# NVIDIA DLSS 5

> Source: https://aiwiki.ai/wiki/nvidia_dlss_5
> Updated: 2026-09-03
> Fact-checked: 2026-09-03
> Categories: AI in Gaming, Generative AI, NVIDIA
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "NVIDIA DLSS 5." aiwiki.ai, 3 Sept 2026. https://aiwiki.ai/wiki/nvidia_dlss_5
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

NVIDIA DLSS 5 is an optional real-time [generative AI](https://aiwiki.ai/wiki/generative_ai) rendering feature developed by [Nvidia](https://aiwiki.ai/wiki/nvidia) for video games. Nvidia calls the underlying method "3D-Guided Neural Rendering." Instead of upscaling an image or inserting an intermediate frame, DLSS 5 changes the lighting and material appearance of the final displayed frame while using the game engine's render as its foundation.[1][2][3]

Nvidia announced DLSS 5 on March 16, 2026. The company scheduled its first public implementation for September 3, 2026 at 9:00 p.m. Pacific in NBA 2K27, together with a required GeForce Game Ready Driver. At the time this article was verified earlier on September 3, that scheduled release time had not occurred. Nvidia lists GeForce RTX 50 Series desktop and laptop GPUs as the supported local hardware at debut, along with GeForce NOW Ultimate sessions running on Nvidia-operated RTX 5080 cloud systems.[1][2]

| Attribute | Detail |
|---|---|
| Developer | NVIDIA |
| Announced | March 16, 2026 |
| Scheduled first availability | September 3, 2026 at 9:00 p.m. Pacific |
| First supported game | NBA 2K27 |
| Product role | Optional final neural-rendering stage |
| Initial local hardware | GeForce RTX 50 Series desktop and laptop GPUs |
| Cloud availability | GeForce NOW Ultimate on Nvidia-operated RTX 5080 systems |
| Developer integration | NVIDIA Streamline or an Unreal Engine 5 plugin |

## Relationship to earlier DLSS features

DLSS originally stood for Deep Learning Super Sampling and began as a resolution-upscaling system. The family later added Frame Generation, Multi Frame Generation, and Ray Reconstruction. DLSS 5's neural-rendering feature has a different purpose: it generates the displayed appearance of lighting and materials after the conventional renderer has produced a frame.[1][3][4]

Nvidia documents DLSS 5 as an independent option that can be used alongside Super Resolution, Multi Frame Generation, and Ray Reconstruction. Those components remain responsible for reconstruction or frame generation. A reported frame rate from a configuration using several DLSS components therefore cannot be attributed to DLSS 5 alone.[2]

## How DLSS 5 works

The NVIDIA Applied Deep Learning Research summary describes DLSS 5 as a one-step pixel-space [diffusion model](https://aiwiki.ai/wiki/diffusion_model) designed for an interactive frame budget. At inference time, it is conditioned on the current rendered frame, engine motion vectors, carried temporal state, and values chosen by the game's artists. During training, renderer-derived scene attributes are used as consistency supervision.[3]

This distinction between inference and training inputs matters. Nvidia's March announcement identifies frame color and motion vectors as runtime input. Its September research summary says renderer-derived attributes help ground the model during training. Public descriptions do not establish that every buffer used for training is separately passed to the model at runtime.[1][3]

The feature is causal, processing one current frame at a time rather than waiting for a future sequence. It also carries temporal state and uses motion vectors to relate one frame to the next. Nvidia describes the result as deterministic and trained for frame-to-frame stability, but independent release testing of those properties was not available before the scheduled debut.[2][3][5]

DLSS 5 runs locally as the last stage of an existing game pipeline. Nvidia says it uses the [Tensor Cores](https://aiwiki.ai/wiki/tensor_core) in GeForce RTX 50 Series GPUs and can operate at output resolutions up to 4K on one GPU. It can follow rasterized, ray-traced, or path-traced rendering. It does not replace the engine's underlying geometry, textures, lighting system, or conventional renderer. Instead, the model uses their rendered result as the source for generated effects such as subsurface scattering, light transmission through hair or foliage, material response, and contact shadows.[2][3]

