Cyberpunk 2077 gets triple the framerate using an unholy combination of Nvidia and AMD tech — Nvidia DLSS Frame Generation plus AMD Fluid Motion Frames delivers impressive benchmark results, but less than ideal real-world utility
There are some serious caveats, naturally.
Benchmarks from QuasarZone show that a near tripling in performance can be achieved in some games by using both DLSS Frame Generation combined with AMD's Fluid Motion Frames. This is possible thanks to a multi-GPU solution that allows an Nvidia 40-series GPU to render using DLSS Frame Generation and then have an AMD GPU to take that output and generate even more frames via Fluid Motion Frames. So pairing two of the best graphics cards for maximum faked frames is possible. Practical? That's a different story, and there are quite a few caveats to this dual Nvidia-AMD setup that makes it unlikely to be useful in practice.
How QuasarZone got this working is a little tedious. It installed both an RTX 4090 and an RX 6600 into its test system, installed drivers for both, and then had the primary display hooked into the 6600. Then in Windows, it selected the 4090 as the rendering GPU, which allowed the 4090's DLSS 3 Frame Generation content to pass through the 6600, which could then allow AFMF (AMD Fluid Motion Frames) to generate even more frames.
In Cyberpunk 2077 and Starfield, the performance boost with both DLSS 3 and AFMF enabled was nearly 3X. Starfield's results were achieved even without DLSS Frame Generation, which isn't supported in the game. Call of Duty: Modern Warfare III and Ratchet & Clank: Rift Apart saw roughly a doubling in performance. It certainly sounds impressive on a surface level. However, this solution isn't exactly without problems.
First, it requires having two GPUs, one from Nvidia and one from AMD. Thankfully, even an RX 6600 is sufficient for the AFMF aspect, but that's still an extra $180, more space taken up, and more heat being generated.
More critically, the performance claims require a bit of an asterisk. Although combining DLSS and AFMF resulted in a big increase to the average framerate, the 1% low framerate either didn't budge much or went slightly down. Ideally, the 1% low framerate or 99th percentile framerate will be as close to the average FPS as possible. Increasing the average FPS while decreasing the 1% low FPS could result in a choppier and less smooth experience.
Finally, there's ample opportunity for visual quality to take a big hit. QuasarZone didn't show any screenshots or videos in its testing, but just thinking logically about what's going on gives plenty of cause for concern. We know that both frame generation technologies don't leave visuals 100% intact, and combining two of them seems like a recipe for disaster. Plus, user input latency should be atrocious. The 4090 would sample user input once for every two frames, and then AFMF would try to double down on that, meaning one input sample would be used for every four frames that get sent to the display.
For all of these reasons, it's very unlikely we'll see gamers picking up two GPUs to see a potentially big boost in performance. It's almost like CrossFire and SLI, which were also multi-GPU technologies that didn't exactly work well. Of course, CrossFire and SLI at least were officially supported; combining DLSS and AFMF isn't and almost certainly never will be.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Matthew Connatser is a freelancing writer for Tom's Hardware US. He writes articles about CPUs, GPUs, SSDs, and computers in general.
-
bit_user We know that both frame generation technologies don't leave visuals 100% intact, and combining two of them seems like a recipe for disaster.
I'd expect more artifacts than using just DLSS Frame Generation, but it's probably not much worse. The reason is that the differences between frames decreases as you increase the frame rate. The more similar they are, the easier it is to interpolate between them. So, if your base framerate is high, then the technologies probably both work pretty well.
Plus, user input latency should be atrocious.
Again, I'm going to call this into question. Both technologies add delays in terms of a couple frames. The higher the input frame rate, the less delay (in milliseconds) they'll add.
I'm not saying I'd like to use such a setup, but if your game is playable with either one, having both probably isn't much worse. -
vanadiel007 I am thinking the key is to generate so many frames that any abnormalities will not be noticeable to the human eye. Like +500 FPS, there's no way your brain will be able to detect inconsistencies between frames.Reply -
atomicWAR
Glad you brought this up. My big question is on the input latency. From what I've read of both techs DLSS 3 has a lot less latency compared to fluid motion frames from AMD. I've heard AMD implementation is really rough on said latency (some reviews were more positive to be fair and I have yet to try fluid motion frames ATP to test it myself). I'd be curious to see input latency tested with both solutions active. If its low enough....I could see myself trying to do the same in combining both techs for game play.bit_user said:Again, I'm going to call this into question. Both technologies add delays in terms of a couple frames. The higher the input frame rate, the less delay (in milliseconds) they'll add.
