HeadlinesBriefing favicon HeadlinesBriefing.com

NVIDIA DLSS 5 Technical Preview Review

TechPowerUp Reviews •
×

NVIDIA DLSS 5 is the first neural rendering technology that adds to a scene rather than reconstructing it, running locally as a generative model. Unveiled at GTC in March and refined through SIGGRAPH and Gamescom, DLSS 5 launches with NBA 2K27 on GeForce RTX 50-series GPUs, RTX 50 laptops, and GeForce NOW. Unlike previous DLSS versions focused on upscaling and frame generation, DLSS 5 adds photorealistic lighting, subsurface scattering, contact shadows, and material effects on top of a game's output—with no ground truth to measure against. This introduces ambiguity, but NVIDIA's approach protects artistic intent by training the model to enhance lighting and materials without altering geometry, silhouettes, or surface normals. Developers gain granular controls to dial effects, apply them selectively, or disable DLSS 5 entirely for specific areas. The technical presentation, led by Edward Liu and Gabriele Leone, was notably more detailed than typical launches, addressing early community concerns about visual correctness. Compatibility is limited to RTX 50-series due to reliance on Blackwell's fifth-generation Tensor cores, with no fallback for older hardware like RTX 40-series.

NVIDIA's Zorah demo demonstrates the technology's capabilities, showing substantial improvements in realism while preserving the original game's aesthetic. The model's complexity means performance costs are significant, though exact figures depend on implementation. Developer tools allow fine-tuning of the neural rendering effects, making DLSS 5 adaptable across different artistic styles and performance requirements. While some feared loss of artistic intent, the ability to constrain changes to lighting and materials rather than geometry appears to address those concerns effectively.

The technology represents a major shift in real-time graphics, moving from reconstruction-based AI to generative enhancement. This approach opens new possibilities for visual fidelity but also raises questions about verification and consistency that NVIDIA is working to resolve through developer controls and careful training methodologies.