Hands-on with Nvidia's new Dynamic Multi Frame Generation: 5x and 6x modes push frame rates even higher than before, though you can have too much of a good thing

Earlier this month, Nvidia announced that RTX 50-series owners would soon be able to use an improved version of Multi Frame Generation (MFG) in games, one that could dynamically switch between modes, including a new 6x option. After our first glimpse of it at the CES show in January, we've now had a chance to test it all out ourselves.

As a very quick recap, DLSS MFG works by having the graphics card render two frames normally, but keeping them both in VRAM. Then, through the power of AI, the GPU interpolates at least one frame that effectively slots in between the two (giving you 2x frame gen mode). Once that's been generated, all three frames get displayed in sequence, and the whole process repeats itself in the background.

The multi part in MFG refers to the fact that the generative stage can generate two frames (3x mode), three frames (4x mode), and with this latest update, five frames for 6x mode. Before you ask, no, there isn't an option to force a 5x override mode, even though Nvidia's MFG supports it.

Anyway, alongside the fixed override options is a new setting that lets DLSS MFG figure out what mode is best to use, based on your monitor's maximum refresh rate and the performance of the game—i.e. Dynamic Multi Frame Generation. For example, if you have a 240 Hz display, DLSS will switch between the various modes to keep the frame rate as close to 240 fps as possible.

The new version of DLSS MFG also includes an updated AI model (aka Preset B) that "enhances in-game user interfaces by incorporating additional game engine data, improving visual quality and clarity of static user interface elements."

However, Nvidia notes that the new model "can only provide a benefit to games which expose a UI depth buffer, so it won’t work on all games, and not all supported games show significant improvement."

Two examples of games that do support it are Hogwarts Legacy and Dragon Age: The Veilguard, though there are only 20 games in total that you can use Preset B with; everything else will just use the standard 'Preset A' model.

Anyway, none of this matters if the dynamic system doesn't work as intended or if the new model actually makes things worse. So let's get on and see it in action with a Ryzen 9 9950X3D and GeForce RTX 5090 combination, and an MSI MPG 321URX 240 Hz OLED monitor for handling the display duties.

Cyberpunk 2077

RT Overdrive | 4K DLSS Performance | No frame gen

To begin with, let's start with a 'ground truth' run of Cyberpunk 2077 at 4K, with RT Overdrive and DLSS Performance enabled (DLSS Ray Reconstruction disabled), to see what kind of frame rates we get. Nvidia Reflex is also disabled to get a sense of the baseline input lag.

I've used an updated version of Nvidia's Frameview (top left), along with a similarly updated Nvidia App statistics overlay (top right), to show accurate real-time performance figures, as well as information about the use of frame generation and the overall system latency (PCL).

As you can see, although the PCL figure is nice and low (so no discernible input lag), and the overall frame rate is around the 60 fps mark, the game comes across as being a little janky. This isn't the video; the game really does look that way in real life, because there's nothing other than the game controlling frame pacing (i.e. when frames are timed for display).

RT Overdrive | 4K DLSS Performance | In-game 4x FG

Now let's see it again, but this time with 4x Multi Frame Generation enabled in-game. You could use the Nvidia app to override it to 6x (FG needs to be activated for any override to work), but for now, let's just stick with 4x.

Cyberpunk 2077 runs a lot smoother with this level of frame generation, because the use of frame gen enables Reflex (which gets a better handle on the frame pacing), but unfortunately, the PCL is also a lot higher, and swinging the camera around feels a little sluggish. Not massively so, and certainly not enough to make the game unplayable, but it's certainly noticeable.

RT Overdrive | 4K DLSS Performance | Dynamic MFG override

The above video shows another test run, but this time with Dynamic Multi Frame Generation enabled. Since native 4x mode couldn't achieve a constant 240 fps (it couldn't reach it full stop), it's no surprise to see that DFMG switches to 5x mode for a good portion of the test run. However, it's quite happy to drop down to 4x—240 fps is a target, not a hard restriction.

However, the slight increase in noticeable input lag with the in-game 4x frame generation is more apparent when DFMG switches to 5x, and you can see this clearly with the rise in the PCL to around 50 or so milliseconds. Again, it's not a game-breaking issue, and it's something that simply cannot be avoided with frame gen.

