GHG
Member
I have huge respect for gamers nexus. Hardware Unboxed on the other hand cry about everyone trying to pay them off or incentivise them for giving positive reviews... Apart from when it involves AMD.
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Digital Foundry: AMD FidelityFX Super Resolution FSR Review:
- Thread starter Arioco
- Start date
I have huge respect for gamers nexus. Hardware Unboxed on the other hand cry about everyone trying to pay them off or incentivise them for giving positive reviews... Apart from when it involves AMD.
alucard0712_rus
Member
And we don't know what Insomniac can do with 2060S GPU, so Alex still can be right.
TrueLegend
Member
Well it's not the AMD customers that are not happy, read the room pls. No sane person ever thought AMD is gonna beat NVidia ML-powered DLSS 2.0 on the first attempt. Everybody is surprised at how FSR has actual relevance and usage as most of us suspected it to be gimmicky after the initial reveal impressions. It's the Nvidia Fanboy who can't swallow the tough pill that Nvidia does what only benefits Nvidia at the cost of progress in the overall gaming industry while AMD's approach is inclusive as these fanboys are dependent upon the feeling of being 'Special'
TrueLegend
Member
Its not a cry when at both occasions both of these companies have apologized to them. It actually speaks of their quality. Nvidia doesnt go after average joe whose opinion it doesnt care about. The very fact that they went after HW reinforces the fact that Nvidia cared about what Hardware Unboxed had to say about their products.
Armorian
Banned
It's good to have options. FSR doesn't break hud clarity and it's decent enough in highest quality option.
Plus it works on any GPU from 2016+
Amiga
Member
FSR made me remember GPU wars are as much a thing as console wars. difficult to get honest opinions whenever this happens.
FSR 1.0 is very usable. software updates and hardware acceleration with next AMD GPUs will make it even better. iNvidia will be forced to stop drip feeding tech or lower prices if they don't have anything more.
FSR 1.0 is very usable. software updates and hardware acceleration with next AMD GPUs will make it even better. iNvidia will be forced to stop drip feeding tech or lower prices if they don't have anything more.
Buggy Loop
Member
As always.
AMD has the benefit to see the runner with tons of R&D go ahead and make a few errors here and there, but what does it say about you if you follow-up next and fuck it up?
Genx3
Member
FFXSR is just another upscaling technique.
It's not AI Upscaling. AI upscaling should be next on the agenda.
Last edited:
Pagusas
Elden Member
As always.
AMD has the benefit to see the runner with tons of R&D go ahead and make a few errors here and there, but what does it say about you if you follow-up next and fuck it up?
it is indeed a head scratcher, which is why I think this was a rushed quick solution, and 2.0 should take its time and get it right (aka use motion data, there is no reason not to if you aren’t going to make it a universal driver level solution, and instead make it per game)
mans that makes me really wonder, why is this solution a per game feature instead of driver level? If it was driver level it would be honestly a cool feature, making every game (within reason) suddenly have a “better than nothing” scaling option. Instead it’s locked into the same per game issue that dlss has, which turns this back into a money and developer support issue, which AMD has been pretty shitty at.
fersnake
Member
this is good for everyone, i have a gtx 1080 and im waiting for this in the future because nvidia wont support us with old cards. and i really dont get why ppl love to throw shit to amd for this. it's a start no one was specting a better tech than dlss 2.0 out of the blue. baby steps guys baby steps.
Buggy Loop
Member
And Durante chimed in on DF's coverage.. Pretty much a nail in the coffin of any argument these other "tech youtubers" would even compare to Alex's analysis.
CitizenZ
Banned
Quote of all reviews is from LTT, "Simply turn your settings down"
SF Kosmo
Al Jazeera Special Reporter
No one was expecting anything competitive with DLSS 2.0, but this doesn't even seem like a particular competitive solution compared to anything other than a basic bilinear upscale. Even TAA upscaling fares better and that's not big praise.
Really disappointing, to be honest.
LiquidMetal14
hide your water-based mammals
First iteration jitters aside, I'm glad their headed hopefully in the right direction.
Thesuffering79
Banned
I wouldnt use it currently but i may use it with DLSS 2.0+ in the future if possible.
Mister Wolf
Gold Member
Elog
Member
Alex must have been smoking weed in Berlin when he wrote that.
