NVIDIA officially announced DLSS 5 at the GTC event today described by CEO Jensen Huang as the GPT moment for graphics, it marks a major shift from performance focused upscaling toward real-time neural rendering to achieve photorealistic visual fidelity.

Unlike previous versions that primarily boosted frame rates, DLSS 5 focuses on enhancing the physical accuracy of the final image.

• Neural Rendering Model: Reconstructs frames with photorealistic lighting and materials in real time by analyzing color and motion vector data.

• Enhanced Materials: Infuses scenes with realistic properties for complex elements like subsurface scattering on skin, the sheen of fabric, and detailed hair interactions.

• Cinematic Lighting: Dynamically generates advanced lighting effects such as rim lighting and contact shadows that were previously limited by hardware constraints.

• Developer Control: Offers specific tools for artists to adjust intensity, color grading, and masking to ensure the AI enhancements align with the game’s intended aesthetic.

Scheduled for release in Fall 2026 Nvidia actually used two RTX 5090s for its demos: one plays the game, the other exclusively runs the DLSS 5 technology.

The use of two GPUs is required right now as DLSS 5 still has a long way to go in terms of optimisation both in terms of performance and its VRAM footprint.

DLSS 5 is designed for use on a single GPU and that’s how it will ship later this year. Quite how scalable it is also remains to be seen, but in common with other DLSS technologies, Nvidia tells us that the computational cost scales with resolution.

This begs the questions of how well this technology can scale down to older GPUs and what official support will be like if this is limited to 50/60 series with DLL requiring a swap on older cards.

Do you think in the future we could see Tensor cards for a second slot to offset AI to like PhsyX cards? As we have already seen DLSS 4.5 being too much of a cost on lower/older RTX cards.