⬢github JavaScript · 579 ★ +52 since we first saw it · pushed 9 d ago · MIT
maanHimself/OpenDLSS-NR
A Vulkan reimplementation of NVIDIA's DLSS 5 Neural Rendering network, bit-exact against the original.
OpenDLSS-NR is a Vulkan reimplementation of NVIDIA's DLSS 5 neural rendering network, matching the original bit-for-bit, including all 71 transformer block intermediates. It runs FP8 GEMMs on tensor cores, includes a browser WebGPU port, a benchmark/profile/parity CLI tool, and a double-clickable demo. Users supply their own model weights; it renders, not upscales.
Why now: It hit Hacker News as a bit-exact, open reimplementation of NVIDIA's just-published DLSS 5 generative neural rendering network — a rare reverse-engineering feat with verified parity against the proprietary model.
Who it is for: Graphics and rendering engineers wanting to study, benchmark, or experiment with DLSS-style neural rendering outside NVIDIA's closed stack.
Stars over our 9 snapshots: 527 to 579, since 2 h ago.
Where people talked about it
API: https://socialmediatrends-api.osmike.com/v1/repos/maanHimself/OpenDLSS-NR