One of the new features NVIDIA announced alongside its new RTX 50 series GPUs is the latest iteration of DLSS, now in its fourth major revision and enables Multi Frame Generation exclusively for the new Blackwell-powered cards.
While there’s more updates to existing features like Super Resolution and Ray Reconstruction, in this article we’ll focus on DLSS 4’s Multi Frame Generation (MFG) component via Cyberpunk 2077, which is one of the 100-plus games featuring MFG currently.
How Multi Frame Generation Works

As detailed NVIDIA’s explainer, DLSS 3’s traditional frame generation technique relies on a hardware-based optical flow accelerator which can’t scale beyond one generated frame per one rendered frame – doing so will cause the native framerate to drop, which isn’t ideal in terms of input latency.

Instead, DLSS 4’s Multi Frame Generation runs on an AI model instead of a traditional hardware accelerator to accelerate the optical flow process, which NVIDIA says uses less VRAM and reduces overall computational cost. This enables the compatible hardware – in this case, the RTX 50 series GPUs – to generate more than one frame per rendered frame, resulting in higher perceived framerates.
To provide enough compute power to do all the AI computations ranging from Super Resolution, Multi Frame Generation and Ray Reconstruction all within the span of milliseconds, the tensor cores onboard these Blackwell GPUs are significantly beefed up; besides that, the generated frames are also delivered to the display using Flip Metering on the GPU itself instead of CPU-based frame pacing to maintain greater timing consistency (less stuttering).
Enabling DLSS 4 MFG


To enable MFG, there are some extra steps required from the user to perform outside the game: in the case of Cyberpunk 2077, you must first enable Frame Generation feature in-game, then head to NVIDIA App settings to enable a setting called “DLSS Override – Frame Generation” and select the appropriate multiplier (up to 4x). The game will follow this setting with the in-game benchmark displaying boosted framerates accordingly.
Test System
CPU | Intel Core i9-13900K |
Cooling | Cooler Master MasterLiquid PL360 Flux 30th Anniversary Edition Thermal Grizzly Kryonaut |
Motherboard | ASUS ROG Maximus Z790 Apex |
GPU | NVIDIA GeForce RTX 5090 Founders Edition |
Memory | Kingston FURY BEAST RGB DDR5-6800 CL34 (2x16GB) *configured to DDR5-6400 CL32 XMP profile |
Storage | ADATA LEGEND 960 MAX 1TB |
Power Supply | Cooler Master MWE Gold 1250 V2 Full Modular (ATX12V 2.52) 1250W |
Case | VECTOR Bench Case (Open-air chassis) |
Operating System | Windows 11 Home 24H2 |
Quality Comparisons
For the visual quality comparisons, we’ll be using the built-in benchmark to maintain consistent scene for easy comparison, and we use 1440p RT Overdrive preset as the baseline, while disabling Super Resolution to eliminate any potential visual variances beyond frame generation itself. We then use NVIDIA’s built-in game recording feature to record the scene in each benchmark runs, and all footage are taken into NVIDIA’s Image Comparison & Analysis Tool (ICAT) for side-by-side comparison.
Frame 1


For the first example we take a look at the bottle placed on the pool table located at the lower-left edge of the screen. Given the principle of perspective distortion (where objects at the edge of the camera moves faster than the objects in the center relative to the camera motion), the bottle shifts faster in frame, meaning it’ll be more susceptible to artifacts in the case of frame generation.
Still, despite that, the artifact is relatively minor despite the relatively low input resolution (~60FPS), where in the case of 3x and 4x MFG you’ll see a slightly more pronounced ghosting near the left side of the bottle. 2x FG sees the same ghosting effect but to a lesser extent, and if you’re putting this in full speed, it’s hardly noticeable to an untrained eye (this pretty much applies to most examples below, unless you’re trying very hard to pixel peep the images).
Frame 2


The worst-case scenario we can find from Cyberpunk 2077’s built-in benchmark scene involves a metal grille, where the artifacts of frame generation are very much visible in still images, and somewhat noticeable in motion. As a side note, the framerate counter on the top left gives you a rough idea what are the framerates you’ll be getting in comparison to native rendering.
Frame 3


Not every fast motion will introduce visual artifacts like the previous two images, however. In this third example, we’re looking at the objects on the pool table at the bottom of the screen – in all four frame generation settings, we see no difference in visual quality.
Frame 4


For the fourth example we’re comparing a slightly more complex object, specifically the barbed wire on the fence around the center-top of the frame. Moving from native to 2x, the difference between quality is mostly minor; however, the barb wires in 3x MFG and 4x MFG gets increasingly fuzzier, with the latter producing visible ghosting on the wires. As for the fence itself, it’s well handled in all scenarios with no observable artifacts in sight.
Frame 5


Finally, we look at the trunk of the tree located at the center of the image, where we see varying degrees of ghosting depending on the frame generation settings. In the case of 2x FG, the ghosting effect is extremely minor, only appearing on the top half of the trunk with a very faintly visible line, whereas 3x MFG and 4x MFG shows some level of fuzziness on the right edge of the trunk.
Verdict

So, is Multi Frame Generation any good? Our verdict is, it depends on how much you’re willing to trade visual quality in exchange for framerates (not necessarily the latency). Generally, 2x FG does a well enough job at keeping the artifacts to a minimum while still giving framerates significant boost, and there’s not much point in going for 3x or even 4x boost if you’re already targeting 60FPS as the native framerate, unless you really want to max out a 240Hz monitor while maxing out all the graphics settings in graphics-heavy games like CP2077 or Black Myth: Wukong.
It’s worth keeping in mind that the extra boost in framerates is merely visually perceptual – your PC will receive inputs as much as it can render a native frame, so any level of frame generation boost will not improve input latency whatsoever (it’s also why NVIDIA Reflex must be turned on when FG/MFG is active).
Ultimately, if you want a framerate boost, start from Super Resolution first; if that’s not enough, I suggest you to try from 2x and go from there – if you’re willing to prioritize smoother motion, go ahead and boost it to 3x or even 4x; otherwise, 2x is still a good balance between smoothness and visual quality.