RT Overdrive | 4K DLSS Performance | 6x FG override

Since Dynamic MFG didn't need to switch to the new 6x mode, I did one more run with it enabled as a fixed override. It's interesting to note that the PCL isn't much worse than with DMFG's 4x/5x modes, and the average frame rate is even higher. However, the stuttery feel to Cyberpunk 2077 without frame gen makes a reappearance here, and that's possibly down to the sheer number of frames that now have to be paced correctly, or just something about my set up that doesn't like this level of frame generation.

Dragon Age: The Veilguard

Ultra | 4K DLSS Performance | No frame gen

Moving on to Dragon Age: The Veilguard, we start once more with a standard run at 4K Ultra DLSS Performance and no frame generation. This game runs pretty well without the aid of generated frames, though the average frame rate doesn't get anywhere 240 fps.

However, the system latency is a tad high for something averaging 130 frames per second, but DLSS is well implemented in Veilguard and enabling 2x frame gen in the game's settings produces an unexpected outcome: The PCL figure is lower.

Ultra | 4K DLSS Performance | In-game 2x FG

The reason for this is almost certainly down to the fact that the use of frame gen requires Nvidia Reflex to be enabled (which isn't for any of the baseline no-FG videos I've created). This system gets rid of the frame queue so that the CPU only prepares and issues a rendering command sequence when the GPU is ready for it.

If the GPU is quite busy trying to churn out a barrage of frames from the CPU, this synchronisation of CPU and GPU results in a lower PCL.

Frame generation in 2x mode produces an overall frame rate pretty close to 240 fps, which is why the use of Dynamic MFG doesn't change things. It stays in 2x throughout the test, giving you the input latency and performance that you need. This is a good thing, because when I first learned about DMFG, I was a little concerned that the system would base when to switch modes on a minimum frame rate.

It's clearly not doing that and is quite 'relaxed' about having the fps fall a little behind the maximum refresh rate.

Ultra | 4K DLSS Performance | Dynamic MFG override

One thing I found in my testing is that setting the DLSS Frame Generation model override to Recommended didn't result in Preset B being enabled, despite Nvidia saying that the game supported it. This model is supposed to help UI elements look better, but they appear completely fine with Preset A anyway.

So I tried the final run once more, but this time with Preset B selected in the override options, and saw absolutely no difference whatsoever: Not in performance, not in system latency, not in the visual fidelity of the UI. The same thing occurs with the next game I tested, too.

Hogwart's Legacy

Ultra | 4K DLSS Performance | No FG

When it first launched, Hogwarts Legacy rapidly garnered a reputation for running like a bag of spanners being dragged over a cobblestone road. It's thankfully an awful lot better these days, but when set to Ultra graphics with ray tracing enabled, the frame rate can still jump about all over the place, especially when you transition from being inside a building out into the open world.

For my test runs, I picked an area of Hogwarts Castle that starts quite simple for the Ryzen 9 9950X3D and RTX 5090 to handle, but then goes through a spot where there is a lot of ray-traced reflections and a whole host of NPCs, which neatly slices a fair chunk of the frame rate. As such, this should be a good exercise for Dynamic MFG to handle.

Ultra | 4K DLSS Performance | In-game 4x FG

Just like Cyberpunk 2077, Hogwarts Legacy natively supports up to 4x Multi Frame Generation, but while it does a great job of lifting the overall performance to a ridiculously high level, the fixed mode makes the heavy area feel a touch janky. If you watch the CPU and GPU utilisation figures carefully, you can see that the drop in GPU usage isn't as bad as before, but it's still quite high.

Ultra | 4K DLSS Performance | Dynamic MFG override

The use of DMFG doesn't eliminate this issue, though it does tame it down a touch, but the main benefit here is that since the frame gen mode never exceeds 3x, the PCL figure is better than with the fixed 4x mode.

I repeated the previous test with a fixed 3x mode and naturally got the same PCL, but it was always at that level. Using Dynamic MFG allows the graphics card to run in 2x mode when it's rendering fast enough, thus giving you a lower system latency.

