Thread: AMD announces FSR (Super Resolution) - Open Source, Cross Platform, June 22
Direct comparison. For me, especially the lower right corner displays how FSR fails to enhance texture detail and only affects edges & overall sharpness.
Hope they'll manage to improve it in the future!

fsruvkkk.jpeg
 
No thank you. I'll just reduce the resolution and smear Vaseline all over my screen if I want to destroy my image quality and depth of field for a bit of performance boost instead of using a post processing effect...

The really troubling thing here for me is, that TAA upscaling in different engines can actually produce better results than FSR.

The articles I've read sounded rather positive, but the actual footage and comparisons that I've seen, it doesn't look that hot or like a game changer.
 
  • This tbh
Reactions: VlaudTheImpaler
I didn't particularly like DF's analysis for this one. When they compare the simple upscaler and the TAA, the texture of the trunk and the moss on the trunk is clearly better on FSR compared to both other options, yet, somehow they failed to see that...

Considering how enthusiastic DF was about DLSS 1.0, and how apathetic and critical they are of FSR, can't say that I can call them trustworthy on this one. And if you look through the comments on their video, there are multiple people that agree. Here's one (scroll down to see the highlighted comment for yourself).

In my eyes, the quality loss of "ultra quality"- and "quality" FSR in 4K isn't actually that substantial (especially when you stop pixel peeping and start playing), and the lower tier modes is basically made for iGPUs, something AMD has been very honest about. I don't necessarily feel like this video portray the benefits of FSR, only the negatives. Which quite honestly isn't right. DLSS has equal amount of drawbacks as FSR. And I am a bit afraid that DF has jumped on the nvidia hype train and now they have a hard time getting off. Especially about FSR since Alex made some pretty bold statements on the technique before seeing it in action. I love DF but I still question their objectivity due to nvidia sponsored videos, their game benchmark selection (which mostly represent nvidia sponsored titles) as well as the nvidia narrative in language use (often using nvidia terminology instead of independent industry standards). The main reason for this concern is not really Alex contributions to DF but rather Richards weird cherry picking in GPU reviews, the nvidia sponsored content in conjunction with the big navi release. The way these things combined creates a narrative that don't sit right with me. I love DLSS as tech, but as a developer I am actually even more excited for open source solutions like FSR and other techniques, since it gives me more freedom and control of the image upscaling.
 
I did notice that DF were by far the harshest critics of FSR, going against most of the rest of the tech reviewer community who seemed to mostly be fairly positive on FSR.

I don't know if I would go as far as to claim they are biased or fanboys or whatever, that kind of thing gets levied at them a lot from fanboys so I would be hesitant to jump on that train. However personally I have noticed a lot of the same things as the commenter quoted by @Ascend.

It is what it is, some outlets will have particular preferences or marketing relationships with companies that they don't want to damage or might potentially influence their editorial one way or another. Just stick to who you trust and over time you will likely notice patterns one way or the other with various outlets.

Having said that I don't necessarily believe DF are just shilling and like I said they often get accused of that from every direction with console warriors etc... I will say that they do seem to have a very close relationship with Nvidia though which is pretty evident to anyone paying attention but people can interpret that whichever way they want, doesn't necessarily mean it would influence their editorial, although I suppose it could potentially plant the seeds of doubt in the audience.

Speaking of FSR itself the general consensus seems to be that it produces quite good results, much better than a lot of people were expecting.

Pros:
- Much better than DLSS1.0
- Much better than lower render resolution + sharpening
- Very good image quality at Ultra Quality and Quality modes targeting 4K output
- FPS performance/performance hit ratio seems to be quite good
- Incredibly easy to implement, one developer said they spent 2 hours implementing it for their game.
- Works on consoles
- Works on AMD/Nvidia/Intel etc.. on PC
- Open source - So can be improved/modified by many more people than just AMD engineers.
- Game support while initially dull seems to be ramping up quickly (at least that is the impression right now)
- Due to working on ratios rather than fixed resolutions offers easy support for 21:9 or other unusual resolutions.
- Does not introduce ghosting that plagues DLSS or other strange artifacts.

