NVIDIAβs DLSS 5 upscaling technology has been in the news a lot lately, although not quite for the reason one might think: modders have been hard at work in getting the technology to (unofficially) work on GPU generations older than the Blackwell-powered RTX 50 Series, with one getting it to run on two RTX 5060 Ti cards at the same time.
Recently, a modder who goes by the handle of MAAN on X posted a working demo of DLSS 5 running directly in a web browser, with a working source code for the demo soon to be published on GitHub.
We cannot emphasise enough that this is an unofficial mod; NVIDIA currently only offers DLSS 5 through its NGX interface and Streamline DSK, the latter basically being an in-between for apps and graphics APIs like DirectX and Vulkan. It is through this devkit that developers will then provide other tools, such as motion vectors, depth data, and the rendered image to DLSS.
MAANβs DLSS 5 demo has clearly circumvented all the loopholes, but based on our brief interaction with it, and it even works on a browser in MacOS. However, it is clear that the upscaler wasnβt designed to work with the neural renderer efficiently. Firing it up on this writerβs laptop, which is equipped with an RTX 5070 Ti, the DLSS 5 Neural Rendering demo took a handful of seconds for it to fully render.
Circling back to the subject of MAAN posting a working source code for the demo on GitHub, the modder also didnβt share any technical details, so it isnβt clear how they managed to bypass the GPU brandβs standard DLSS binaries β as a native bridge uses? Did they convert browser-rendered resources for processing outside the browser? All questions that could use an explanation.
In the mean time, you can try it out yourself via the link we’ve provided in the source.
(Source: DLSS5 WebGPU Demo, Videocardz, Techspot)