NVIDIA Unveils DLSS 5: Real‑Time Photoreal AI Upscaling

NVIDIA has introduced DLSS 5, the latest version of its AI‑driven upscaling and neural rendering system, during its GTC 2026 keynote. The company says the new model will bring photoreal lighting and material detail to games in real time, aiming to push visual fidelity further than previous DLSS iterations.

In on‑stage demonstrations, CEO Jensen Huang showcased DLSS 5 running on a dual RTX 5090 setup with titles like Resident Evil: Requiem, Hogwarts Legacy and Starfield. The clips highlighted more lifelike hair strands, nuanced skin tones and richer lighting across scenes. NVIDIA explained the model consumes per‑frame color and motion vectors, then infuses pixels with lighting and material cues anchored to original 3D content to keep results consistent from frame to frame.

The company notes DLSS 5 can operate in real time and supports up to 4K resolution. During the demo, the feature was compared to versions of games with no DLSS enabled; what remains unclear is the relative gain when pitted against DLSS 4.5 with its path‑tracing and advanced feature set fully active.

Huang’s demo ran on two RTX 5090 GPUs, though NVIDIA says the technology will eventually be able to run on a single card. That said, the firm’s own demonstration suggests top‑tier hardware will be needed to match the showcased fidelity. NVIDIA also framed DLSS 5 as a major leap, calling it the largest breakthrough in graphics since real‑time ray tracing.

For players and developers, DLSS 5 could mean a new path to richer visuals without the same raw rendering cost, but key questions remain about performance tradeoffs, single‑GPU viability and how it compares directly to DLSS 4.5 in equivalent conditions. NVIDIA plans to ship DLSS 5 this fall; until then, hands‑on comparisons from third parties will better reveal its true impact.