Alex states the following: "AMD's FSR is not an image reconstruction technique like checkerboard rendering, DLSS 1.0, DLSS 2.0, Temporal Anti-Aliasing Upscaling, or a variety of techniques which look to reconstruct the image's higher level detail beyond the spatial realm while Anti-Aliasing that new image information.
FSR is similarly not Anti-Aliasing - FSR comes after a game has already been anti-aliased and inherits the qualities, faults, and benefits of the anti-aliasing technique of the game in question.."
This is horse shit/ snake oil at the highest level.
All these techniques are interpolation techniques. Full stop.The question is what mathematical function you use for the interpolation. And since they are interpolation techniques you upscale the digital signal, i.e. it is a type of upsampling/upscaling. To claim that one technique is not using interpolation but instead 'reconstruct the Image's higher level of detail beyond the spatial realm' is some serious gibberish.
Alex states the following: "AMD's FSR is not an image reconstruction technique like checkerboard rendering, DLSS 1.0, DLSS 2.0, Temporal Anti-Aliasing Upscaling, or a variety of techniques which look to reconstruct the image's higher level detail beyond the spatial realm while Anti-Aliasing that new image information.
FSR is similarly not Anti-Aliasing - FSR comes after a game has already been anti-aliased and inherits the qualities, faults, and benefits of the anti-aliasing technique of the game in question.."
This is horse shit/ snake oil at the highest level.
All these techniques are interpolation techniques. Full stop.The question is what mathematical function you use for the interpolation. And since they are interpolation techniques you upscale the digital signal, i.e. it is a type of upsampling/upscaling. To claim that one technique is not using interpolation but instead 'reconstruct the Image's higher level of detail beyond the spatial realm' is some serious gibberish.
Last edited:
Bulletbrain
Member
Ok it may have not lived up to expectations or the marketing hype (it never does) but, this is the version 1.0 of a technology available for BOTH AMD and NVIDIA cards! It's another tool for the PC toolbox to further customize your PC gaming experience. As Anthony from LTT said "enjoy your free upgrade".
Bulletbrain
Member
I think you should read up a bit more in detail about DLSS 2.0 and its use of machine learning trained on a per game basis.
Elog
Member
What I wrote is 100% correct.
The ML creates the key numerical input for the interpolation calculation. Simplistically you have a low res image and image motion vectors as parameters to the MLed algorithm and in return you get a number that you use for the interpolation. In allowing the interpolation to vary with the specific game/scene/movement you improve the interpolation and get fewer artefacts.
But it is still interpolation.
Ev1L AuRoN
Member
Digital Foundry was the only review that truly analyze with some depth the tech. Nothing unexpected, actually. It's pretty much useless in anything but 4k Ultra Quality. For low-end GPU it loses all-purpose since those gpu's will most likely need to reconstruct sub-1080p to 1080p and with that resolution the FSR is horrible.
In the end, there is no such a thing as a free meal. Nvidia like it or not develop silicon to put in their GPU with the purpose of image reconstruction with the Tensor cores. DLSS to achieve what it does needs image training, motion vectors and specialized hardware to reconstruct the final image. I don't think FSR will be competitive until AMD put some hardware functions to help the process, like Nvidia, maybe in RDNA3.
In the end, there is no such a thing as a free meal. Nvidia like it or not develop silicon to put in their GPU with the purpose of image reconstruction with the Tensor cores. DLSS to achieve what it does needs image training, motion vectors and specialized hardware to reconstruct the final image. I don't think FSR will be competitive until AMD put some hardware functions to help the process, like Nvidia, maybe in RDNA3.
Bulletbrain
Member
Don't you think that a method using motion vectors and predictive elements (such as DLSS2.0) is different from a method that does not? Sure, you can call them under the broad banner "interpolation" if you want, but the underlying technology, accuracy and effectiveness of the end solution is wildly different.
Thats DLSS 1.0
DLSS 2.0 ditched per game training and introduced TAA.
Depends on the game. TAA can get this very distracting shimmering effect. And its only available on UE4 games.
Heres a direct comparison in Godfall.
The true test will be against UE5's TSR.
Last edited:
Unknown Soldier
Member
This is why DF only posts to Era now instead of GAF. Some of you people should be fucking ashamed of yourselves.
Auf Wiedersewho gives af.
ethomaz
Banned
He is right.
FSR will just upscale the image… you still needs to use AA before or after like any other scaling.
vpance
Member
From tests I've seen people do with the Valley of Ancients demo, UE5 TSR needs at least 1440p input res to reach near 4K native clarity. Still has some of the usual TAA motion issues too, at least for now.
