NVIDIA Shows DLSS 5 Progress and Technical Details at SIGGRAPH 2026 - TechPowerUp

What the report says

TechPowerUp reports that NVIDIA used SIGGRAPH 2026 in Los Angeles to present further progress on DLSS 5, its neural-rendering technology for games. The site says the company is continuing work on the system after an earlier demonstration drew criticism from some gamers, particularly over concerns that AI-driven changes could interfere with a game’s intended look.

According to the report, NVIDIA described three selectable DLSS 5 models — identified as Model A, Model B and Model C — with different levels of structural intensity, global illumination, texture detail and computational demands. Developers would be able to apply these models differently by scene, character or object, rather than using one uniform setting across an entire game. TechPowerUp says NVIDIA emphasized controls meant to protect artistic intent, including customization tied to lighting, motion, object meaning, character poses and overall art direction.

The article also highlights two technical challenges NVIDIA discussed: temporal streaming and performance. Unlike offline AI video tools that generate frames in batches, real-time games require frame-by-frame output with minimal delay. TechPowerUp says NVIDIA is using motion vectors from game engines so DLSS 5 can attach generated texture detail to the correct moving elements and avoid artifacts such as shimmer or corruption.

For performance, the report says NVIDIA is targeting 4K rendering at more than 60 frames per second within a sub-16 millisecond frame budget, with most time still spent by the game engine rather than the DLSS model. System requirements, including VRAM needs, remain unclear. TechPowerUp says more details are expected before a fall launch.