Cons:
- While image quality is quite good, in a best to best scenario DLSS edges it out with better overall image quality.
- The lower quality modes produce noticeable burring and worse image quality than the native target.
- DLSS is much better with lower resolution input data/lower performance modes.
- Introduces some shimmer artifacts, gets worse as quality modes decrease.
- 4K target with Ultra or Quality modes looks good, but lower res like 1440p target looks noticeably worse with only Ultra being worth using at this resolution.
- Initial game support sucks
 
Here is a cool comparison with Godfall:




Additionally it seems DOTA2 now supports FSR, which was pretty damn fast. There are tons of people posting comparisons and if anyone owns it you should be able to check it out for yourself now.




In addition The Riftbreaker has a free demo you can download from Steam with FSR enabled so you can toggle on/off and check out the difference yourself.

Seems to be a lot of excitement about it in the tech twittersphere.

Speaking of DF and the TAAU comparison, it seems that TAAU disables Depth of Field, but the DoF was present in the FSR comparison, hence additional blur. Alex from DF has mentioned he will update their article to reflect this in the twitter thread.

Interesting twitter thread here where they go over it:



Here is a reddit thread where it was pointed out the DoF issue with the TAAU comparison for Kingshunt that DF used.

 
Last edited:
I'm not sold on this. My understanding from DF (which, even if it goes against the grain, backed its reasoning thoroughly and made an excellent argument) is that if a game doesn't use TAA, or temporal data, to increase detail in a lower-res render, this is fantastic and basically does it for that game.

If it does, however... And the problem is, that's most major engines and games out there.

So, where does this fit in? Don't get me wrong. The only Nvidia tech I own and am aware of, resides in my Switches. With the consoles, I have a vested interest in them getting this right. I just don't think it's there yet, for most games that already have similar solutions in place.
 
  • Hell Yeah!
Reactions: Ascend
I'm not sold on this. My understanding from DF (which, even if it goes against the grain, backed its reasoning thoroughly and made an excellent argument) is that if a game doesn't use TAA, or temporal data, to increase detail in a lower-res render, this is fantastic and basically does it for that game.

If it does, however... And the problem is, that's most major engines and games out there.

So, where does this fit in? Don't get me wrong. The only Nvidia tech I own and am aware of, resides in my Switches. With the consoles, I have a vested interest in them getting this right. I just don't think it's there yet, for most games that already have similar solutions in place.
So FSR happens after the AA phase, meaning a game is free to use whatever AA solution it wants including TAA.

FSR has been tested on multiple tech channels vs simple render scale and render scale+sharpening and FSR wins out handily.

I don't really watch much of DF content but by all accounts they are normally quite good, knowledgeable and thorough but in this case as the evidence mounts it has become clear they are showing their ass with their FSR take.

I'm assuming it was likely because their video was rushed, but for a channel that is all about pixel counting, fine details and graphics techniques to not notice that DoF was turned on on FSR and off with TAAU is pretty damning, especially given that it had to be pointed out by random redditors. The DoF change is the main difference between their comparison of TAAU and FSR which makes FSR appear more blurry and which was the main evidence used by them to claim TAA/TAAU based solutions were superior.

So how did this DoF bug/mistake happen? Well both Godfall and Kingshunt are UE4 games, TAAU was added in to the engine and can be forced via command line arguments/config changes. So the games render pipelines were not designed to take into account TAAU, hence there are bugs with DoF where TAAU diasbles it, showing a much cleaner image compared to FSR or the "simple native upscale".

In addition the DoF effect seems to be kind of bugged anyway as the character models in focus shouldn't be getting affected by the effect but anyway, not really the point I'm making.

