DLSS 4.5: with Dynamic Frame Generation and MFG 6X NVIDIA raises the stakes

Technological Suites for GPUs

The technological suites accompanying graphics cards have become a key element to consider when purchasing a GPU, desktop PC, or notebook. The advent of ray tracing and later path tracing, combined with an increasing number of polygons and geometric complexity, has raised the bar for performance required to enjoy games at their best.

The silicon, the term we use to refer to the actual hardware made up of shader units and the rest, does not allow for high resolution and frame rate handling in the most demanding games with such quality settings, nor would it be beneficial for manufacturers to do so because it would incur excessive costs. This is where suites like DLSS come into play.

The Deep Learning Super Sampling is looking ahead to its fifth version, expected later this year, but for some time now it has been possible to access version 4.5. Initially born as an upscaling algorithm, DLSS has now evolved into much more and, with version 5, it will take on even more significant features.

DLSS 4.5 was announced at CES 2026 in Las Vegas and introduced two new features that strengthen NVIDIA's suite's role as a reference point for the entire industry: Dynamic Multi Frame Generation and Multi Frame Generation 6X.

DLSS 4.5 can generate up to five additional frames for every traditionally rendered frame, potentially allowing for over 240 FPS gaming with path tracing on GeForce RTX 5000 series GPUs.

If Frame Generation, introduced with DLSS 3 and then expanded with Multi Frame Generation 4X with DLSS 4, doesn’t need much introduction, what is Dynamic Multi Frame Generation? The name says a lot, if not everything.

It is an intelligent system that works like an automatic transmission. Instead of relying on a fixed multiplier, it automatically switches between frame multipliers to find the perfect balance between frame rate, image quality, and responsiveness. In other words, it generates only the frames necessary to maximize the desired frame rate or display refresh rate, whether it be 120 Hz, 144 Hz, 240 Hz, or higher.

By constantly monitoring the gap between GPU performance and the maximum display refresh rate, the system adapts in real time. In graphically intensive scenarios, it increases frame generation to fill performance drops, ensuring that the high refresh rate display remains smooth. Conversely, when the workload lightens, it gradually reduces the multiplier to process only what is necessary.

How to Activate Dynamic Multi Frame Generation and MFG 6X

Dynamic Multi Frame Generation and MFG 6X are supported in over 200 games and apps through NVIDIA App. And like DLSS 4, they are accessible only with GeForce RTX 5000 GPUs. Some games, like 007 First Light, support both new technologies directly from the game engine; otherwise, it can be forced through NVIDIA App, the new control panel for GeForce GPUs.

To activate, download the latest version of NVIDIA's Game Ready Drivers and NVIDIA App, go to NVIDIA App, select “Graphics” from the left bar, and find the game you want to test/play. Once selected, find the option “Override DLSS – Frame Generation Mode” and select Dynamic. Set “Target FPS” to the maximum refresh rate and the multiplier up to 6X, then click Apply.

Another New Feature in DLSS 4.5

There’s also another new feature in DLSS 4.5, a new frame generation model, activable by RTX 4000 and 5000 users by setting Preset B in NVIDIA App, which improves user interfaces in some titles, leveraging additional data in the game engine. Recently, NVIDIA also announced the arrival of a second-generation Transformer model for Ray Reconstruction in August.

DLSS 4.5, Some Tests with Cyberpunk 2077

To analyze Dynamic Multi Frame Generation and Multi Frame Generation 6X, we used a Razer Blade 16 notebook, equipped with an AMD Ryzen AI 9 HX 370 processor, 32 GB of RAM, and the top-of-the-line NVIDIA GeForce RTX 5090 Laptop GPU (24 GB).

Tests with Cyberpunk 2077 were conducted at a native resolution of 2560x1600 pixels with a quality preset that sets Ray Tracing to Low (the DLSS test used Balanced setting). We gathered some data in various modes, with results acquired via NVIDIA FrameView: the numbers collected were useful not only to evaluate the average frame rate but also the consistency of frame distribution and the overall input lag of the system.

Before reviewing the results, it is essential to clarify some terms: 1% Low FPS, 0.1% Low FPS, and AvgPCLatency:

1% Low FPS: One of the most important metrics, representing the average of the slowest 1% of frames. If this number is close to the Avg FPS, the game is extremely smooth. If it is much lower, it means noticeable stuttering.
0.1% Low FPS: Similar to the previous one but even more stringent, calculating the average of the slowest 0.1% of frames to measure sudden micro-stuttering. If this value drops drastically, it indicates spikes in lag.
AvgPCLatency (ms): The average time (in milliseconds) it takes the PC to process a frame, from when the mouse/keyboard sends the input to when the video card finishes rendering the frame. The lower, the better.