Thesuffering79
Banned
You probably dont understand these techniques if you think these are mere interpolation. Especially DLSS 2.0 and FSR.
DLSS uses a high res 16k source image to upscale with some line recognition thrown in hence why sometimes it looks better than native 4k.
FSR has preset structures it recognizes and draws them at higher detail with the rest looking worse than regular upscaling.
The two is nothing like each other. Hence why i would use both at the same time when FSR gets better. As it stands it is straight trash and i dont see how the way the generate the image will get any better. But it may be useful coupled with o0ther techniques such as TAA or DLSS.
Last edited:
Shmunter
Member
It upscales based on single frame data? This is no better than a bread & butter upscale feature on a Sony TV surely.
How can you not use additional data from past frames when you have all the data readily available? I can’t see this being better than any custom solutions unless they’ve done something completely new and radical here.
How can you not use additional data from past frames when you have all the data readily available? I can’t see this being better than any custom solutions unless they’ve done something completely new and radical here.
ZehDon
Gold Member
So, FSR is not a technique designed to actually understand the image being processed, and, as a result, it cannot construct image information intelligently when upscaling, which is where DLSS gets its magic from. It's literally a form of "make this one pixel into a four pixel block, using the surrounding pixels to guesstimate the colour of those new pixels". The result is as expected; a blurry mess of subpixel information, especially in motion.
No wonder it can run on old hardware - it's not doing anything worth using.
No wonder it can run on old hardware - it's not doing anything worth using.
Boy bawang
Member
I hope that they'll improve the tech. I'm very much rooting for an hardware agnostic method of upscaling.
But as it is, DLSS is just much better, and it's not surprising as it's been around longer, runs on dedicated hardware, and is made by nvidia which has more ML experience than AMD.
But as it is, DLSS is just much better, and it's not surprising as it's been around longer, runs on dedicated hardware, and is made by nvidia which has more ML experience than AMD.
Shmunter
Member
You can still still re-construct single frame data intelligently by doing some AI pattern matching against known samples in a database. E.g. recognizing faces and treating appropriately, trees, grass etc. Gaps can be filled and edges enhanced with some smarts, but it cannot be nowhere as good or accurate as using actual data from past frames. I have no idea if this even does that tho.
Sony TV's do, do the above and do produce better than native results, but that's a TV, not a gfx card with a lot more at it's disposal and hooks directly into the game code.
Spukc
always chasing the next thrill
That’s a lot of text for not as good.
TrueLegend
Member
still more valuable than your entire existence.
Spukc
always chasing the next thrill
JeloSWE
Member
Source?
RobRSG
Member
I want not to be biased about Hardware Unboxed. But every video I watch, it’s clear that they favor AMD with passion when it comes to GPUs.
Downplaying DLSS, Ray Tracing, recommending a 5700XT dispite its inability of RT and early obsolescence, creating weird bottlenecks as “experiments”, calling out to Nvidia publicly and others to name a few.
Bulletbrain
Member
Yup. HU has a sliiight whiff of AMD bias. DF has a sliiight whiff of NV bias. GN seems to be the most objective of the lot (so far. Let's hope it stays that way).
mrBarrelNut
Member
It's very niche. Not big benefits over simply lowering resolution and letting your monitor do shitty upscale to native resolution.
And UE4/5 games should use their TAAU solution for equal performance (id needed) but with much better image quality. And most other game engines already have existing TAAU upsamplers as well.
To be able to compete with DLSS/TAAU this needs to be temporal as well, and use motion vectors to do actual reconstruction at desired quality level. And this also means that such version won't be drop in replacement with now released FSR version and will need proper integration to game engines.
Last edited:
CuNi
Member
I guess now we know why you need mod approval to post.
There is nothing wrong with accepting that DLSS 2.0 is better than FSR.
It does its job especially for lower end hardware but that's all there is to it.
DLSS also had to start somewhere so all we can do is hope that they will continue to build off of this and not just leave it at that.
alucard0712_rus
Member
You call one term "wrong" and introduced another term but it changed nothing.
By "upscaling" he mean resampling the image to a higher resolution using information from that image only. It can be used for still image (one frame).
By "reconstruction" he mean resampling the image to a higher resolution using information from that image + information from other frames + any other information.
I think you can call both "interpolation".