DLSS 5 runs on RTX 4000 with a mod—but performance is the catch

Yes—an RTX 4000 GPU can run DLSS 5 unofficially. A patched version of NVIDIA’s leaked neural-rendering library has been verified on an RTX 4080, proving that Ada Lovelace hardware can execute the code. But that is the compatibility answer, not the buying-advice answer: the current modded implementation is experimental, unsupported and often brutally expensive in performance.

The RTX 4080 verification

The breakthrough came from adapting the leaked nvngx_dlssnr.dll, a library found in an early-access build of NBA 2K27. The file was reported as 158 MB and identified as NVIDIA DLSSNR 310.8.0.0.

The original library was built around CUDA binaries targeting Blackwell, the architecture used by the RTX 5000 series. A community patch replaced the incompatible binaries with versions that could run on Ada Lovelace. That change allowed the same neural-rendering model to execute on an RTX 4080 without turning the card into an RTX 5000 overnight—sadly, no such miracle patch exists for your electricity bill.

The result matters because it separates two ideas that are easy to confuse:

• Technical compatibility: the patched library can execute on at least one RTX 4000 model.
• Supported product functionality: NVIDIA and game developers provide the integration, testing and controls needed for a reliable feature.

The first has been demonstrated. The second has not.

What the patch changes

DLSS 5’s neural rendering is designed to work with information supplied by the game engine, including scene data, masks and artistic controls. A community injector such as RenoDX can place the leaked library into a game, but it does not automatically give that library every piece of developer-level information intended for an official integration.

That helps explain why a mod can produce striking results in one scene and strange ones in the next. The early Control implementation exposed seven Neural Rendering styles and mainly affected character models. It demonstrated the feature’s visual ambition, but it was still a rough injection rather than a finished game-specific implementation.

The community workflow is visible in the DLSS 5 Swapper interface, which includes a game library, compatibility badges and controls for adding or rescanning titles. Those controls document the tool’s workflow; they do not establish that every listed game works reliably or that the software is safe to use.

The performance penalty

The early numbers are the part that should make RTX 4000 owners pause. They come from different GPUs, games and conditions, so they are not a universal benchmark for every Ada card. They do, however, show a consistent warning: the mod can cost far more performance than its visual effects are worth.

The RTX 5070 Ti result is especially useful because it keeps the resolution and before-and-after figures together. The RTX 5090 results show that even the hardware originally targeted by the leaked build can suffer a substantial penalty in early community packages. The RTX 3080 example is much more extreme, dropping into single-digit performance, but it should not be transferred directly to RTX 4000 cards.

A separate community test also showed Cyberpunk 2077 falling from 138 to 68 FPS at 1440p on an RTX 5070. The settings and implementation differ from the other tests, so it reinforces the direction of the result rather than creating a clean cross-GPU comparison.

A video demonstration shows RenoDX forcing the mod to run on a Zotac GeForce RTX 4060 Twin Edge 8GB GDDR6 in Cyberpunk 2077. The footage includes an on/off comparison, configuration screens and a live performance overlay. It is useful evidence that the mod can operate on another Ada card, but it is not a controlled RTX 4000 performance benchmark.

Why the visuals divide players

The appeal is easy to understand. Neural rendering can improve the apparent treatment of lighting, shadows, materials and character detail without changing the game’s underlying geometry. In the best-looking scenes, the result can resemble a more advanced lighting pass.

The trade-off is consistency. Community users have reported altered faces, noise, shimmering and lighting that changes unpredictably during motion. Some characters can look noticeably different from their intended designs. That is more than a minor blemish when the effect is applied to a story-driven game: a sharper face is not an upgrade if it stops looking like the same character.

There is also a basic frame-time issue. Generated frames can make motion appear smoother, but they do not remove the input latency of the underlying rendered frames. A large FPS number in an overlay therefore does not automatically mean the game feels more responsive. A solid base frame rate still matters.

Official support versus community execution

The patched RTX 4080 result does not mean that RTX 4000 has official DLSS 5 support. It shows that a leaked, pre-release library can be adapted to run on Ada hardware through unofficial tools and modified binaries.

Official support would require a supported software path, game-level integration and behavior that has been tested across the relevant hardware and titles. The modded implementation has none of those guarantees. Its results depend on the game, the injection method, the selected rendering style and the hardware involved.

That distinction also explains why the original Blackwell-only behavior and the later RTX 4000 demonstration are not contradictory. The unmodified leaked binary targeted Blackwell; the later patched version added an Ada-compatible route.

What RTX 4000 owners should take away

If your question is whether an RTX 4000 card can execute leaked DLSS 5 code, the answer is yes, unofficially. The RTX 4080 verification and RTX 4060 demonstration make that clear.

If your question is whether the mod is currently a sensible way to improve gaming performance, the evidence points the other way. Early tests show large frame-rate losses, image-quality problems and game-dependent behavior. No supplied result establishes a consistent performance profile for the entire RTX 4000 family, so an RTX 4080 result should not be treated as a promise for every Ada model.

The practical verdict is simple: this is an intriguing look at what RTX 4000 hardware may be capable of, not a finished feature worth planning a build around. Compatibility has been cracked. Usability is still the boss fight.