AI models now clean up the noisy lighting left behind after a game engine has already cast its rays, producing a final image with far fewer samples than brute-force rendering would need. Introduced with NVIDIA DLSS 3.5, it replaces the hand-tuned denoisers that traditional ray tracing depends on. A trained neural network studies partial lighting samples and rebuilds the missing detail, letting developers get cleaner reflections, global illumination, and shadows from far fewer cast rays.
Beyond that quick definition, this article breaks down how ray reconstruction actually works inside a real-time renderer, where it sits between ray tracing and path tracing, and which engines and games already ship with it.
The Short Version: Ray Reconstruction as a Smarter Denoiser
Real-time lighting in modern games starts with the engine casting rays into a 3D scene to simulate how photons bounce, reflect, and cast shadows. The result arrives noisy and incomplete because the GPU only has a few milliseconds per frame to work with. A denoiser then cleans up that mess before it reaches your monitor.
Older denoisers used fixed rules written by programmers, things like temporal accumulation (blending frames together) and spatial filtering (smoothing nearby pixels). Ray reconstruction swaps those rules for a neural network trained on high-quality offline renders. The network reads the noisy data, recognizes patterns from its training, and reconstructs detail that was never physically calculated.
Why a Denoiser Is Even Necessary
Tracing every light ray to a clean, noise-free result is wildly expensive. A single path-traced pixel in a modern game can require hundreds or thousands of ray bounces to converge, and at 60 frames per second the math simply does not fit in a consumer GPU’s budget. Denoising is the shortcut: cast fewer rays, accept a grainy image, then clean it up with smart algorithms before display.
What Makes the Neural Approach Different
Hand-tuned denoisers struggle when lighting gets complex. They blur motion, smear edges, and let “fireflies” (those bright flickering specks) linger in reflections. A trained network evaluates lighting, motion vectors, and color data together to fill in gaps that rule-based code would either smooth away or leave noisy. In practice, this means sharper reflections, smoother global illumination, and fewer visual artifacts from the same ray budget.
Where Ray Reconstruction Fits Between Ray Tracing and Path Tracing
These three terms get tangled constantly, so the table below sorts them out by what each one actually does at runtime.
| Technique | What It Casts Rays For | Typical Cost | Visual Result |
|---|---|---|---|
| Ray Tracing | Specific effects: shadows, reflections, ambient occlusion | Moderate | Selective realism layered over rasterized lighting |
| Path Tracing | Nearly all lighting interactions, including indirect bounce light | Very high | Full physically based lighting, but extremely demanding |
| Ray Reconstruction | The cleanup pass for whichever rays were already cast | Low (runs on Tensor Cores) | Cleaner output from a smaller ray budget |
Ray Tracing as a Selective Effect
Most shipping games that advertise “ray tracing” use a hybrid approach. The engine rasterizes the scene normally, then casts a limited number of rays for specific effects: a reflection probe here, a shadow trace there. Traditional denoisers smooth these out, and the result looks better than pure rasterization, but it is still mostly faked lighting with selective realism.
Path Tracing as the Heavy Hitter
Path tracing goes further. Instead of casting rays for a handful of effects, it casts rays for nearly all lighting interactions, including the indirect bounce light that fills a room when sunlight hits a red wall. The result is the most physically accurate lighting possible, but it is also brutally expensive. Without aggressive denoising, path-traced scenes look like static on a bad TV channel.
Ray Reconstruction as the Bridge
Sitting between raw ray casting and the final image, this step turns sparse, noisy samples into a polished frame. It takes the noisy output from a heavily path-traced scene and rebuilds a clean frame using a trained model. That quality-and-performance bridge is exactly what makes aggressive lighting modes viable on consumer hardware today.
That bridge is not magic, though; it depends on a neural network doing the actual image work, and the next section unpacks how.
The Neural Network Step That Changes the Image
Raw ray-traced frames arrive at the denoiser noisy, incomplete, and full of missing samples. The engine simply did not have time to compute every bounce, so pixels end up as statistical guesses rather than finished colors. Traditional denoisers try to fill those gaps with rules, but rules fail when the lighting is dynamic and the motion is fast.
Why Traditional Denoisers Hit a Wall
Temporal accumulation blends the current frame with previous ones to hide noise, but fast-moving objects leave ghost trails. Spatial filters smooth nearby pixels, but they also smear the sharp edges of reflections on a wet floor. Fireflies (the bright single-pixel flickers) linger because the rules do not know which bright specks are real and which are statistical noise.
How the Trained Model Handles It
The neural network inside ray reconstruction was trained on thousands of high-quality, fully path-traced reference frames. When it sees a noisy input, it matches the pattern to what a “correct” version would look like and outputs the cleaned image directly. Lighting, motion vectors, and color data all feed into the decision at once, so the result handles motion, edges, and flickering better than hand-written heuristics ever could.
Think of it as autocomplete for lighting: the network guesses the missing detail based on patterns it has seen before, then refines the guess with the partial data the engine actually provided.
Hardware, Engines, and the Games That Ship It
it is not a free toggle you can flip in any game. It requires specific hardware, deliberate engine integration, and a supported title.
GPU Requirements
The feature runs on Tensor Cores, the dedicated AI acceleration hardware inside NVIDIA’s RTX lineup. GeForce RTX 20-series, 30-series, and 40-series cards all support it, including the laptop variants. AMD Radeon and Intel Arc GPUs do not currently support it, because it depends on NVIDIA’s DLSS SDK and the Tensor Core execution path.
