Sunday 4 October 2026
Afternoon EditionRSS
ZOTPAPER

News without the noise

ZOTPAPER
Hardware & Devices

Modder Brings Nvidia's DLSS 5 to Web Browsers via WebGPU, Works on Non-Nvidia GPUs and macOS

Proof-of-concept demo runs on Cloudflare Workers but delivers only one render every two seconds

By LineZotpaper
Published
Read Time2 min
A modder named MAAN has reportedly gotten Nvidia's DLSS 5 neural rendering working inside a web browser by reimplementing it with WebGPU compute shaders, according to a report by VideoCardz and Tom's Hardware. The demo, hosted on Cloudflare Workers, works on macOS and runs on non-Nvidia GPUs, but each frame takes roughly two seconds to render, far slower than real-time use.

The modder, known as MAAN, extracted model weights from a leaked DLSS 5 library file and rebuilt the neural network using WebGPU compute shaders. The weights total around 147MB, with a 1MB compressed JavaScript runtime. The demo includes default scenes such as "Cowboy Gramps" and offers adjustable settings and a comparison view.

Nvidia officially launched DLSS 5 earlier this month, with NBA 2K27 as the first supported title. The technology is restricted to RTX 50-series hardware and GeForce NOW, though modders have already ported it to RTX 40-, 30-, and 20-series cards, as well as to AMD's RDNA 4 GPUs. MAAN's browser port bypasses Nvidia's driver-level restrictions entirely, using WebGPU for compute which works across operating systems and GPU vendors.

In a quick test on an RTX 40-series desktop, the 3D viewer remained smooth to rotate, but each DLSS 5 render took one to two seconds in live mode. MAAN plans to release the source code on GitHub this weekend.

The developer noted they are unsure whether the leaked library file differs from the official release. Nvidia typically provides DLSS through its NGX interface or Streamline SDK, neither of which supports WebGL or WebGPU natively.

§

Analysis

Why This Matters

  • Cross-platform accessibility: DLSS 5's AI-powered upscaling can now run on macOS and non-Nvidia GPUs, breaking Nvidia's hardware lock-in for this feature.
  • Proof of concept for browser-based neural rendering: If performance improves, in-browser gaming or cloud streaming could leverage similar techniques without vendor-specific drivers.
  • Modding community persistence: The rapid porting of DLSS 5 to multiple platforms shows that software restrictions are often temporary, pressuring Nvidia's ecosystem strategy.

Background

DLSS (Deep Learning Super Sampling) is Nvidia's AI-driven technique for upscaling and improving image quality in real time. DLSS 5, launched in September 2026, introduced new neural rendering capabilities and was initially exclusive to the RTX 50-series GeForce RTX cards. Shortly after launch, modders began porting it to older Nvidia hardware and even AMD GPUs, primarily by patching CUDA instructions. This browser-based approach is the first to bypass the need for Nvidia's proprietary runtime entirely, relying instead on the cross-platform WebGPU standard.

Key Perspectives

Modders and enthusiasts: The browser demo represents a technical achievement in reverse engineering and WebGPU programming. It offers a glimpse of vendor-neutral AI graphics, though current performance makes it unusable for actual gaming. Nvidia: The company has not commented on the port. Officially, DLSS 5 is marketed as an RTX 50-series exclusive, and the leak of library files may prompt discussions around intellectual property and software reverse engineering. Critics and skeptics: The 2-second-per-frame rendering speed is orders of magnitude slower than real-time demands. Even with future optimizations, WebGPU compute shaders lack the dedicated tensor core hardware acceleration that makes DLSS efficient on Nvidia GPUs, so practical use remains distant.

What to Watch

  • Source code release: MAAN's GitHub publication this weekend will reveal the implementation details and allow others to test or improve performance.
  • Nvidia's response: Whether the company will take technical or legal action to prevent further distribution of extracted weights or browser-based ports.
  • Performance benchmarks on other hardware: Testing on AMD RDNA 4 or Intel Arc GPUs could show whether the port scales beyond the RTX 40-series desktop used in the initial test.

Sources

Zotpaper

Written by software from the reporting listed above, scored by an automated standards desk, and published without a person reading it first. If something here is wrong, tell the editor and it will be put right.

How we workReport an errorSubscribe