Similar to how FreeSync in a monitor works, a chip could be made to function in the same capacity on a GPU. USB-C can be used to facilitate the passing of generated frames. At present, USB4 Version 2.0 (published in October 2022) can handle up to 80GBPS. I belive NVidia's latest version of NVLink in 2022 is capable of 120GBPS, however, when passing already rendered frames, 80GBPS should be enough for the current/next generation of GPUs.
The CPU gets the frames to generate, it passes each alternating frame to each GPU.
Going off trace length, from CPU to GPU, then GPU to output, it would be best to use the 2nd GPU (if 2 GPU’s, use 2nd as Primary) as the output device. This way we alleviate as much latent problems as possible.
The Radeon software would see that 2 GPU’s are present, and would disable frames to pass from the “?” to the other outputs on the 1st GPU (Secondary). Instead, routing them through a Type-C connection, where the frames inter the 2nd GPU.
The 2nd GPU (Primary) would see that it’s connected to both “?” on 1st GPU, and Radeon software would accept incoming frames. It would interlace the frames to the in use Primary output device.
The “?” would work similar to FreeSync. It would take each alternating frame, check a quick time stamp, and interlace frames from 1st and 2nd GPU, to primary output.
The “?” is more or less a dumb device, that would work as pass through or interlace and nothing more. The output on 1st GPU would be disabled, as the “?” reroutes all frames generated, out the Type C connection, to the 2nd GPU to the “?” there, where all the frames are interlaced.
In this configuration, the dropped or discarded frames should be mitigated to around 5%-10% (guessing) allowing for high refresh rate 4K, 8K and beyond, gaming.
It would also allow for 2x Ray Tracing, allow lower end GPU owners to nearly double their FPS w/ the add in of a 2nd GPU, possibly double workload task and or video encoding for streamers.
The big crutch here, is AMD (RTG) would have to take initiative in NOT making it up to game dev's to have to enable CrossFire, RTG would have to make their software dictate that 2x matching GPU's are present, and alternate frame generation out to each.
This is all just a theory. I guess feel free to tear appart my theory in the comments.