I'm not saying I'd like to use such a setup, but if your game is playable with either one, having both probably isn't much worse. -
bit_user
That's true if the anomalies aren't consistent from frame-to-frame and are somewhat evenly distributed. However, this usually isn't the case. Frame interpolation algorithms tend to make systematic errors on certain kinds of content.vanadiel007 said:I am thinking the key is to generate so many frames that any abnormalities will not be noticeable to the human eye. Like +500 FPS, there's no way your brain will be able to detect inconsistencies between frames.
For instance, my TV has a built-in motion smoother that's normally very good. However, there are some cases it consistently struggles with, for instance a person walking in front of a regularly-patterned background, such as a brick wall. There's a sort of halo around the person, where the bricks appear jumbled up and jumping around. It sticks out like a sore thumb. Yet, I leave motion smoothing on the maximum setting, because the effect normally delivers such a big improvement in how clearly camera pans or moving objects appear. -
bit_user
Other than reading through the details on DLSS Frame Generation, when it first launched, I haven't followed either tech.atomicWAR said:From what I've read of both techs DLSS 3 has a lot less latency compared to fluid motion frames from AMD. I've heard AMD implementation is really rough on said latency (some reviews were more positive to be fair and I have yet to try fluid motion frames ATP to test it myself).
I do like motion smoothing so much that I would enable it on my TV, for a couple games that didn't require fast-twitch reaction times, on my PS3. If I played PC games and had a card that supported it, I'd probably try pretty hard to use it, unless it was just untenable. The latency on my TV was really bad, since I couldn't use its "Game Mode" if I wanted motion smoothing. Game Mode circumvents other processing the TV does that adds latency, so that other processing latency stacked on top of what the motion smoother itself added. Yet still I enabled it, because I'm such a framerate junky. -
Colif it might look good but the reaction time wouldn't match. Both of them add latency so it depends, do you want a really smooth movie or do you want to actually play it?Reply -
bit_user
Allow me to use a numerical example, to make my case.Colif said:it might look good but the reaction time wouldn't match. Both of them add latency so it depends,
Let's say each doubles the frame rate, but adds a frame worth of latency, at the input frame rate. If you start with 30 fps, then the first stage adds 1/30th of a second latency (i.e. 33.3 ms) and outputs 60 fps. Then, the second stage adds a further 1/60th of a second of latency (i.e. 16.7 ms) and outputs 120 fps. Now, we're up to 50 ms of latency. Not great, but better than the 66.7 ms you might have assumed.
Note that this was only hypothetical, but I think it's a fair characterization of how stacking would impact latency.
I should add that stacking should be completely unnecessary. If you're doing frame interpolation, then you should be able to interpolate more than one frame. There might be practical reasons why the hardware lacks the resources to do it, but there's no logical reason why you shouldn't be able to just have DLSS interpolate a 30 fps stream to 120 fps. In that case, you'd avoid the second-stage latency altogether. -
Colif Fluid Motion Frames is so new I haven't got any games that would use it. Its not in the Adrenaline drivers yet unless you get a beta pack. So I can't comment on either of their effects.Reply -
bigpinkdragon286 Fluid Motion Frames doesn't require support be built into the game. Been using it since the first beta was released bringing support for 6 series cards. It pretty much tripled the frame rate in Alan Wake 2 from 60 to 180, with obviously expected ups and down depending on the scene, but it's clearly working as intended.Reply
The driver has been stable enough I haven't bothered upgrading from the 2nd beta to the third yet.Colif said:Fluid Motion Frames is so new I haven't got any games that would use it. Its not in the Adrenaline drivers yet unless you get a beta pack. So I can't comment on either of their effects. -
Order 66 I’ve used it and from what I can tell, it is pretty good, but I have also never used any frame generation tech before. The input lag increase is substantial, but I think that is to be expected. For games that don’t require fast input, it works very well.Reply