TL;DR
Get the latest gadgets delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
OpenDLSS is a public Vulkan reimplementation of Nvidia’s DLSS 5 neural rendering network, with its developer reporting byte-for-byte agreement at all 75 tested block boundaries. The code does not include the model weights, supports only specified newer Nvidia GPUs and drivers, and is not an implementation of DLSS Super Resolution.
A GitHub project called OpenDLSS has published a Vulkan reimplementation of Nvidia’s DLSS 5 neural rendering network, with its developer reporting byte-for-byte agreement with the original at all 75 tested block boundaries. The implementation is presented as research and engineering code, not a complete Nvidia product: users must supply the model weights, and the project has specific hardware and driver requirements.
The project describes the network as a 71-block shifted-window transformer with a global vision transformer at its lowest level, arranged across six pooling levels. Its account says the model uses E4M3 FP8 activations with FP16 accumulation and has 141 MiB of weights. OpenDLSS takes a rendered frame along with noise, temporal-history data and conditioning values, then produces an RGB residual and a temporal-blend logit for each pixel.
The developer says parity covers all 75 block boundaries, not just the final rendered result. The repository provides a Vulkan route using shader kernels and a faster route using generated Nvidia PTX kernels. A separate browser WebGPU port is described as an independent implementation that matches the same captures without tensor cores or FP8; the project reports 72 milliseconds for that version at 512-by-512, compared with 2.7 milliseconds for its Vulkan implementation at that resolution.
On an RTX 4070 SUPER, OpenDLSS reports minimum per-frame times of 2.8 milliseconds at 768-by-768, 7.8 milliseconds at 1920-by-1080, 12.6 milliseconds at 2560-by-1440 and 29.3 milliseconds at 3840-by-2160. Those are project-supplied measurements from a minimum-over-40-frames test; the repository says sustained GPU load can shift clock states and make median times a few percent higher. They are not an independent benchmark or a comparison with Nvidia’s own implementation.
What OpenDLSS Makes Testable
If the reported parity holds for other users and setups, the code offers developers a way to inspect and run a DLSS 5-style neural rendering graph outside Nvidia’s proprietary software stack. Its block-level checks could help researchers study how the network’s computations fit together, while the browser port provides a separate implementation path that does not rely on tensor cores or FP8 hardware.
The practical reach is narrower than the open-source label alone might suggest. The project requires users to provide the model files, and its stated requirements include Windows, an Nvidia Ada-generation or newer GPU, and drivers exposing several specified Vulkan extensions. The source does not establish that the code is endorsed by Nvidia, nor that its reported parity has been independently verified.
Nvidia DLSS 5 neural rendering GPU
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Neural Rendering, Not Upscaling
OpenDLSS describes the network as a generative neural rendering system: it processes a frame already rendered by a game engine, generating or adjusting image detail and tone rather than increasing the input resolution. The project explicitly says it is not an upscaler and that DLSS Super Resolution is not implemented; that is a different network.
The repository includes a demo built around the Filament renderer, where its temporal path uses reprojected history and the network’s blend output. Its command-line tool runs single frames without history, matching the conditions used for the project’s reference captures. That distinction matters when interpreting results: the parity claim is tied to those captures, and demo behavior involves an additional temporal feedback loop.
“The intermediates match too, not just the final image: all 75 block boundaries, byte for byte.”
— OpenDLSS project description on GitHub
Vulkan compatible Nvidia graphics card
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Limits of the Parity Claim
The supplied project description does not identify an independent test of its bit-exactness claim, or explain who created the reference captures and how they were obtained. It also does not provide a publication date, a comparison against Nvidia’s runtime performance, or evidence that the implementation is officially authorized or supported by Nvidia. The reported frame times are developer measurements on one named GPU, not results from a broader hardware survey.
Users also need model weights, but the source material does not say where those files can be obtained or whether they are distributed by Nvidia. Compatibility beyond the listed Windows and Nvidia hardware and driver requirements is not established. OpenDLSS’s results should therefore be understood as claims about this project and its test setup, not as confirmation of general compatibility or equivalent behavior across commercial games.
As an affiliate, we earn on qualifying purchases.
Access, Testing and Compatibility
The project’s next practical step for interested developers is to build it with the documented Windows toolchain, provide a model directory and run the supplied parity or verification commands against fixtures. The repository also provides a demo path using Filament and a browser WebGPU port, giving technically equipped users ways to test different implementations.
Further confirmation would depend on independent checks of the reference comparisons, more performance results across supported GPUs, and clearer information about access to the required weights. The source material does not announce a release schedule or a response from Nvidia, so it remains unclear whether either will follow.
Nvidia RTX 4070 SUPER graphics card
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is OpenDLSS?
OpenDLSS is a GitHub project that reimplements Nvidia’s DLSS 5 neural rendering network using Vulkan. Its developer says the implementation matches reference captures byte for byte at 75 block boundaries.
Does OpenDLSS include DLSS Super Resolution?
No. The project says DLSS Super Resolution is a different network and is not implemented. OpenDLSS is described as a same-resolution neural rendering network, not an upscaler.
Are the model weights included?
No. The project says users must supply the model weights in a directory following its documented format. The supplied source does not say where to obtain them.
What hardware does it require?
The listed requirements include Windows, an Nvidia Ada-generation or newer GPU, and a driver exposing several named Vulkan extensions. The source does not establish compatibility with other operating systems or GPU vendors.
Has the bit-exactness claim been independently verified?
The project reports byte-for-byte agreement at 75 block boundaries, but the supplied material does not identify independent verification. The claim should be attributed to the project until outside testing is documented.
Source: hn
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