Once DoF is disabled you can see that the FSR image looks way better than it did before and that FSR and TAAU are both quite close with each having different pros/cons in the image and how they handle certain elements of it.

Speaking of TAAU for a second (which is Unreal Engines new temporal upscaling method/plugin) it suffers in motion from artifacts and ghosting, such as any TAA based solution will as such FSR doesn't introduce these artifacts and would (in theory) look better in motion compared to TAAU. Granted I'm not an expert on either tech, just explaining my understanding.

I'm concentrating a lot on the DF review here but that is because this is where most of the misconceptions and negativity regarding FSR come from in the tech community at the moment. So for now lets continue talking about DF's take.

They show 3 side by side comparisons, the first one is some undefined "simple upscale". Nobody knows what this means or how it has been achieved. Do they mean they took a still image and applied some upscale/sharpening filters in something like photoshop/after effects or a similar program? If so then it is not really an apt comparison as things like FSR/TAAU/DLSS all work in real time in a game and whatever this is would not.

So if it is not post processing on a single still image....are they using some kind of render scale? Are they using sharpening? Nobody knows as they have been completely opaque about it.

Just to be clear, I'm not angry about this or having a go at you or anything, just stating some inconsistencies and issues I have with DF's methodology in this case.

So back to FSR vs TAAU, they both seem to be great options so I'm very happy we are getting more options and not locked in to one hardware vendor or another. However in DF's comparison (even if we remove the terrible DoF effect which is 95% of what makes the FSR image look worse) they show the worst case scenario for FSR which is a 1080p->4K upscale using the performance mode.

This is fine from an academic standpoint vs TAAU 1080p->4K, however they mention the performance (FPS) are essentially the same but don't show any frame graphs or metrics, which seems pretty odd given that is there whole schtik, I'm assuming it was likely due to lack of time rather than any kind of malice but it does seem odd that they seemed very reluctant to show any FPS charts/metrics with any of the FSR modes (not specifically with relation to vs TAAU).

So in other words we don't actually know if the performance in FPS is close/the same with those two modes compared, and some other people running independent tests to recreate this are getting different results so I'm not sure what the deal was with this from DF, but again I put it down to lack of time most likely.

EDIT: Actually upon further inspection it looks like Alex from DF locked the games to 60FPS during his FSR vs TAAU test, then looked at GPU utilization % and used that as a "performance" comparison to say that "perfomance is essentially identical", rather than....you know actual FPS performance. Worse than I thought...

So back to resolutions, we know FSR works best with higher input data/resolutions and we know nobody recommends using performance mode, so in a realistic scenario they should have compared best to best (Ultra Quality FSR vs best case TAAU) to get a fuller picture as well as the lower res upscale comparison. I'm happy they did the lower resolution input comparison as it does show some of the weaknesses of FSR, but they should have also balanced that with higher resolution comparisons to show how people would actually use both in games and for a more well rounded overview.

I mean if graphics techniques and comparisons are your whole deal and purpose of your channel then it presents some pretty huge oversights and problems with methodology all of things I've mentioned above. Again I'm assuming it is likely due to the time constraints for most of it, but some of these are pretty inexcusable oversights, especially when you have the reach/audience of DF and people take their word on tech as gospel.

Anyway, sorry if this seemed like a sperg out, I was just reading a ton about the issues with DF's video today and lots of people have pointed out those issues, which makes sense as to why almost the entire rest of the tech reviewer community seems to have an opposite take to DF here. I don't generally have an issue with DF and I don't think they are fanboys or paid shills or whatever, but they had a bad take here and for the sake of trying get an objective view on these new technologies they should be called out on where they might have messed up to improve in future.
 
Last edited:
I didn't particularly like DF's analysis for this one. When they compare the simple upscaler and the TAA, the texture of the trunk and the moss on the trunk is clearly better on FSR compared to both other options, yet, somehow they failed to see that...

