Preview: AMD FidelityFX Super Resolution

There is atleast nice boost in performance even with ultra quality. I can imagine with a some sharpening it could look very close to native.

 

can we get some slider comparisons ?
those I found do not look flattering even in quality mode,horribly blurry
https://www.purepc.pl/misc/img_comp...pl/files/Image/red_temp/1624299906_16_ncb.jpg
https://www.purepc.pl/misc/img_comp...pl/files/Image/red_temp/1624299955_16_ncb.jpg
dafuq are those games too,they look bad on native
was there not a single triple A supported on launch ?

i still don't know how people want sharpening to compensate for lost detail.

 

Only I can say is gosh this is cooool!
Thanks AMD for giving us some extra boost that is not limited to specific brand, make etc!

 

It removes some of the blurriness so i dont see why not use RIS+FSR ultra quality mode.

 

it's a shame they are "behind" the curve in technology however this is very promising and by next gen it should be just as good as DLSS so it seems. One other thing to note is that it can add life to older cards from both sides of the table so I would gladly buy an AMD card in support of them.

 

Good job AMD. I would say in a couple years when this technology matures a bit more this will be a true competitor to DLSS.

So exciting the times we're living in now. Technology abound!!

 

Imagine if we had a toggle button inside the driver to make it work in every game.

 

It would be helpful if there were comparison shots between 1080p/1270p/1440p/1662p native res and 4K perf/balanced/quality/ultra SR to see how the image is being upscaled. Especially 1080p to 4K with perf SR would tell us a lot, I think.

Nobody seems to do this comparison

 

It's fine for a first attempt but not good enough for me to want to use it.

 

My big issue with Sharpening (even CAS sharpening) is that my observation's been it makes any aliasing present in the image look noticeably worse.

The other issue is, if it's not integrated into the game itself, you get sharpening affecting the UI elements which is noticeable to me. For example, Overwatch + Nvidia's control panel sharpening makes the UI reticule look like crap, but this isn't an issue in say RAGE 2 where AMD's CAS Sharpen is integrated into the game.

I will say at least in my local tests, I've found the integrated AMD CAS Sharpen to look better than Nvidia's Control Panel Sharpen forced externally, but I'm not sure about forcing AMD's externally as I lack an AMD GPU.

To clarify, I do think "good" sharpening filters have their place when used with restraint at least. A modern CAS Sharpening filter like AMD's integrated at the engine-level (so you're not sharpening things you don't want to be) set at a mild strength in conjunction with something like modern TAA produces what I think is a pretty good looking final result given TAA tends to soften the image somewhat.

 

I'm waiting for Digital Foundry and other outlets to do some testing, but I'm hoping at least we've got a "good upscaler" even if it's not doing as much effective reconstruction as DLSS (which was of course rather poor at its launch in fairness). From what I gather AMD isn't currently using motion vectors or the AI inference route at this point (though Nvidia's solution requires dedicated hardware in the form of tensor cores to work correctly from what I've read) so i'm curious to see how AMD aims to improve their solution down the line.

What Nvidia has done with DLSS from 1.0 -> 2.1 has been transformative, but if AMD wants their solution to run in a more generalized fashion on any hardware (no tensor cores to accelerate the AI inference/"ground truth" comparison stuff) then I'd love to hear more about their plans. I'd expect they could still use motion vectors and pull data from prior frames like Unreal's Temporal Upsampling does (assuming the game in question has TAA at least), but I don't know if they're doing something like that yet either in FSR's initiate form.

 

Here is AMDs and Nvidias one dirty little secret they don't want you to realise....
How you get "Pseudo FSR" for unsupported GPUs or games and this one trick works in everything
Enable GPU scaling in your drivers, run games in lower resolutions. GPU outputs to your displays native resolution while you're rendering at a lower resolution.... Use post processing AA or filters as necessary....

