[KongchangAI]
ProductOpenDLSS

OpenDLSS NR

An open-source Vulkan reimplementation of the DLSS 5 neural rendering network, released under the MIT license. It supports native execution on RTX 40-series GPUs and cross-platform execution in web browsers via WebGPU.

Timeline (last 90 days)

Oct 4

Developer MAAN rewrote DLSS 5's neural rendering network in Vulkan and open-sourced it under the project name OpenDLSS NR, using the MIT license.

Unverified50%
Oct 4

OpenDLSS NR's neural network contains 71 blocks, uses FP8 activations, and has total weights of approximately 141 MB.

Unverified50%
Oct 4

OpenDLSS NR achieves bit-exact intermediate output consistency with the original at 75 block boundaries.

Unverified50%
Oct 4

On the RTX 4070 SUPER, OpenDLSS NR takes 7.8 milliseconds per frame at 1080P, 12.6 milliseconds at 2K, and 29.3 milliseconds at 4K.

Unverified50%
Oct 4

The above performance figures are from the developer's own testing and are not independent third-party benchmarks.

Unverified50%
Oct 4

OpenDLSS NR provides a browser version that runs using WebGPU and does not depend on TensorRT.

Unverified50%
Oct 4

The browser version is approximately 27 times slower than the native version, but theoretically can run on AMD and Intel graphics cards.

Unverified50%
Oct 4

OpenDLSS NR's network weights must be extracted from NVIDIA's closed-source runtime by the user, which is the highest-barrier step in the entire process.

Unverified50%
Oct 4

OpenDLSS NR currently has no game integration whatsoever; incorporating it into games would require the community to create mods.

Unverified50%
Oct 4

The native version of OpenDLSS NR still requires specific hardware support and cannot run on AMD graphics cards.

Unverified50%

All Facts (10)

Source Articles