DLSS-NR-on-AMD ships one installer binary and no source, and it needs anti-cheat switched off to load at all

Anti-cheat has to be off, and that is a term not a setting

The project's own requirements list ends with the sentence that decides whether you can use it at all. Games with anti-cheat will block the DLL, and the instruction that follows is to use them only with anti-cheat off.

That is not a configuration option. Disabling anti-cheat in a game with matchmaking means joining public sessions without the mechanism that exists to keep modified clients out, in a title where other players are present and where the modification is running a vendor's frame generation runtime on hardware the vendor does not target. Nothing in the page limits the guidance to single player.

The legal section beside it states that the project is not affiliated with or endorsed by NVIDIA or AMD, that DLSS is a trademark of NVIDIA Corporation, that the software is provided as-is without warranty, and that use is at your own risk.

So the terms come before the capability here, and they are the reason this article describes what the project is rather than how to run it. Anyone who has anti-cheat in a competitive game should close the page at this point rather than at the next one.

The repository holds four entries and none of them is code

The top level of the repository is `.github/`, `LICENSE`, `README.md` and `THIRD_PARTY.txt`. There is no source directory, no build system, no project file, and the project records its primary language as unknown.

What the page claims to be is considerably larger than that. The FAQ answers the question of whether this is a compatibility layer with a flat no, and describes it instead as a ground-up, full reimplementation of the neural rendering runtime for AMD GPU architectures, naming HIP kernels, memory layouts and game integration, written from scratch. So the repository documents a reimplementation of a proprietary runtime on a different vendor's hardware and then ships the reimplementation as a compiled Windows installer, with the code staying on the author's machine.

There is a licence conflict here that the repository does not resolve. The project metadata records no licence at all, and a `LICENSE` file is present at the root. Nothing on the page says which terms apply to the binary being distributed, and the legal section is a disclaimer rather than a grant.

For anyone deciding whether to look further, the practical consequence is that there is no way to check the claim about what is or is not in the binary.

A reimplementation that depends on a specific Nvidia DLL build

The FAQ makes an exclusion claim: no NVIDIA code is included or translated at runtime. The requirements section makes a dependency claim: the project needs your own copy of a named NVIDIA file, at a named build, `nvngx_dlssnr.dll` version 310.8.0.0, which the page says you can get from games that support DLSS 5.

Those two statements sit about ten lines apart, and the second one is the operative one. Whatever the reimplementation replaces, the page's own instructions have a vendor binary loaded into the game process alongside it, at a specific version, alongside a specific driver floor for the AMD side.

The rest of the dependency chain is just as specific. Windows 11 is required, as is a DirectX 12 game that uses FSR, and any FSR 3 or FSR 4 quality level is said to work. The AMD driver floor is Adrenalin 26.1.1 or later. Vulkan is described as planned for a later release, so the Vulkan path does not exist yet.

The testing statement is narrow: tested on Cyberpunk 2077 and GTA V Enhanced, on one specific card.

The newest alpha was published two days after the last commit

Three releases landed inside five days, and all three are titled as alphas while carrying version numbers on a zero line. v0.5.0 is dated 2026-09-29, v0.5.1 on 2026-10-01, and v0.6.0 on 2026-10-03.

The default branch is `master`, the repository is not archived, and its most recent push is dated 2026-10-01. So the newest artefact was published about two days after the last commit recorded on the branch. Whatever produced v0.6.0, it was not built from the tip of the branch as the repository records it, and with no source in the tree there is no way to tell what changed between v0.5.1 and v0.6.0.

The funding note at the end of the page explains the cadence. It asks for donations, with tokens described as expensive and GPUs to test the mods on as more expensive, and it asks for reports from people with a 7000 series card or other titles, with a log file named for the purpose attached to any issue.

A project shipping binaries at that rate, with no source, and asking for hardware donations to test on, is best read as one person's work in progress rather than as a maintained tool.

One tested card, one hopeful generation, and no older support

The hardware line is drawn in three tiers and only one of them is a claim of fact. The RX 9000 series is the supported target, and the page names a 9070 XT as the card it was tested on. The RX 7000 series is expected to work, with reports invited. Older GPUs are described as not currently supported.

So the tested configuration is a single card from the newest generation, on two named titles, with a single driver floor. Everything else on the list is a prediction or a request for data.

The image quality claim is hedged in the same register. Output quality is described as very similar to what DLSS produces on the original hardware, immediately followed by the admission that there are still some effects that may not be working fully. There is no list of which effects, no comparison captures, and no way to tell from the page whether the gap is cosmetic or structural.

The in-game overlay is the part that shows the feature set honestly, and it is small. It is opened with the End key, adjusted with the arrow keys and toggled with Enter, and it exposes four options: a mode that is inline or async for photo mode only, and three intensity sliders for tone, structure and skin structure. Model selection is listed as planned rather than implemented.

The one performance figure is a frame rate at one resolution

There is exactly one performance number on the page, and it comes with two qualifications. Performance is described as a work in progress, and the figure given is roughly 33 FPS at 1080p on an RX 9070 XT. The stated goal is to get it running as fast as a 5070Ti.

Neither half of that sentence is a benchmark. There is no frame generation figure at all, which is the number a reader would most want from a project whose entire purpose is a frame generation runtime. There is no CPU cost, no VRAM cost, no input latency, no comparison against FSR 3 or FSR 4 on the same card, and no indication of which settings produced the figure. The 5070Ti comparison is a target, not a measurement, and it names no configuration on either side.

The page also asks readers to check back for new releases rather than measuring anything themselves, which is a reasonable thing to say for a hobby project and also the reason there is nothing here to verify.

Treated as a single data point, roughly 33 FPS at 1080p tells you the thing runs and is not fast. Treated as a claim about quality, the comparison to native DLSS output is unsupported by anything on the page.

Editorial conclusion

This is a hobby project with a legal section, a funding appeal and a single number for performance, and that combination is enough to decide against it for anything but reading. It is not suitable for real play, and the project says so in pieces: the quality claim comes with an admission that some effects may not work fully, the performance claim is a work in progress measured on one card at one resolution, three of the five listed features are marked planned, and the requirement list asks you to disable anti-cheat in a multiplayer game. If you want to understand the approach, the README's account of HIP kernels, memory layouts and game integration is the only description available, because the repository contains no code to read.

Frequently asked questions

Official sources

1. danielblnc/DLSS-NR-on-AMD on GitHub
2. Issues
3. README
4. Releases