These effects are inferred from learned appearance priors rather than calculated as additional conventional ray samples. Nvidia says a higher-quality source frame, such as one containing ray-traced or path-traced lighting, gives the model better information and can improve its result. Claims that the output is photorealistic, preserves structure, or eliminates temporal artifacts are Nvidia's characterizations, not independently established properties.[2][3]

## Developer controls and game integration

Nvidia exposes several controls to game developers rather than treating the model as a fixed filter. Studios can select among model variants and can use different variants in different scenes. A Structure Intensity setting adjusts high-frequency effects such as ambient occlusion, reflections, and subsurface scattering. A Tone Intensity setting changes broader lighting and color response.[2]

Two kinds of masking limit where the effect is applied. Nvidia describes semantic masking that recognizes classes of scene content and engine-level masks that developers can assign to particular objects or asset groups. A studio could, for example, apply more of the effect to scenery while reducing it on faces, or isolate foliage and glass from the rest of a frame. In the documented NBA 2K27 integration, players receive an on/off setting labeled `DLSS Neural Rendering`; the model, intensity, and masking decisions remain with the developer.[2]

Integration uses the NVIDIA Streamline framework or an Unreal Engine 5 plugin. Streamline is also used for other DLSS components, but adding DLSS 5 still requires a game-specific implementation and art-direction decisions. Nvidia's March list of intended games is not evidence that the feature can be enabled in any game without developer support.[1][2][4]

## Performance evidence

Nvidia said in September that it had made DLSS 5 five times faster over six months, moving from a March demonstration on two RTX 5090 GPUs to support across the RTX 50 Series on one GPU. The company attributed the improvement to pipeline optimization and model changes. Nvidia did not publish a reproducible benchmark for that fivefold figure, so it should be understood as a vendor-reported development comparison.[2]

The launch material advertises up to 370 displayed frames per second for NBA 2K27 on an RTX 5090 at 4K, the Ultra preset, and ray tracing. That configuration also uses Performance-mode Super Resolution and 6x Multi Frame Generation. It does not isolate the cost or benefit of the neural-rendering stage.[2][5]

TechSpot calculated from Nvidia's chart that the 370 displayed-frame result implies a render rate of roughly 62 frames per second before 6x Multi Frame Generation. It similarly derived about 39 frames per second from Nvidia's 233 displayed-frame result for an RTX 5080. The publication noted that actual render performance with frame generation disabled would differ because frame generation has its own overhead. These values are TechSpot's inferences from Nvidia's data, not measurements from an independent test.[5]

PC Gamer separately reported that Nvidia representatives estimated a 50 to 60 percent performance reduction when DLSS 5 is enabled. That estimate was relayed from Nvidia staff and was not the result of PC Gamer benchmarking a public build.[6] Nvidia had not supplied an otherwise-identical DLSS 5 on/off performance comparison when TechSpot published its September 1 analysis. Exact latency, frame-time behavior, memory use, artifact frequency, and the cost of the neural-rendering stage therefore remained unverified before launch.[5]

## Reception

Nvidia's March demonstrations prompted debate because DLSS 5 sometimes changed faces, lighting, and the perceived style of a scene rather than only reconstructing detail. The Associated Press reported widespread criticism and memes, including complaints that the examples weakened the games' existing art direction. Nvidia responded that developers would have detailed controls over the effect.[7]

The September launch material placed those controls at the center of the product description. In a controlled in-person demonstration, a PC Gamer reporter found some restrained settings convincing but said an aggressively adjusted model could produce overdone shadows, shiny skin, and facial features that looked substantially different from the source. That is a qualitative account of a vendor demonstration, not a general image-quality test.[8]

