zoomed out during gameplay, the subtle details that DLSS 5 NR adds to characters are not always clearly visible. The most notable changes are to the tonemapping and differences in shadows, particularly contact shadows. During testing, using a hotkey to toggle the effect in real time, it was sometimes difficult to tell if it was actually on or off based on image quality alone. It was often easier to determine the state by observing the frame rate and motion smoothness.

Performance Impact and Hardware Requirements

Toggling DLSS 5 on and off has an immediately visible effect on the game's performance, even when using Dynamic Multi-Frame Generation (MFG) to smooth things out. Rendered frames and generated frames look similar but are not the same to the game engine. The actual performance effect of toggling DLSS 5 is stark. Most performance data to this point for DLSS 5 has been recorded using the GeForce RTX 5090, which has enormous tensor throughput. For this testing, two different systems were used: a test bench with a Ryzen 9 9900X processor and a GeForce RTX 5070 Ti graphics card, and the Alienware 18 Area-51 with a GeForce RTX 5090 Laptop GPU.

Frame rate results were generated by carefully verifying game settings, restarting the game, and capturing an entire quarter of an NBA game with CapFrameX. The chart includes both game frame rates and output frame rates as multiplied by DLSS Dynamic Frame Generation. On the GeForce RTX 5070 Ti system, which has a 4K display refreshing at 160 Hz with G-SYNC enabled, the Dynamic MFG algorithm attempted to get as close as possible to the display refresh rate without exceeding it. Meanwhile, the laptop did not have G-SYNC enabled, and with vertical sync off, MFG was free to crank the frame rate as high as it would go.

The Alienware laptop with the mobile version of the GeForce RTX 5090 cannot effectively use DLSS 5 Neural Rendering at its native QHD+ resolution. Lowering DLSS to "Performance" does almost nothing for the frame rate. The 175W mobile GPU is bottlenecked by its power limit, and the performance limiter is the DLSS 5 Neural Rendering processing itself. Even setting DLSS to 'Ultra Performance' did not help more than a few FPS. The final output resolution is what affects DLSS 5 Neural Rendering performance. While it looks smooth with Dynamic MFG cranking the frame rate to 284 FPS, it does not always feel great in actual gameplay because it was often running below 60 FPS. The game engine has no knowledge of the MFG frame rate.

The more powerful GeForce RTX 5070 Ti desktop card, with its 300W power limit, struggles in 4K UHD with DLSS 5 Neural Rendering enabled. With a 1% low frame rate under 30 FPS, the Dynamic MFG has to crank up all the way to 6x at times, producing brief but noticeable artifacts, particularly when UI elements are sliding around the screen. With DLSS 5 NR off, the game runs beautifully in 4K "Quality" on the GeForce RTX 5070 Ti.

NVIDIA recommends a maximum resolution of 2560×1440 for this GPU. The RTX 5070 Ti has 70 shader modules, versus 82 for the RTX 5090 Laptop, and it does not quite have the tensor throughput to handle Neural Rendering at 4K. At 2560×1440 with DLSS set to 'Quality', the performance level is much more reasonable, with an average game framerate of 68.9 FPS and a 1% low frame rate of 52.3 FPS. Dynamic MFG takes this up to 152.3 FPS. This is a perfectly acceptable way to play the game. However, when Neural Rendering is turned off, Dynamic MFG disables itself because the 1% Low FPS of the game skyrockets to 168.2 FPS, with an average of nearly 240 FPS, which appears to be a cap built into the game.

DLSS 5 NR has a significant performance cost. While high frame rates can be reached through frame generation, NVIDIA recommends getting to at least 50-60 FPS before using MFG. Users are likely to have to lower their resolution from what they are accustomed to if they want to make use of this tech in the latest, most demanding games.