 

I'm not totally sure about the inner workings of the tech, but I imagine it's because these modern reconstruction techniques require access to certain information from the game engine that's not readily available without some integration from the dev. Or perhaps older APIs don't support it or some such. If there weren't some sort of issues with allowing it as an external option to be forced then I would hope they would let us do that. Another thing would be integrating it into the game itself means you can do the FSR reconstruct before film grain/the HUD are rendered.

If it was merely a "good upscaler" where all it needed was the current frame then you probably could force it externally for the same reasons you can force FXAA externally. Namely that you just need the final frame to operate on and no other information really.

DLSS on the other hand from what I've read needs to have the particular game tested and run through on their local "super computer" at super-resolutions so that the AI portion has "ground truth" comparison data to make its inferences from. Now, I know I recall them stating at some point that they had made their AI solution more generalized/that all of that wasn't maybe as necessary anymore as they could use info from other datasets or some such, but I would bet they're still generating comparison data for specific games and I'm not sure the tech works as well without it so if they don't have that comparison data set they maybe wouldn't want you to use DLSS for fear of getting subpar results.

The other thing we know is DLSS uses motion vectors and works similarly to modern TAA in some capacity (pulling info from prior frames/temporal accumulation). The only game I know of that supports DLSS which does not have modern TAA (to my recollection at least) is Nioh 2 and I'd be curious to know whether or not DLSS does not work as well in that title since it's got no temporal component to its stock anti-aliasing/probably doesn't have motion vectors either unless they reworked their engine specifically for DLSS or some such.

A lot of what I'm saying is speculative, I acknowledge, perhaps I'm mistaken on some portions, but my point is you may not be able to generalize a solution if it requires integration at the game level due to certain prerequisites. We'd need to know more about how exactly FSR works I imagine. Still, if all it's doing is upscaling using the current frame and nothing else though then yeah, I don't understand why that couldn't be forced externally.

 

Wonderful tech, bravo AMD.
I hope there will be a way to use it on older titles as well.

Skyrim with a gazillion of mods + FSR = yes please

 

I feel AMD has pushed the Scale factor too far, 3 quality modes would be enough: 1.25 for quality; 1.5 for balanced; and 1.75 for performance.
At 4k the quality mode would be almost placebo quality, the balanced would offer a working compromise and the performance in case you're desperate (but 2x just looks awful to use).

 

Yep, that's the thing that bothers me most about FSR, but looking at the other side it's the same with DLSS. Wish that FSR relies purely on driver/GPU support, not on game FSR implementation, in that case FSR would lift of like a rocket ship.

It'll be interesting to see if NVidia is going to do anything about GPUOpen/FSR integration in future GeForce drivers. Probably it would boost older Pascal GPU's quite noticable in games and NVidia won't be thrilled to see some of their non-RTX GPU's still keeping up with RTX cards in latest games. Guess they'll prolong implementing this in drivers as much as they can and brute force DLSS whenever and wherever they can. They did the same with G-Sync backthen and after a while with cheaper FreeSync monitors flooding the market they had to respond via drivers in form of Adaptive VSync.

Wishing all the best to AMD with future FSR development but as Hilbert already stated AMD need to do something on hardware level to became real competitor to DLSS.

 

Watching reviews it looks like it is superior to CAS or DLSS 1.0 as long as you are using Quality or Ultra Quality mode. As expected not as good as DLSS 2.0 but still better than what we had with CAS and DLSS 1.0.

If it was not the case then it would probably not be better than CAS. What makes it better is probably the fact it must be implemented by the dev and the dev can fine tune it instead of using generic settings that might not work for this particular game.

 

They probably think about 5-6 years from now. 4k to 8k will probably look good enough. But yeah atm i would not touch Performance and balanced more with a 100 feet pole. In fact i would personally not touch FSR unless i have no other choice (like the game performing under 60 fps at medium settings then yeah i mean ...). Still the ultra quality mode looks close enough to be a viable options for sure.