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 GitHub project that reimplements NVIDIA’s DLSS 5 Neural Rendering network using Vulkan and also provides a separate browser-based WebGPU port. Its author reports byte-for-byte agreement with reference captures, but the project does not provide the model weights and requires a recent NVIDIA GPU with specific driver support.
A GitHub project called OpenDLSS has published a Vulkan reimplementation of NVIDIA’s DLSS 5 Neural Rendering network, alongside a separate browser-based WebGPU version. The project author says both implementations reproduce reference outputs byte for byte; the code does not include the model weights, which users must provide separately.
The Vulkan version targets Windows systems with an NVIDIA Ada-generation or newer GPU and driver support for several specified Vulkan and NVIDIA extensions. The project describes the network as a 71-block shifted-window transformer and global vision transformer, with FP8 E4M3 activations, FP16 accumulation and 141 MiB of weights. It says the implementation also matches intermediate results at all 75 block boundaries, not only the final image.
According to the project’s benchmark, the full network took a minimum of 7.8 milliseconds at 1920 by 1080 on an RTX 4070 SUPER, measured over 40 frames. Reported minimum times were 2.8 ms at 768 by 768, 12.6 ms at 2560 by 1440 and 29.3 ms at 3840 by 2160. The author cautions that the GPU alternated between two clock states under sustained load and says median times were a few percent higher.
The independent WebGPU port runs without tensor cores or FP8. The project reports 72 ms at 512 by 512 for that version, compared with 2.7 ms for its Vulkan implementation at the same resolution. The code describes the network as a same-resolution rendering process, not an upscaler: it takes a rendered frame and additional inputs, then generates an RGB residual and a temporal-blend value.
What OpenDLSS Makes Testable
The project makes a claimed reproduction of a proprietary neural-rendering pipeline available for technical examination and experimentation. Its reference captures, parity and verification tools, and description of intermediate outputs give developers ways to compare implementations at a level beyond judging final images by eye. The separate browser version also illustrates how the same model can be run without specialized tensor-core hardware, albeit at much lower reported speed.
These results do not establish that OpenDLSS is an official NVIDIA release or that it will work as a drop-in replacement in commercial games. Its stated hardware and driver requirements are narrow, and users need an appropriately formatted model directory. The project also says it does not implement DLSS Super Resolution, a different network, so its name should not be read as a reimplementation of every DLSS feature.
NVIDIA RTX 4070 SUPER graphics card
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Network Behind the Project
NVIDIA describes DLSS 5 as generative neural rendering. In the project’s account, the network processes a frame already rendered by a game engine at the same resolution. Inputs include a low-dynamic-range frame proxy, three lanes of Gaussian noise, a reprojected prior output and five conditioning values. The network returns four values per pixel: an RGB residual and a logit used for temporal blending.
OpenDLSS divides its implementation into Vulkan kernels and a separate WebGPU port. Its demo integrates the network with Filament, while the command-line tool can benchmark, profile and compare output against fixtures. The project says the command-line path processes individual frames without history; temporal feedback is implemented in the demo. Its documentation identifies NVIDIA’s DLSS 5: Generative Neural Rendering project page as information about the model.
“A Vulkan reimplementation of NVIDIA’s DLSS 5 Neural Rendering network, bit-exact against the original.”
— OpenDLSS project documentation
As an affiliate, we earn on qualifying purchases.
What the Accuracy Claim Establishes
The supplied material presents bit-exactness as the project author’s claim, supported by its parity and verification tools and reference fixtures. It does not include an independent test, an NVIDIA statement endorsing the reimplementation, or details about how the reference captures were obtained. The reported timings are also project benchmarks; the material does not provide results from other systems or an independent reproduction.
The weights are not supplied with the code. The source material does not establish their availability, distribution terms or whether a user can obtain them through an officially supported channel. It also does not specify how the implementation performs across games, scenes or driver versions beyond the documented test setup.
As an affiliate, we earn on qualifying purchases.
Testing Depends on Weights
The next practical step for interested developers is to review the project’s model-directory format and hardware requirements, then run its parity or benchmark tools with compatible weights and reference fixtures. The project documentation provides build instructions for the Vulkan tool and demo, as well as a separate route for the browser port.
Further independent testing could clarify whether the reported output matches hold across additional scenes and systems, and whether the performance figures are reproducible. Until such evidence or further project updates appear, OpenDLSS should be treated as a community implementation with developer-reported results, not an NVIDIA-supported product.
WebGPU compatible graphics hardware
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 with Vulkan and provides a separate WebGPU version. The project author reports byte-for-byte agreement with reference captures.
Does OpenDLSS include the model weights?
No. The project documentation says users must supply a model directory containing the weights in the specified layout. The supplied material does not confirm where those weights can be obtained or their distribution terms.
Is OpenDLSS an upscaler?
The project says the network operates at the input frame’s resolution and is not an upscaler. It describes the process as neural rendering that modifies a frame the engine has already rendered.
What hardware does the Vulkan version require?
The documented requirements include Windows, an NVIDIA Ada-generation or newer GPU, and a driver exposing the Vulkan and NVIDIA extensions listed by the project. The WebGPU port runs without tensor cores or FP8 but is reported to be slower.
Has NVIDIA endorsed the implementation?
The supplied project information does not say that NVIDIA endorses or supports OpenDLSS. Its accuracy and benchmark statements are claims made in the project documentation.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
