TL;DR
OpenDLSS is a GitHub project that reimplements Nvidia’s DLSS 5 Neural Rendering network using Vulkan, and its authors report byte-for-byte matches at all 75 block boundaries against reference captures. It requires users to supply the model weights, runs on supported Nvidia Ada-or-newer GPUs, and is not an implementation of DLSS Super Resolution. The supplied performance figures and parity claims come from the project, not an independent assessment.
The OpenDLSS GitHub project has published a Vulkan reimplementation of Nvidia’s DLSS 5 Neural Rendering network, with the developers reporting byte-for-byte matches against reference captures at all 75 block boundaries. The work offers a way to examine and run the network outside Nvidia’s own implementation, but users must supply the model weights, and the parity and speed figures remain project-reported rather than independently verified.
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
The project describes the model as a 71-block shifted-window transformer and U-net, with a global vision transformer at its deepest level and six pooling stages. It reports using E4M3 FP8 activations with FP16 accumulation and lists the weights at 141 MiB. OpenDLSS says its matching checks cover intermediate outputs as well as the final image; its documentation includes tools for testing parity against fixtures.
The network takes a rendered frame, a low-dynamic-range proxy, three lanes of Gaussian noise, a reprojected previous-frame output and five conditioning values. It returns four channels per pixel: an RGB residual and a temporal-blend logit. The project calls this generative neural rendering: it modifies an image already rendered by a game engine, rather than increasing its resolution. The repository explicitly says DLSS Super Resolution is not implemented.
A separate WebGPU implementation is included for browsers. The project says it matches the same captures without tensor cores or FP8, but reports substantially slower performance: 72 milliseconds at 512-by-512, compared with 2.7 milliseconds for the Vulkan route. The Vulkan version’s listed minimum frame times on an RTX 4070 SUPER are 2.8 ms at 768-by-768, 7.8 ms at 1920-by-1080, 12.6 ms at 2560-by-1440 and 29.3 ms at 3840-by-2160, measured as minima over 40 frames, according to the repository.
What OpenDLSS Makes Testable
A public implementation could make it easier for developers and researchers to inspect how a neural rendering pipeline operates, test compatible scenes and compare implementations. The project supplies a CPU arithmetic reference, shader and PTX kernel routes, profiling tools, and a demo built around the Filament renderer. That provides more than a visual demonstration: the claimed intermediate-level parity gives users a way to check computations throughout the network.
Access remains constrained by hardware, drivers and weights. The listed requirements include Windows, an Nvidia Ada-or-newer GPU, and drivers exposing several specified Vulkan and Nvidia extensions. The network files are not presented as bundled weights; users must provide a model directory in the documented format. The project is therefore not a drop-in alternative for anyone who lacks supported hardware or access to suitable weights.
Products Worth Considering

Kindle Paperwhite Signature Edition 32GB (2024 Model) – Metallic Black

Bose QuietComfort Wireless Noise-Cancelling Headphones, Twilight Blue

Apple AirPods Pro (2nd Generation) Wireless Earbuds with MagSafe Charging Case (Lightning)
As an affiliate, we earn on qualifying purchases.
How the Network Differs from Upscaling
OpenDLSS distinguishes this network from DLSS Super Resolution, a separate Nvidia network. The implementation processes an input frame and returns output at the same resolution, using temporal information and conditioning inputs to alter the rendered image. The project says its demo implements a history-feedback path, while the command-line tool processes single frames without history, matching the setup used for its reference captures.
The repository attributes the model design to Nvidia’s report, DLSS 5: Generative Neural Rendering. OpenDLSS’s contribution is its own Vulkan implementation and a second, browser-based WebGPU port; the material supplied does not establish that Nvidia endorses or supports either project. Its performance table describes tests on one listed GPU, an RTX 4070 SUPER, and should not be read as a benchmark across hardware.
“A Vulkan reimplementation of NVIDIA’s DLSS 5 Neural Rendering network, bit-exact against the original.”
— OpenDLSS project README
Questions About Parity and Access
The supplied material does not include independent verification of the reported bit-exact results, benchmarks or model description. It also does not identify the source or availability of the weights users need, beyond documenting the expected model-directory format. The reference captures and Nvidia materials are mentioned, but the source does not provide enough detail to assess how broadly the tests cover different scenes or operating conditions.
Performance figures are limited to one GPU and reported minimum times over 40 frames; the README notes that GPU clocks alternate under sustained load and that medians run a few percent higher. It is also unclear from the supplied information whether the project will support additional GPUs, operating systems or driver configurations.
Testing the Published Implementations
Readers can inspect the repository’s network documentation, build instructions and parity tools to see how the authors describe the implementation and reproduce checks, provided they have compatible hardware and model files. The project includes a demo and profiling commands, while the browser port offers a separate route to compare results without tensor-core FP8 execution.
No independent review, follow-up release schedule or Nvidia response is included in the source material. Further confirmation would depend on external testing of the parity claims, benchmark conditions and weight requirements. Until then, the reported results should be treated as claims made by the project authors.
Key Questions
What is OpenDLSS?
OpenDLSS is a GitHub project that reimplements Nvidia’s DLSS 5 Neural Rendering network in Vulkan and provides a separate WebGPU version for browsers.
Does it implement DLSS Super Resolution?
No. The project says DLSS Super Resolution is a different network and is not implemented. This network produces output at the input frame’s resolution.
Are the bit-exact results independently confirmed?
The project reports matching outputs at all 75 block boundaries against reference captures. The supplied material does not include an independent verification of that claim.
What hardware does the Vulkan version require?
The listed requirements include Windows, an Nvidia Ada-or-newer GPU, and a driver exposing the Vulkan and Nvidia extensions named in the repository. Users must also provide the model weights.
How fast is the implementation?
On an RTX 4070 SUPER, the project reports minimum Vulkan frame times ranging from 2.8 ms at 768-by-768 to 29.3 ms at 3840-by-2160, measured over 40 frames. These are project-reported results for that GPU, not independent or cross-hardware benchmarks.
Source: hn
