A developer known as MAAN has created a browser-based implementation of Nvidia's DLSS 5 technology using WebGPU compute shaders. The project allows users to run neural rendering models directly in a web browser, bypassing the requirement for specific Nvidia RTX hardware or proprietary drivers.
The implementation consists of model weights totaling 147MB and a compressed JavaScript runtime of approximately 1MB. The demo supports macOS and non-Nvidia GPUs, making the technology accessible for 3D model previews and architectural visualization tasks where real-time performance is not critical.
During testing on an RTX 40-series desktop, the viewer demonstrated smooth rotation, though individual renders required one to two seconds to complete. The developer plans to release the project's source code on GitHub, enabling users to experiment with their own 3D models via the web interface.
Nvidia officially launched DLSS 5 earlier this month for the game NBA 2K27, initially restricting support to RTX 50-series hardware. While the technology typically relies on the NGX interface or Streamline SDK, this WebGPU port demonstrates a new approach to neural rendering outside the standard ecosystem.