The Trade-Off: Image Quality vs. Resolution

"Image quality" in 3D rendering can refer to the clarity of the rendered image, expressed in terms of spatial resolution, or the quality of the rendered image in terms of visual fidelity, enhanced by improved lighting detail, higher-quality textures, and advanced rendering features. DLSS 5 NR can absolutely ramp up the image quality of a scene and make the output look more photoreal. However, it may also detract from the image quality by forcing users to lower the resolution. Playing in 2560×1440 with DLSS upscaling on a 4K monitor resulted in some imprecise, fuzzy geometry edges and visible geometry aliasing along certain surfaces.

Higher resolution helps resolve fine details at a distance. In 4K, player names on jerseys are readable even when players are standing halfway downcourt, and there is a relative lack of aliasing on the lines on the court. To attain this higher render resolution, however, one must give up the lush details of DLSS 5 Neural Rendering. The irony of this necessary decrease to render resolution is that a major part of DLSS 5 NR is the incredible level of detail it adds to surfaces, including skin, fabrics, natural materials, and metals. Without sufficient render resolution, these details cannot always be made out unless they are very close to the camera. In a game like NBA 2K27, where the camera is often far from the relevant objects, the value proposition of DLSS 5 NR is diminished.

Computational Intensity and Power Consumption

DLSS 5 NR is computationally intense. It puts a heavy workload on the GPU by maxing out the chip's tensor cores, which are normally nearly idle in games even if other DLSS features are used. With DLSS 5 NR running on a ray-traced game, every part of the GPU die (Raster, CUDA, RT, Tensors) aside from the video block is fully utilized. This generates a power load similar to running FurMark. While testing NBA 2K27 with DLSS 5 NR enabled, the GeForce RTX 5070 Ti was at its 300W power limit for the entire duration of the testing.

On the Alienware laptop, temperatures remained in the low 70°C range, but the GPU was slammed at its 175W power limit for most of the test and managed brief excursions up to 240W when it could "borrow" TDP from the CPU. The GeForce RTX 5090 Laptop is actually a bigger GPU than the RTX 5070 Ti desktop chip, so the laptop power limit is more stringent, limiting performance. This means that when using DLSS 5 Neural Rendering, the GPU works harder, draws more power, creates more heat, and generates more fan noise. This is a consideration not present with previous DLSS technologies, which mostly decreased the workload on the GPU rather than increasing it.

Conclusion: A Work in Progress

The fundamental aspect of DLSS 5 Neural Rendering that gives many people pause is that it is not like other previous NVIDIA DLSS technologies. Other DLSS tech generally works to make the game image better than the renderer could afford to produce, attempting to be a smarter way to render the same scene while working toward the ground truth of the game's own output. DLSS 5 NR is different. Even NVIDIA's Edward Lee acknowledged that DLSS 5 NR isn't working from any ground truth. Instead, the model is applying its own knowledge from its training to the game's output. With previous DLSS technologies, the game remains authoritative, but with DLSS 5, the model becomes authoritative to an extent, depending on what the developer configures.

DLSS 5 NR is a clever and impressive technology. The original marketing message may not have landed properly, and there is misinformation regarding what it is and its capabilities. It is a technology that, at least for now, requires trade-offs. Some people will dismiss Neural Rendering out of hand, while others will look down on games that do not implement it as being outdated. For personal preferences, the performance penalty is a tough sell currently. However, many people are likely to be impressed by the photorealistic details it adds to game graphics. If one prefers the nearly photographic visuals that DLSS 5 NR can enable over the higher frame rates and resolutions achievable without it, that is a valid choice. DLSS 5 Neural Rendering is extremely cool technology and is obviously just the first step in what may well be a new rendering paradigm. It is likely that DLSS 5 may be a sneak peek at the capabilities of NVIDIA's next-generation graphics hardware, which is likely to ramp up tensor performance significantly. NVIDIA has informed that the feature is coming to Ada Lovelace GPUs "once RTX 50 Series performance is more fully tuned." In its current form, the recommendation is conditional, depending on hardware, sensitivity to rendering resolution, and personal enjoyment of near-photorealistic graphics.