Engine Integration Is Required
Developers must add it to a title during development. It cannot be retrofitted through a driver update, a mod, or a community patch. When you see “DLSS it” in a graphics menu, that label means the studio wired the neural denoiser into the engine’s lighting pipeline.
Supported Games and How Each Uses It
it shows up in a growing list of path-traced and heavily ray-traced titles. A few standouts:
- Cyberpunk 2077: Overdrive Mode uses path tracing for almost all lighting, with it cleaning up the noise.
- Alan Wake 2: Path-traced reflections and indirect light rely on it for stable frames.
- Portal with RTX: Fully path-traced scenes use the neural denoiser to stay playable.
- Diablo IV: Ray-traced reflections and shadows use it for cleaner results at higher frame rates.
Consoles Are Out for Now
PlayStation 5 and Xbox Series X|S lack the dedicated AI hardware and SDK path that NVIDIA provides. The consoles have capable GPUs, but it is a PC-only NVIDIA feature in its current form.
The Performance and Image Quality Tradeoff in Practice
it can reduce the number of rays a game needs per pixel, easing GPU load and potentially raising frame rates in path-traced scenes. In demanding titles like Cyberpunk 2077 Overdrive, enabling the feature is often the difference between a slideshow and a playable experience.
Where the Visual Gains Show Up
Side-by-side comparisons reveal the difference clearly. Reflections on wet streets stay stable instead of flickering. Indirect light in dark interiors fills corners smoothly rather than appearing as grainy noise. Shadow gradients soften and darken gradually instead of stair-stepping across surfaces.
Artifacts to Watch For
No denoiser is perfect. Fast-moving objects can leave ghost trails because the network blends new data with old. When new geometry suddenly enters the frame, the model has no prior information and can smear the reveal. In highly chaotic scenes like explosions or particle-heavy effects, rare temporal instability still occurs.
Heads up: a faint smudge trailing a character sprinting across the screen means the neural denoiser is catching up, not that the game itself is broken.
When to Turn Ray Reconstruction On, Off, or Leave It Alone
Recommended settings depend on what you’re playing and what hardware you have.
Keep It On for Story-Driven and Path-Traced Games
Slow-paced, visually heavy titles benefit the most. Single-player adventures, horror games, and cinematic experiences lean on stable lighting and clean reflections, exactly what the neural denoiser delivers. If your card handles the path-traced mode at all, it is almost always the right choice there.
Consider Turning It Off for Competitive Play
Every frame of latency and any ghosting artifact matters in fast multiplayer. If you’re playing a competitive shooter and your GPU is already struggling, the safest path is to drop to a lower ray-tracing preset or disable the feature entirely. Native resolution plus traditional denoising can feel more responsive, even if it looks less impressive.
Practical Checklist for Enabling the Feature
- Confirm an RTX GPU: RTX 20-series, 30-series, or 40-series from NVIDIA.
- Update GeForce drivers: it landed in drivers supporting DLSS 3.5.
- Open the graphics menu: Look for a “DLSS it” toggle, usually under Ray Tracing or DLSS settings.
- Pair it with DLSS Super Resolution: Running path-traced lighting at native resolution is rarely practical; upscaling keeps frame rates usable.
- Test in motion: Walk through a reflective hallway or a dark interior and confirm the result looks clean before committing.
Final Thoughts
it is the quiet workhorse behind the most demanding lighting modes in modern games. It does not make ray tracing possible, that hardware has been here for years, but it makes path-traced lighting look finished on a consumer GPU. If your rig is RTX-equipped and your favorite titles support it, leaving the feature on is almost always the right call. The neural denoiser delivers cleaner images at playable frame rates, and the occasional ghosting artifact is a small price for what you gain.
FAQ
What does ray reconstruction do in games?
it is an AI-based denoiser that cleans up noisy lighting in real time. It runs after the engine casts rays for reflections, shadows, or full path-traced scenes, using a trained neural network to reconstruct detail from partial samples and produce a stable, finished image.
How is ray reconstruction different from ray tracing?
Ray tracing is the act of casting rays to simulate light, while the follow-up step uses trained models to fill in the missing detail. it is the cleanup step that runs after the rays are cast. One creates the raw lighting data; the other turns noisy data into a clean final image using a neural network rather than hand-written rules.
Does ray reconstruction improve FPS?
Yes, often significantly. By letting developers cast fewer rays per pixel and still produce a clean image, it can raise frame rates in heavily path-traced scenes, in some cases turning an unplayable mode into a smooth one.
What games support DLSS ray reconstruction?
Supported titles include Cyberpunk 2077 (Overdrive Mode), Alan Wake 2, Portal with RTX, and Diablo IV, along with a growing list of other path-traced and heavily ray-traced releases. The feature must be integrated by the developer, so not every DLSS-compatible game has it.
Is ray reconstruction worth turning on?
For most single-player and visually driven games, yes. The image quality gains and performance savings usually outweigh the rare ghosting artifacts. For competitive multiplayer where every frame matters, you may prefer a lower ray-tracing preset with the feature off.
Do consoles support ray reconstruction?
No. it requires NVIDIA Tensor Cores and the DLSS SDK, both of which are PC-only. PlayStation 5 and Xbox Series X|S can use traditional ray tracing and other denoisers, but not the neural path-traced cleanup that defines it.