The test without upscaling and frame generation technologies returned an average frame rate of 90.3 FPS, with a “1% Low” value of 62.3 FPS, ensuring a smoothness that doesn’t drop below the critical threshold of 60 frames per second, while the 0.1% Low drops to 35.3 FPS. The system latency (AvgPCLatency) settled at 48.3 ms.

Then we turned on Super Resolution, meaning upscaling, using the Balanced setting. This session displayed similar behavior to the previous one, recording an average of 125 FPS. The 1% Low reached 78.9 FPS, while the 0.1% Low stopped at 41.4 FPS. The latencies, however, exhibited optimization compared to the first native test: the AvgPCLatency dropped to 25 ms, outlining significantly more responsive gameplay despite the similar average frame rate.

From here on, we saw a vertical performance increase generated by activating DLSS with Frame Generation. Enabling Dynamic Frame Generation (DFG) up to 6X, we recorded an average frame rate of 236.3 FPS, with excellent behavior of 1% Low (142.2 FPS) and 0.1% Low (90.4 FPS). The PC latency stood at 35.5 ms, considerably lower than the native rendering.

Next, switching to DLSS 2X, the original frame generation for better understanding, the frame rate settled at 202.8 FPS, with a 1% Low of 131.8 FPS. Despite the FPS decrement compared to DFG, system latency was notably lower at 33.4 ms.

DLSS 3X raised the average frame rate to 274.6 FPS. The 1% Low climbed to 168.7 FPS, and the 0.1% Low stabilized at 120.1 FPS, showing remarkable visual consistency despite a latency of 34 ms. The DLSS 4X setting pushed the average frame rate to 330.6 FPS, with 1% Low at 187.9 FPS. In our case, average latency increased to 35.8 ms.

With the DLSS 5X and 6X settings, even higher performance peaks were reached. In the first case, we hit an average of 382.7 FPS, with a 1% Low that exceeded 216 FPS. Latency stood at 37.1 ms. In the second case, the average reached 421.1 FPS. The most relevant data is the 1% Low at 240.8 FPS, meaning the fluctuation of the slowest frames remains tied to the native refresh rate of the notebook's display (240 Hz). Conversely, system latency shows a value of 38.6 ms.

As the options for generating more artificial frames increase, the frame rate grows as we rise with the multiplier, with 420 FPS on average in 6X mode allowing us to increase ray tracing settings and even set path tracing. It can be noted that Dynamic Frame Generation is well-suited to ensure a constant balance in every gaming scenario. Moreover, it can be set “up to” 6X, allowing for the best option depending on quality settings and the available system, ranging from up to 3X to, precisely, 6X.

Taking the integrated panel of the Razer Blade 16, characterized by a refresh rate of 240 Hz, in our case, the ideal setting was with DLSS 6X, as it represents the qualitatively perfect session for a 240 Hz display. With 1% Low at 240.8 FPS, the system ensures that the minimum perceived smoothness coincides steadily with the monitor's refresh.

Conversely, the session with DFG is essentially the most efficient: with 236.2 Avg FPS, it almost perfectly aligns with the physical limit of 240 Hz of the panel, with reduced latency (35.5 ms) and a 1% Low preserving smoothness at 142.2 FPS.

Traditional rendering placed last in the comparative analysis. The 90.3 Avg FPS and peak latency of 48.3 ms outline the worst scenario of all.

DLSS 4.5, Some Tests with 007 First Light

The performance test conducted on the same system with 007 First Light, at a resolution of 2560x1600 pixels with Ultra graphical preset, shows that our GeForce RTX 5090 Laptop can push native rendering of the game to an average of 64.7 FPS, with a 1% Low of 51.4 FPS. This scenario guarantees playability, but it is certainly not ideal and especially latency reaches 53.1 ms, the highest registered in the entire test.

Activating Super Resolution in Quality mode raised the average frame rate to 76.9 FPS. A stability improvement was noted with a 1% Low of 57.1 FPS and a marked reduction in latency, dropping to 43.2 ms, making the action more responsive to user input.

Introducing Dynamic Frame Generation produces a remarkable performance leap, nearly tripling the average frames to 232.1 FPS. The 1% Low rose to 139.4 FPS, ensuring extreme smoothness, while PC latency saw a vertical drop to only 29.3 ms.

With 2X, 3X, up to 6X settings, we observed the following trends:

2X: Average of 135.4, the lowest latency in the entire suite at 26.2 ms.
3X: Average rises to 188.6 FPS, with 1% Low at 119 FPS and latency of 27.8 ms.
4X: Achieves 234.4 FPS. The 1% Low reaches 143.3 FPS, maintaining latency at a very reasonable 28.8 ms.
5X: The system hits an average of 272.3 FPS with a 1% Low of 163.5 FPS, with latency remaining stable under thirty milliseconds at 28.6 ms.
6X: Represents the performance peak with an average of 309.8 FPS and a 1% Low of 183.8 FPS. Despite the high number of generated frames, system latency stabilizes at 29.4 ms.

