NVIDIA explains DLSS 4.5 model presets, and shows how to confirm what is running
NVIDIA has published a short DLSS 4.5 Super Resolution FAQ after users ran into confusion with the NVIDIA App’s DLSS Override controls.
The company says DLSS 4.5 Super Resolution moves to a second-generation transformer model. NVIDIA links the image quality gains to more training and algorithm changes, plus a large jump in raw compute demand. RTX 40 and RTX 50 can offset part of that cost with FP8 acceleration, while RTX 20 and RTX 30 do not support FP8 and can take a larger performance hit. N VIDIA says RTX 20 and RTX 30 owners may prefer staying on the older Model K preset for higher FPS.
NVIDIA also confirms two new Super Resolution presets. Model M is tuned for DLSS Performance mode, and Model L is tuned for 4K DLSS Ultra Performance mode. Both can still be selected in other DLSS modes, including DLAA, but NVIDIA frames Performance and Ultra Performance as the intended targets. NVIDIA also says Ray Reconstruction is not moving to the same second-generation transformer architecture in this update, so the gains discussed here are tied to Super Resolution.
Here’s where users have been getting confused by the NVIDIA App UI.. NVIDIA’s own app post says that picking “Latest” under DLSS Override, Model Presets will apply the newest model, and it notes that “Latest” for Super Resolution currently maps to preset M.
NVIDIA also added a simple way to verify what is active. Turn on the NVIDIA app overlay statistics view via Alt+Z, then Statistics, then Statistics View, then DLSS to see the model in use.
We got some questions from the community on DLSS 4.5 Super Resolution and wanted to provide a few points of clarification.
DLSS 4.5 Super Resolution features a 2nd generation Transformer model that improves lighting accuracy, reduces ghosting, and improves temporal stability. The new model delivers this image quality improvement via expanded training, algorithmic enhancements, and 5x raw compute. DLSS 4.5 Super Res uses FP8 precision, accelerated on RTX 40 and 50 series, to minimize the performance impact of the heavier model. Since RTX 20 and 30 Series don’t support FP8, these cards will see a larger performance impact compared to newer hardware and those users may prefer remaining on the existing Model K (DLSS 4.0) preset for higher FPS.
DLSS 4.5 Super Resolution adds support for 2 new presets:
- Model M: optimized and recommended for DLSS Super Resolution Performance mode.
- Model L: optimized and recommended for 4K DLSS Super Resolution Ultra Performance mode.
While Model M and L are supported across DLSS Super Resolution Quality, Balanced modes, and DLAA mode, users will see the best quality vs. performance benefits in Performance and Ultra Performance modes. Additionally, Ray Reconstruction is not updated to the 2nd gen transformer architecture benefits are seen using Super Resolution only.
To verify that the intended model is enabled, turn on the NVIDIA app overlay statistics view via Alt+Z > Statistics > Statistics View > DLSS.
We look forward to hearing your feedback on the new updates!
The FAQ helps, but the override flow still needs clearer in-app wording. A gamer should not need a forum thread to learn that “Latest” can map to a specific model preset and change the performance profile on older cards.
One good thing from this situation is that NVIDIA is still letting RTX 20 and RTX 30 run DLSS 4.5 Super Resolution with an explicit “stick to Model K if you want FPS” escape hatch. AMD could learn a thing from this, their AMD’s FSR “Redstone” branding for FSR 4 remains officially tied to RDNA 4.
Source: NVIDIA
