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 in Vulkan and says its outputs match reference captures bit for bit, including intermediate block outputs. It also provides a separate browser-based WebGPU implementation. The project requires compatible Nvidia hardware and model weights; its claims and performance figures come from the project’s own documentation.
A GitHub project called OpenDLSS presents a Vulkan reimplementation of Nvidia’s DLSS 5 neural rendering network, with a separate browser-based WebGPU port. The project author says the implementation is bit-exact against reference captures, including outputs at all 75 block boundaries, offering developers a way to examine and run the network outside Nvidia’s original implementation.
The project describes the model as a 71-block shifted-window transformer and global vision transformer, arranged across six pooling levels. It says the network uses E4M3 FP8 activations with FP16 accumulation and has 141 MiB of weights. Users must supply the model weights in the directory format documented by the project; OpenDLSS does not provide them in the supplied material.
OpenDLSS says its Vulkan route runs on Nvidia Ada-generation or newer GPUs with drivers exposing several specified Vulkan and Nvidia extensions. The author reports minimum per-frame times on an RTX 4070 SUPER of 2.8 milliseconds 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. These are project-reported measurements, not independently verified benchmarks.
The project also includes a browser WebGPU implementation that it says matches the same captures without tensor cores or FP8. Its documentation reports 72 ms at 512 by 512 for the browser route, compared with 2.7 ms for the Vulkan implementation at that resolution. The repository includes source code, shaders, build scripts, a profiling tool and a demo that integrates the network into a Filament-based renderer.
What OpenDLSS Offers Developers
If the reported parity holds, OpenDLSS could give graphics researchers and engine developers a practical way to inspect the network’s computation and test its behavior across implementations. The claim that intermediate outputs match byte for byte is particularly relevant for debugging: it offers a more granular comparison than checking only the final rendered image.
The project is not a general replacement for Nvidia’s graphics features. Its documentation says the network re-renders an image at its existing resolution, adding or changing visual detail under a style setting; it is not an upscaler. The repository also says it does not implement DLSS Super Resolution, a separate network. Its hardware requirements and need for separately supplied weights limit who can run the Vulkan version, while the slower WebGPU port provides a different route for testing in a browser.
Nvidia RTX 4070 SUPER graphics card
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How the Network Fits DLSS
Nvidia describes DLSS 5 as generative neural rendering. In OpenDLSS’s account, the network receives a rendered frame along with noise inputs, a reprojected prior output and five conditioning values, then produces an RGB residual and a temporal-blend value for each pixel. The model therefore operates on an engine-rendered image rather than generating a scene from scratch.
The repository separates its implementation into a Vulkan route using GLSL and generated PTX kernels, and a WebGPU port that prioritizes matching the reference behavior without the same hardware-specific acceleration. Its demo includes a temporal feedback path, but the command-line tool runs single frames without history. These distinctions matter when interpreting the project’s parity and speed claims: the tools do not all exercise the same runtime path.
“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.
Parity and Access Still Need Checking
The bit-exact parity and performance results are claims in the repository, and the supplied material contains no independent test results or external review. It does not establish how the implementation performs across different compatible GPUs, driver versions or scenes. The published timings are minimum measurements over 40 frames, and the author notes that sustained GPU clock changes can make median results a few percent higher.
The source also says users must provide the model directory, but the supplied report does not establish the weights’ availability, licensing terms or provenance. It is not clear whether Nvidia has endorsed, audited or formally assessed OpenDLSS. The project’s description of visual changes, including adjustments to tone, structure and skin, should be understood as its account of the network rather than an independently evaluated assessment of output quality.
As an affiliate, we earn on qualifying purchases.
Testing the Public Implementation
The immediate next step for interested developers is to inspect the GitHub source, model-format documentation and test fixtures, then run the parity and profiling tools on supported hardware with compatible weights. The project documents separate commands for comparing reference captures and measuring individual dispatches, as well as a demo build for rendered scenes.
Further independent tests could establish whether the reported matching results and frame times hold across systems and inputs. The supplied source does not announce a release schedule, Nvidia response or planned roadmap, so those developments remain unknown.
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, with a separate browser-based WebGPU port.
Does OpenDLSS include DLSS Super Resolution?
No. The project says it implements the neural rendering network, not DLSS Super Resolution, which it describes as a different network.
What hardware does the Vulkan version require?
The project lists Windows and an Nvidia Ada-generation or newer GPU, along with a driver exposing specified Vulkan and Nvidia extensions. It also requires users to supply the model weights.
Are the performance numbers independently verified?
Not in the supplied source. The RTX 4070 SUPER timings are reported by the project author and are described as minimum times over 40 frames.
Can the network be tested in a browser?
Yes. The repository includes a WebGPU port that the author says matches the same reference captures, though its published 512-by-512 runtime is slower than the Vulkan route.
Source: hn
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
