NVIDIA is taking another major step toward AI-powered graphics with the launch of DLSS 5, a new generation of its Deep Learning Super Sampling technology that moves beyond traditional image upscaling and frame generation. The company says the technology introduces a new approach called 3D-Guided Neural Rendering, designed to use artificial intelligence to dramatically improve lighting, materials and visual realism in games.
NVIDIA announced Tuesday that DLSS 5 will officially launch on September 3 at 9 p.m. Pacific Time, with NBA 2K27 becoming the first game to support the technology. DLSS 5 will initially be available across NVIDIA’s GeForce RTX 50 Series desktop and laptop GPUs, as well as through GeForce NOW.
Unlike earlier versions of DLSS, which primarily focused on reconstructing images, generating additional frames and improving ray-traced graphics, DLSS 5 is designed to become an additional stage of the rendering pipeline. NVIDIA says the technology uses information generated by the game engine—including the rendered image, motion vectors, surface information and other scene data—to produce more realistic lighting and material responses.
The result is intended to make traditionally difficult visual effects appear more lifelike without requiring developers to render every detail through conventional graphics techniques. NVIDIA highlights effects such as light passing through hair and foliage, realistic subsurface scattering in human skin, deeper contact shadows and more natural interactions between light and materials.
The introduction of DLSS in 2018 was initially focused on using AI to reconstruct higher-resolution images from lower-resolution renders. NVIDIA subsequently expanded the technology with features such as Frame Generation, Multi-Frame Generation and Ray Reconstruction.
DLSS 5 represents a significant change in that philosophy. NVIDIA says the technology can generate visual details that traditional real-time rendering often has to simplify or eliminate because of performance limitations.
Rather than replacing the game’s underlying graphics, however, NVIDIA says DLSS 5 works from the frame created by the game engine. The original geometry, textures and other artist-created elements provide the foundation, while the neural-rendering model adds visual detail on top of that foundation.
NVIDIA is also emphasizing consistency. Unlike conventional generative AI models that can produce different results from the same prompt, DLSS 5 is designed to be deterministic and temporally stable. The company says it uses motion vectors and other information from the game engine to prevent visual elements from randomly changing between frames.
NBA 2K27 will provide the first public demonstration of DLSS 5 when the technology launches September 3.
NVIDIA says Visual Concepts has used the technology to enhance the appearance of NBA and WNBA players, spectators and environments. Improvements include more realistic skin lighting, facial detail, hair illumination and contact shadows.
According to NVIDIA, the RTX 5090 can reach up to 370 frames per second at 4K with the Ultra preset, ray tracing and the full DLSS feature suite enabled in NBA 2K27. At 1440p, NVIDIA claims up to 590 FPS on the RTX 5090, 410 FPS on the RTX 5080 and 350 FPS on the RTX 5070 Ti.
These figures, however, combine DLSS 5 with other DLSS technologies, including Super Resolution and Multi-Frame Generation. They therefore should not be interpreted as the performance gain provided by DLSS 5 alone.
The biggest question surrounding DLSS 5 may ultimately be its performance cost.
Early reports indicate that enabling the neural-rendering component can impose a substantial GPU workload, with some testing pointing to performance reductions of roughly 50% to 60% depending on the configuration.
That represents a significant departure from the way consumers traditionally think about DLSS. Earlier versions were primarily associated with increasing frame rates by allowing games to render fewer pixels and reconstruct the rest. DLSS 5 instead uses some of that additional GPU capability to improve the quality of the final image.
NVIDIA has nevertheless made significant progress since first demonstrating the technology. The company says the original DLSS 5 implementation required two RTX 5090 GPUs, while the current version can run on a single RTX 50 Series GPU. NVIDIA says those optimizations have produced a fivefold performance improvement over the original implementation in approximately six months. The company also says additional model and performance improvements are planned for the fall.
The technology has also generated controversy.
Early demonstrations of DLSS 5 raised concerns that AI could alter the appearance of characters rather than simply improve their rendering. Some demonstrations produced noticeably different facial features and appearances, prompting criticism that neural rendering could interfere with the artistic decisions made by game developers.
NVIDIA is now stressing that DLSS 5 gives developers significant control over the technology. The company says developers can use granular controls and semantic masking to determine what parts of an image the neural model is allowed to modify while preserving important elements of the original scene.
The company also describes DLSS 5 as deterministic, meaning identical inputs should produce consistent results rather than the unpredictable outputs associated with many generative AI systems.
The significance of DLSS 5 extends beyond simply making individual games look better.
For decades, graphics developers have had to balance visual quality against the enormous computational cost of rendering increasingly complex lighting, materials and geometry. NVIDIA’s approach suggests that neural networks could eventually take over some of the work traditionally performed entirely by the GPU’s rasterization and ray-tracing hardware.
DLSS 5 does not eliminate traditional rendering. Instead, NVIDIA describes it as a combination of conventional graphics and neural graphics, with AI becoming another stage in the rendering process. The technology can operate alongside rasterization, ray tracing, path tracing, Super Resolution, Multi-Frame Generation and Ray Reconstruction.
That could represent a broader shift in the gaming industry. Rather than simply rendering more pixels with increasingly powerful GPUs, future graphics architectures may increasingly focus on rendering a smaller amount of fundamental scene information and allowing specialized AI hardware to determine how that information should ultimately appear on screen.
For NVIDIA, that strategy also reinforces the importance of its Tensor Cores and AI-focused GPU architecture. DLSS 5 is designed specifically for RTX 50 Series hardware and uses those Tensor Cores to execute its neural-rendering model in real time.
The technology remains in its early stages, with NBA 2K27 serving as its first major test and broader developer adoption still to be determined. But if NVIDIA can continue reducing the computational cost while maintaining image quality, DLSS 5 could mark a significant transition from AI-assisted rendering to AI-driven rendering—and potentially reshape how video games are built and displayed in the years ahead.