Whether a DLSS 5 result is preferable is partly an artistic judgment. Technical questions also require moving footage and repeatable measurements: a still comparison cannot show temporal instability, input latency, or frame-time consistency. Before the scheduled NBA 2K27 release, independent reviewers had not yet tested the official implementation under controlled conditions.[5][8]

## Announced game support

NBA 2K27 was the only first-day implementation named in Nvidia's September release plan. It was scheduled to expose the feature after installation of the September 3 Game Ready Driver, with an `F9` shortcut for toggling neural rendering during play and replays.[2]

In March, Nvidia also announced intended DLSS 5 support for the following games:[1]

- AION 2
- Assassin's Creed Shadows
- Black State
- CINDER CITY
- Delta Force
- Hogwarts Legacy
- Justice
- NARAKA: BLADEPOINT
- NTE: Neverness to Everness
- Phantom Blade Zero
- Resident Evil Requiem
- Sea of Remnants
- Starfield
- The Elder Scrolls IV: Oblivion Remastered
- Where Winds Meet

The March announcement did not give release dates for those integrations. Their inclusion in the list represents planned support, not confirmation that each implementation has shipped.[1]

## Publicly documented limitations

At the scheduled debut, Nvidia listed only RTX 50 Series hardware and GeForce NOW Ultimate as supported platforms. The company's launch materials did not announce official RTX 40 Series or older-GPU support.[2]

The public research summary reviewed for this article does not give a model parameter count, training-corpus inventory, precision format, per-model execution time, or a complete account of training data. It states that the model learns appearance priors from real-world visual data and is grounded through renderer-derived supervision, but those high-level descriptions are not enough to independently reproduce it.[3]

Claims based on leaked software, user-made modifications, or unofficial operation in unannounced games and hardware are not part of the documented product record and are not treated here as evidence of support.

## References

[1] NVIDIA Newsroom. "NVIDIA DLSS 5 Delivers AI-Powered Breakthrough in Visual Fidelity for Games." March 16, 2026. https://nvidianews.nvidia.com/news/nvidia-dlss-5-delivers-ai-powered-breakthrough-in-visual-fidelity-for-games

[2] Lin, Henry. "NVIDIA DLSS 5 Available September 3rd: 3D-Guided Neural Rendering Debuts in NBA 2K27 for All GeForce RTX 50 Series GPUs, and on GeForce NOW." NVIDIA GeForce, September 1, 2026. https://www.nvidia.com/en-us/geforce/news/dlss-5-3d-guided-neural-rendering/

[3] NVIDIA Applied Deep Learning Research. "DLSS 5: Generative Neural Rendering." September 1, 2026. https://research.nvidia.com/labs/adlr/DLSS5/

[4] NVIDIA Developer. "NVIDIA DLSS." Accessed September 3, 2026. https://developer.nvidia.com/rtx/dlss

[5] Schiesser, Tim. "This Is How DLSS 5 Actually Performs (According to Nvidia)." TechSpot, September 1, 2026. https://www.techspot.com/article/3169-nvidia-dlss-5-release/

[6] Edser, Andy. "DLSS 5 comes with a massive 50-60% performance hit." PC Gamer, September 1, 2026. https://www.pcgamer.com/hardware/graphics-cards/dlss-5-comes-with-a-massive-50-60-percent-performance-hit-so-i-hope-you-like-lower-frame-rates-with-your-dose-of-ai-image-enhancement/

[7] Huamani, Kaitlyn. "Nvidia's DLSS 5 is the glossy subject of memes and backlash from gamers." Associated Press, March 18, 2026. https://apnews.com/article/nvidia-dlss-5-memes-jensen-huang-c0042421aba959f1433ba8798553bf18

[8] Edser, Andy. "Nvidia showed me DLSS 5 in person." PC Gamer, September 1, 2026. https://www.pcgamer.com/hardware/graphics-cards/nvidia-snuck-me-into-a-room-to-show-off-dlss-5-in-person-and-its-somehow-beautiful-ugly-impressive-and-troubling-all-at-the-same-time/