TL;DR
OpenDLSS is a public project describing a Vulkan reimplementation of NVIDIA’s DLSS 5 neural rendering network. Its GitHub documentation claims byte-for-byte agreement across 75 block boundaries and reports RTX 4070 SUPER timings, but the project requires users to provide model weights and compatible NVIDIA hardware. Independent verification and the weights’ availability are not established in the supplied material.
The project describes the model as a 71-block shifted-window transformer network with a global vision transformer at its deepest level, spread across six pooling levels. Its documentation says the implementation runs FP8 E4M3 activations with FP16 accumulation and uses 141 MiB of weights. OpenDLSS says its parity tests match not only the final image but intermediate results at every block boundary, byte for byte. Those are claims made by the project; the supplied report does not include independent validation.
Play games on Amazon Luna for your game nights, included with Prime
- A rotating selection of games, no download needed
- Play on TV, laptop or phone
- Fast, free delivery for your gear, too
OpenDLSS characterizes DLSS 5 as generative neural rendering, rather than upscaling. The network takes a rendered frame at its existing resolution, alongside noise inputs, a reprojected prior output and five conditioning values, and returns four values per pixel: an RGB residual and a temporal-blend logit. In the project’s account, this process can generate detail and alter tone, structure and skin according to a style setting. The report says the model’s input and output resolutions are the same.
The project reports timings for the full network on an RTX 4070 SUPER: 2.8 milliseconds at 768×768, 7.8 ms at 1920×1080, 12.6 ms at 2560×1440 and 29.3 ms at 3840×2160. These are project-reported minimums over 40 frames; it says GPU clock changes under sustained load make median results a few percent higher. They are not independent benchmark results, and the report does not provide comparisons with NVIDIA’s own implementation.
A second implementation in browser WebGPU is described as matching the same captures without tensor cores or FP8. OpenDLSS reports 72 ms at 512×512 for that port, compared with 2.7 ms for its Vulkan implementation at that resolution. The report identifies the WebGPU figure as a specific test, not a general browser performance guarantee. A demo also embeds the network in Filament, a real-time rendering engine, and implements a temporal feedback path; the command-line tool instead processes single frames without history.
A New Route to Test Neural Rendering
The project could give graphics developers and researchers another way to inspect and experiment with a neural rendering network, including its intermediate computations. Its claimed block-by-block parity matters because matching only final images can hide differences inside a model. If independently reproduced, the implementation may help others study how the network behaves across different execution paths.
Access remains conditional. The project says the Vulkan route requires Windows, an NVIDIA Ada-or-newer GPU and drivers exposing several specified Vulkan extensions, including cooperative-matrix and shader-float8 support. It also requires model weights supplied separately by the user. The project is not described as a drop-in replacement for NVIDIA’s software stack, and it explicitly says DLSS Super Resolution is not implemented.
Products Worth Considering

LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDR True Black 400

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

Logitech MX Mechanical Mini for Mac Wireless Keyboard, Space Gray
As an affiliate, we earn on qualifying purchases.
How OpenDLSS Describes the Model
The GitHub report identifies its target as the network in DLSS-NR build 310.8.0 and points to NVIDIA’s DLSS 5: Generative Neural Rendering project page for the model description. OpenDLSS organizes its work into a C++20 host program, Vulkan shader kernels, generated PTX kernels and the browser WebGPU port. It also provides a graph document and commands for benchmarking, profiling and checking parity against fixtures.
The project says its reference Vulkan route uses cooperative-matrix FP8 operations and fused kernels, while its faster route uses NVIDIA PTX features, including E4M3 matrix operations and asynchronous data movement. That implementation detail helps explain why the reported Vulkan timings are tied to NVIDIA hardware. The report lists Filament v1.77.0 for the optional demo and says the engine is patched to provide motion vectors and Vulkan interoperability.
““The intermediates match too, not just the final image: all 75 block boundaries, byte for byte.””
— OpenDLSS GitHub project documentation
Parity, Weights and Access Questions
The available source is the project’s own GitHub report. It does not establish that an independent party has reproduced the claimed byte-exact results or confirmed the reported benchmark methodology. The source also does not say where users can obtain compatible model weights, whether their distribution is authorized, or what licensing terms apply to those weights. The implementation’s repository and model files should not be treated as having the same availability or licensing status without further documentation.
The report gives requirements for the Vulkan route but does not establish support for other operating systems or GPU vendors. It also does not provide a direct performance comparison with NVIDIA’s DLSS 5 implementation, nor does it say how results vary across scenes or driver versions. The WebGPU port’s matching claim is tied to the project’s captures; broader behavior and performance are not demonstrated by the information supplied.
What Users Need to Run It
According to the project instructions, users can fetch its toolchain and build the Vulkan program on a supported Windows system, then run it against a model directory they provide. Separate commands are listed for benchmarking, profiling and parity checks using fixtures. The optional Filament demo requires additional downloads and a separate build, while the browser port offers another route to examine the network without tensor cores.
The next useful evidence would be independent parity tests, clear information on model-weight sourcing and licensing, and benchmarks that identify driver versions and compare results on matched hardware. Until then, OpenDLSS is best understood as a technically detailed project report with significant claims that remain attributable to its authors.
Key Questions
What is OpenDLSS?
OpenDLSS is a GitHub project describing a Vulkan reimplementation of NVIDIA’s DLSS 5 neural rendering network, plus a separate browser WebGPU implementation. The report says users must supply the model weights.
Does OpenDLSS implement DLSS Super Resolution?
No. The project explicitly says DLSS-SR is not implemented and describes it as a different network.
What hardware does the Vulkan version require?
The project lists Windows, an NVIDIA Ada-generation or newer GPU and a driver exposing the Vulkan extensions it uses. It also requires development tools to build the software.
Has the byte-for-byte parity claim been independently verified?
The supplied information presents this as an OpenDLSS project claim. It does not provide independent verification.
How fast does OpenDLSS run?
On an RTX 4070 SUPER, the project reports minimum full-network times of 7.8 ms at 1920×1080 and 29.3 ms at 3840×2160, among other resolutions. These are author-reported measurements, not independent benchmarks.
Source: hn
