Artificial frame generation technology. Even typing those words feels vaguely science-fiction, but ever since the release of Nvidia’s DLSS 3 Frame Generation (and much later, AMD’s Fluid Motion Frames), it’s become a concept that’s gone beyond the theoretical into the everyday. However, Korean technology website QuasarZone (via Videocardz) has taken things to a new extreme by managing to run both methods at the same time via an impressive bit of GPU wrangling, and the performance gains are, on the surface at least, rather impressive.
To achieve this mind-bending result, both an RTX 4090 and Radeon RX 6600 were installed on the same motherboard, with the monitor connected to the AMD GPU. The RTX 4090 was then forced to render a game with DLSS 3 Frame Generation enabled, while the RX 6600 was used as an output source to the monitor using AMD Fluid Motion Frames. This means the eventual output was rendered at a higher frame rate using DLSS 3 Frame Generation, then interpolated once more through the RX 6600 to even higher levels still.
The results at first glance seem very good indeed. Cyberpunk 2077 at a 4K output resolution jumped from an average of 71.7 fps all the way up to a gigantic 209.3 fps with both DLSS 3 FG and AFMF enabled. That’s nearly triple the frames, and quite an impressive demonstration, as proof of concepts go.
It wasn’t just Cyberpunk either. Call of Duty: Modern Warfare III nearly doubled its frames using the technique, with a leap from just under 128 fps all the way to nearly 226 fps, while Ratchet and Clank: Rift Apart managed 233 fps with both methods enabled, up from an average native frame rate of just over 121 fps. Even the notoriously CPU-bound Starfield received a significant performance jump, from 72 fps at native to just over 200 fps with DLSS quality mode and AFMF working together.
The fact that this even works at all is very impressive, but as you might expect, there are some caveats. First of all, beta drivers are currently required to enable AFMF on any AMD card, and some games reportedly didn’t recognise the full-screen mode required for Fluid Motion Frames to work correctly.
Secondly, sticking two GPUs next to each other and making them work hard in conjunction means that power and heat are factors to be considered, so a motherboard was used that allowed them to sit side by side while still receiving enough space for adequate cooling.
And the third? 1% lows were an issue, and in all the games tested they remained very low compared to the lofty heights achieved in average framerate alone. This was caused by AFMF turning itself off dynamically under fast motion, as under these conditions it doesn’t have enough motion vectors to determine the direction of the changes to the generated frames, which would result in a very messy output if it stayed on at all times.
As such, with such large jumps and drops in the framerate the overall smoothness of the experience was less than ideal, meaning it’s unlikely anyone’s going to be using this as a genuine way of drastically improving the frame rate on their machines at home.
Nevertheless, credit has to be given not just for the conceptual idea, but the implementation of a technique that on the surface doesn’t seem like it would work at all, nevermind with the results achieved. While it’s an excellent demonstration of a bit of lateral thinking to make two distinct frame generation technologies work in relative harmony, perhaps it’s better to leave this one as an experiment rather than something to try yourself.