The Elder Scrolls 4: Oblivion Remastered

Ultra | 4K DLSS Performance | No FG

The last game I tested Nvidia's new Dynamic Multi Frame Generation with was Oblivion Remastered, something that desperately needs as much help as possible to run well. Without Reflex enabled, the system latency is pretty awful (over 50 milliseconds), even though the average frame rate is fine.

Oblivion Remastered doesn't natively support HDR, but you can force it on via the game's config files. However, I haven't quite got the settings right for my monitor, which is why the videos look a touch washed-out compared to those for the other tested games.

Ultra | 4K DLSS Performance | Dynamic MFG override

Alas, where the use of frame gen and Reflex makes a big difference in Dragon Age: The Veilguard, it does little to improve how the remastered game feels. The PCL is lower, and the average frame rate is a lot higher, but the 1% low fps figure still lurks in the disappointing zone. It's only when you're basically looking at nothing but rocks and grass that things pick up, but that's hardly praise.

Before testing DMFG, I briefly checked out the fixed 4x and 6x modes. Neither made a wealth of differences (especially the latter), so once I completed a Dynamic Multi Frame Generation run, the result wasn't surprising. Switching between 3x and 4x, the system gets the required average frame rate, but it's just not enough to overcome the inherent jankiness of the whole game.

Dynamic MFG: The Verdict

So, what to make of Nvidia's update to Multi Frame Generation? Well, it clearly works as intended, and any concerns you might have over how switching modes could affect gameplay don't appear to be an issue: It's practically instantaneous. But that doesn't mean it's something you should have enabled for every game, all the time.

That's because there are more costs to using frame generation than just an increased input latency, and the AI interpolation algorithm can't work from thin air and wishful thinking. Each generated frame requires a fair chunk of calculations, plus a smattering of extra VRAM to store the two pre-rendered frames and the extra AI ones.

What you can't do is enable Dynamic Multi Frame Generation on a graphics card like an RTX 5060 and expect your games to now happily run at 4K with path tracing. I tried to record footage of a GeForce RTX 5070 running Cyberpunk 2077, using the same settings as I did for the RTX 5090 and DMFG, but it ran so badly that the video stream collapsed after a few seconds.

RTX 5070 | RT Overdrive | 1080p DLSS Performance | Dynamic MFG override

Dropping the resolution to 1080p (but still using RT Overdrive and DLSS Performance) solved that problem, but as you can see in the video capture, it doesn't come across as being particularly smooth, even though the frame rates and PCL figures are generally fine (though the latter is a bit too high in places).

Frame generation can't fix performance issues inherent to a given gaming PC. It's best to think of it as being something that can lift 'decent' into the realms of 'great', e.g. a consistently smooth 60 fps is more than playable, so interpolating it up to 120 fps or higher shouldn't cause too many problems.

However, I do wonder a little just who Dynamic MFG is really for. Does it really matter that a game runs at your monitor's refresh rate when we have systems like G-Sync and FreeSync to remove screen tear? Yes, it can improve frame pacing for smoother gameplay, but that's mostly thanks to Reflex, anyway.

Few competitive gamers are going to use frame generation, let alone DMFG, to get super-high frame rates because of the increased input latency. Sure, on a high-end gaming PC, it's only a small increase, but esports shooter pros do everything they can to reduce latency, not increase it.

PC gamers with a budget or mainstream setup, using a 144 Hz 1080p or 1440p monitor, might be tempted to try it out, but if they can already get 60 fps without frame generation, then DMFG is only going to use 2x mode for most of the time, perhaps 3x in some cases. Given that Nvidia's new system requires a game to have frame gen in the first place, you might just prefer to use that.

Having said that, Dynamic Multi Frame Generation is entirely optional, and it has to be employed on a per-game basis. You don't have to use 6x mode; you can set the cap to something much lower. In other words, you can choose to use it only where it gives you genuinely better performance and a nicer gaming experience. In games that don't do this, you can just ignore it.

You might not be interested in using the new feature, but if you have a GeForce RTX 50-series graphics card, you've now got something else to play around with for free, and there's nothing wrong with that.