News AMD releases FSR 3 source code for DirectX 12 and Unreal Engine 5 game developers, aiming to increase interest and adoption

Status
Not open for further replies.
This has a lot of potential. We might well see some more mods where you enable both AMD and Nvidia's frame gen tech at the same time (like we have already seen). At the very least maybe we can see modders/other corps lower the latency on AMDs solution or further it at least as I heard they made great strides with the new Avatar game. I have it and while I haven't used FSR3 with frame gen, honestly with a 4090 @4K it frequently sits at the 96-128FPS range. I only got it last night to play with the wife and haven't tested it thoroughly yet...she is gone for the weekend so I might well tear into some benchies.
 
  • Like
Reactions: ivan_vy
Smart move. Now the community can do the work to make it bettter
100%. Just look at what modders have already done (ie jerry-rigged it so both frame gens techs to work together). Peeps on 160+hz might finally use all those frames lol. But the real crown would be to lower the latency AMDs fluid motion frames gives you. Do that and it might 'effectively' kill off Nvidia's proprietary solution.
 
  • Like
Reactions: ivan_vy
This all well and good, but I’m still waiting on AMD to finally allow a 4x super resolution setting so I can play older games at 4x super sample. This has been the only thing I miss since switching from nvidia to AMD…being able to super sample from 3440x1440 to 6880x2880. Right now AMD limits me to 5120x2160 for some arbitrary reason
 
upscaling and frame generation will do wonders for PS3 emulation.
It seems unlikely that it would help much. From what I understand, PS3 emulation is CPU-limited more than anything. It shouldn't take all that much graphics hardware to render 10+ year old games that were originally designed to run on a graphics chip mostly equivalent to a (circa 2005) Nvidia 7800 GTX, even at much higher render resolutions. The real performance limitations come from trying to emulate the PS3's 8-core IBM Cell processor on other CPUs, and upscaling isn't going to help with that.

And FSR upscaling was already added to the leading open-source PS3 emulator well over 2 years ago. At the time, the developers stated that they would not be supporting DLSS, because it required access to motion vectors, something their emulator (and perhaps PS3 games in general) didn't use. The main advantage of FSR2+ is that it is also able to utilize motion vectors for better results, but much like DLSS, if the emulator doesn't have access to that, it likely won't be implemented. FSR3 Frame Generation also utilizes motion vectors. It's technically possible to implement frame generation without access to that, but the results likely won't be particularly good.

Even if it were possible to implement frame generation in a PS3 emulator, that feature is usually best reserved for games that already run well, to make them look a bit smoother on high refresh (120+ Hz) displays, rather than something to fix performance issues with games running at low framerates. Since the games are still effectively running at or below their frame rate with the feature disabled, but then have to wait until a subsequent frame is rendered before generating an interim one between the two, it effectively increases input latency rather than reducing it. So while a game's motion might look smoother to an observer, it won't really feel smoother while gaming, in terms of responsiveness. The generated frames also tend to look notably worse then the rendered ones, and while that might not affect visual quality so much at high frame rates, at low frame rates the artifacts can become more noticeable. So if an emulated game has unplayable performance now, frame generation isn't likely to help with that.
 
  • Like
Reactions: ivan_vy
If we're going to keep getting higher and higher refresh rate displays this type of technology is going to be a must. I'd really like to see some successful hardware agnostic approaches, but I can't help but look at FSR/DLSS/XeSS. DLSS and XeSS (on Arc) look better than FSR in most circumstances below 4k resolution. I can't help but wonder if frame generation is going to be similar in that it works best on a dedicated codepath that can leverage proprietary hardware.

edit: speaking of Intel and frame generation I just saw this https://asia.siggraph.org/2023/presentation/?id=papers_744&sess=sess155
 
Last edited:
It seems unlikely that it would help much. From what I understand, PS3 emulation is CPU-limited more than anything. It shouldn't take all that much graphics hardware to render 10+ year old games that were originally designed to run on a graphics chip mostly equivalent to a (circa 2005) Nvidia 7800 GTX, even at much higher render resolutions. The real performance limitations come from trying to emulate the PS3's 8-core IBM Cell processor on other CPUs, and upscaling isn't going to help with that.

And FSR upscaling was already added to the leading open-source PS3 emulator well over 2 years ago. At the time, the developers stated that they would not be supporting DLSS, because it required access to motion vectors, something their emulator (and perhaps PS3 games in general) didn't use. The main advantage of FSR2+ is that it is also able to utilize motion vectors for better results, but much like DLSS, if the emulator doesn't have access to that, it likely won't be implemented. FSR3 Frame Generation also utilizes motion vectors. It's technically possible to implement frame generation without access to that, but the results likely won't be particularly good.

Even if it were possible to implement frame generation in a PS3 emulator, that feature is usually best reserved for games that already run well, to make them look a bit smoother on high refresh (120+ Hz) displays, rather than something to fix performance issues with games running at low framerates. Since the games are still effectively running at or below their frame rate with the feature disabled, but then have to wait until a subsequent frame is rendered before generating an interim one between the two, it effectively increases input latency rather than reducing it. So while a game's motion might look smoother to an observer, it won't really feel smoother while gaming, in terms of responsiveness. The generated frames also tend to look notably worse then the rendered ones, and while that might not affect visual quality so much at high frame rates, at low frame rates the artifacts can become more noticeable. So if an emulated game has unplayable performance now, frame generation isn't likely to help with that.
'fake frames' (done with no motion vectors required) would help in RPGs games (latency is not so important here). 20-30 fps are just abysmal to play with.
 
Status
Not open for further replies.