Nvidia DLSS 5 Arrives September 3 on RTX 50 GPUs
Nvidia's DLSS 5 launches September 3 on RTX 50 GPUs, adding neural rendering for lighting and materials. Here is what the new hardware-dependent technology actually changes.
On this page
Nvidia will release DLSS 5 on September 3, bringing its new 3D-Guided Neural Rendering system to NBA 2K27 on GeForce RTX 50-series graphics cards and laptops. The change matters because DLSS is moving beyond simply reconstructing or generating frames: Nvidia is now using a dedicated neural model to alter lighting and material detail in the final rendered image.
That makes DLSS 5 a hardware story as much as a graphics-software update. The technology is designed around the Tensor Cores in RTX 50-series GPUs, while Nvidia says it has reduced the processing cost enough to run on a single GPU rather than the dual-RTX 5090 setup used during its original development.
DLSS 5 changes what the GPU does after rendering
Earlier versions of Deep Learning Super Sampling, or DLSS, primarily used artificial intelligence to reconstruct a higher-resolution image, generate additional frames, or improve ray-traced lighting. DLSS 5 adds another stage to that pipeline. It takes the game's rendered frame and engine data such as color, motion vectors, surface information and lighting information, then uses a neural model to enhance lighting and material behavior.
The distinction is important. Nvidia is not describing DLSS 5 as a replacement for the game engine's geometry or textures. Instead, the original frame remains the foundation, while the neural renderer adds effects such as more convincing light passing through hair, deeper contact shadows and subsurface scattering, the way light penetrates and spreads through materials such as skin. That approach is intended to keep the result tied to the developer's original scene rather than allowing an AI model to freely redraw it.
RTX 50 hardware is central to the launch
DLSS 5 is launching across the GeForce RTX 50 series, including desktop graphics cards and laptops. Nvidia says the system uses the Tensor Cores in these GPUs to run its specialized neural network locally, with support for resolutions up to 4K. For owners of older RTX generations, that means the official launch is not simply a driver update that makes the same feature available everywhere.
Unofficial experiments have already shown why the hardware distinction matters. Modders have managed to adapt leaked DLSS 5 components to some RTX 40-series and even older RTX hardware, but independent testing has reported substantial performance penalties in those unsupported configurations. Those experiments are useful for showing what might be technically possible, but they should not be treated as official compatibility or representative retail performance.
Nvidia claims a fivefold performance improvement
The most striking hardware claim is not a higher clock speed or a new graphics architecture. Nvidia says the DLSS 5 neural-rendering pipeline has become five times faster over the six months since its March announcement. The company says the technology originally required two RTX 5090 GPUs during development and can now run on a single RTX 50-series GPU.
That figure is an Nvidia claim rather than an independent benchmark, so it needs context. It describes the improvement in the DLSS 5 system itself during development, not a claim that an RTX 50 graphics card is suddenly five times faster in ordinary games. Independent measurements of the final release will be more useful for determining how much of that optimization translates into frame time, latency and image quality for players.
The first real test arrives with NBA 2K27
DLSS 5 will debut in NBA 2K27 at 9 p.m. Pacific Time on September 3. Nvidia says the feature will be available on RTX 50-series PCs and laptops, while GeForce NOW Ultimate users can access it through cloud systems powered by RTX 5080 GPUs. Players will also need the new GeForce Game Ready driver released at the same time.
Nvidia's own numbers show how aggressively the company is combining DLSS technologies. At 4K with the Ultra preset and ray tracing enabled, it reports up to 370 frames per second on an RTX 5090 when DLSS 5 is combined with Super Resolution and Multi Frame Generation. At 1440p, Nvidia lists up to 590 frames per second on the RTX 5090, 410 on the RTX 5080, 350 on the RTX 5070 Ti and 260 on the RTX 5070. These are vendor-provided results, so they should not be read as independent performance benchmarks.
The bigger point is that those frame-rate figures combine several technologies. DLSS 5 itself is responsible for neural rendering and visual enhancement, while other DLSS components handle image reconstruction and frame generation. A high output frame rate therefore does not mean the GPU is rendering hundreds of complete conventional frames every second.
Developers get more control than the early demos suggested
One concern surrounding DLSS 5 has been whether neural rendering could change the appearance of characters or scenes in ways artists did not intend. Nvidia's final implementation gives developers several controls intended to address that problem. Studios can choose between different neural models, adjust structure and tone intensity, and use semantic masks to control which objects receive the enhancement.
That control is particularly relevant for character rendering. A developer could allow stronger neural enhancement for foliage, materials or environmental lighting while limiting the effect around a character's face. Nvidia says NBA 2K27 uses the technology to improve skin lighting, hair detail and contact shadows while preserving the facial geometry captured from real athletes.
Better input data still means better output
DLSS 5 does not remove the importance of the underlying rendering pipeline. Nvidia explicitly says the neural result depends on the quality of the information supplied by the game. Traditional rasterization, ray tracing and path tracing can all provide the foundation, but richer lighting data gives the neural renderer more information to work with.