The analysis highlights that the implementation of frame generation technologies in this title not only increases visual smoothness but also drastically optimizes system responsiveness.

Moving from 53.1 ms native down to about 26-29 ms during Frame Generation sessions effectively halves input lag, an exceptional result for a notebook.

Given the refresh rate of 240 Hz of the Razer Blade 16, we see a scenario similar to that observed in Cyberpunk 2077, where the test with Multi Frame Generation 6X stands as the best session for those wanting to leverage every single hertz of their monitor. With an average of 309.8 FPS and a 1% Low of 183.8 FPS, it offers the highest possible smoothness without sacrificing latency (29.4 ms).

In this case, the session with Dynamic Frame Generation records performance similar to 4X (232.1 FPS) but with a dynamic management that ensures excellent consistency of minimum frames.

And the Graphics? Other DLSS 4.5 Features

There are also other news concerning DLSS 4.5, specifically improving the graphics for GeForce RTX 4000 and 5000 players by selecting specific presets in NVIDIA App.

Using Cyberpunk 2077, we set the Preset L for Super Resolution to verify if we actually achieve better quality with the Ultra Performance upscaling setting, and selected Preset B for Frame Generation. The latter enhances user interfaces in some titles, leveraging additional data in the game engine. Recently, NVIDIA also announced the arrival of a second-generation Transformer model for Ray Reconstruction in August.

As we are now discussing specifics and details, we closely observed the matte sign bearing the word “GARAGE.” With DLSS 4, the edges of the white letters show a smooth transition and some blurriness along the curved profiles (especially on the letter “R”) and on the oblique vectors of the “A.” This behavior tends to soften the overall image.

Left: DLSS 4.5, Right: DLSS 4

In contrast, moving to version 4.5 highlights a clear optimization in edge reconstruction: geometric profiles are markedly more linear, sharp, and free of aliasing artifacts, precisely defining the chromatic separation between the white characters and the background.

A second indicator of the greater efficiency of the updated neural network can be found in the retention of microscopic texture detail. The background panel of the sign showcases a raw and porous metallic surface. With DLSS 4, such granularity undergoes a micro-smoothing that evens out the surface, misinterpreting high-frequency detail as background noise. The sampling performed by DLSS 4.5, instead, allows for the preservation of the original porosities and imperfections of the material, delivering a greater material richness without introducing flickering or artificial post-process sharpening artifacts.

Finally, we observe similar findings in the stability of the 2D elements of the user interface (HUD). The yellow informational texts placed to the right of the screen and the mini-map paths show a higher definition level with DLSS 4.5. The individual glyphs maintain high local contrast, avoiding visual cross-talking phenomenon (blurring of edge pixels) noticeable in the captured frame with the earlier technology version.

This second image also allows us to isolate the improvements brought by the latest revision of the DLSS technology. Observing the metallic tank, the major distinction element is represented by the rendering of the graffiti. Under DLSS 4.5, the outlines of the writings have superior definition and more pronounced local contrast. In contrast, the DLSS 4 implementation shows a tendency for pixel dispersion along the margins of thin lines, resulting in a more “softened” and less incisive visual perception. This difference suggests refinement in temporal reconstruction management, which in version 4.5 is more effectively able to separate applied level graphic information (decal) from the base texture.

Left: DLSS 4.5, Right: DLSS 4

Surface texture analysis confirms this trend. Rust and degraded cement textures benefit in version 4.5 from preservation of the original grain, which in DLSS 4 appears partially subdued by more aggressive filtering. The stability of the HUD elements remains a strength of the new release, ensuring text and icons with sharp edges and no chromatic blurring compared to the previous version.

Conclusion

Overall, the tests conducted with DLSS 4.5 and the new Multi Frame Generation modes highlight a further step forward in NVIDIA's strategy to increasingly separate traditional rendering from the final perception of the gaming experience. The increase in achievable frame rate, especially under heavier loads with active ray tracing and path tracing, is evident and allows for attaining fluidity levels that native rendering alone could not ensure on high-end mobile hardware.

At the same time, the data show that the benefits are not exclusively tied to average FPS values but also to the overall stability of frame pacing, which in the analyzed tests remains generally solid even at higher multipliers. Latency has improved compared to native rendering.

However, the role of the usage context remains central: the effectiveness of the more advanced modes, such as the higher multipliers of Dynamic Frame Generation, is closely related to the presence of high refresh rate displays and careful balancing of image quality, perceived latency, and performance goals. In the absence of these conditions, the advantages may prove less significant or coherent.

On a qualitative level, the observed differences in image reconstruction indicate further refinement of upscaling and temporal reconstruction algorithms, with better management of fine details and greater stability of high-contrast elements, without introducing revolutionary changes over the previous generation.

Ultimately, DLSS 4.5 represents an effective incremental update, capable of elevating overall frame stability and readability of minute details, reducing the visual gap from native resolution.