Considering how enthusiastic DF was about DLSS 1.0, and how apathetic and critical they are of FSR, can't say that I can call them trustworthy on this one. And if you look through the comments on their video, there are multiple people that agree. Here's one (scroll down to see the highlighted comment for yourself).

In my eyes, the quality loss of "ultra quality"- and "quality" FSR in 4K isn't actually that substantial (especially when you stop pixel peeping and start playing), and the lower tier modes is basically made for iGPUs, something AMD has been very honest about. I don't necessarily feel like this video portray the benefits of FSR, only the negatives. Which quite honestly isn't right. DLSS has equal amount of drawbacks as FSR. And I am a bit afraid that DF has jumped on the nvidia hype train and now they have a hard time getting off. Especially about FSR since Alex made some pretty bold statements on the technique before seeing it in action. I love DF but I still question their objectivity due to nvidia sponsored videos, their game benchmark selection (which mostly represent nvidia sponsored titles) as well as the nvidia narrative in language use (often using nvidia terminology instead of independent industry standards). The main reason for this concern is not really Alex contributions to DF but rather Richards weird cherry picking in GPU reviews, the nvidia sponsored content in conjunction with the big navi release. The way these things combined creates a narrative that don't sit right with me. I love DLSS as tech, but as a developer I am actually even more excited for open source solutions like FSR and other techniques, since it gives me more freedom and control of the image upscaling.
Well, I don't agree with this poster about the tech. Whatever his opinions of DF is doesn't really matter here. The big elephant in the room is can this FSR best or even match DLSS 2.0. That answer is clearly a "no". Nvidia put a lot of work into their hardware supported DLSS solution with ML. AMD is behind on that and RT. It's just the way it is at the moment. Watching the article seems to portray that AMD's solution has a LOT of drawbacks. The video comparisons don't lie and we can all see the implementation in use. It's just not good enough.
 
Well, I don't agree with this poster about the tech. Whatever his opinions of DF is doesn't really matter here. The big elephant in the room is can this FSR best or even match DLSS 2.0. That answer is clearly a "no". Nvidia put a lot of work into their hardware supported DLSS solution with ML. AMD is behind on that and RT. It's just the way it is at the moment. Watching the article seems to portray that AMD's solution has a LOT of drawbacks. The video comparisons don't lie and we can all see the implementation in use. It's just not good enough.
Agreed that image quality is not quite as good as DLSS2.0+ (although clearly much better than 1.0). I don't think anyone has claimed otherwise, DLSS definitely wins out with overall IQ in a best to best scenario. However FSR1.0 seems to get pretty close and produce a very good image without necessarily introducing ghosting in movement that DLSS is notorious for or any of the other weird artifacts issues that most people gloss over.

DLSS is definitely the superior overall tech, however it is also more difficult/time consuming to implement as a developer vs FSR, and more importantly FSR is cross platform with consoles as well and doesn't require dedicated hardware to run which means it can be used on almost all GPUs, which is a pretty big advantage.

Where FSR seems to really fall down is lower resolution inputs, to be clear DLSS loses quality here too as it is only natural that lower input data in will produce a lower quality output, but DLSS seems to be far better than FSR here when it comes to making a good image from a lower quality input, really great work from Nvidia here.

As for articles and opinions, there seem to be tons out there but almost everyone agrees that FSR produces good results and they are happy to have it as a feature. The only mostly negative review came from DF who seemed to go against the consensus of the rest of the tech community, which makes it worse that their methodology was so flawed, but I don't want to keep harping on about that or going into ever further detail.

But articles and videos are all well and good, we can actually use it right now if you own any of the supported games, and if not anyone can download the Riftbreaker demo from steam for free and test it out with their own eyes, results by all accounts seem to be pretty good.
 
Agreed that image quality is not quite as good as DLSS2.0+ (although clearly much better than 1.0). I don't think anyone has claimed otherwise, DLSS definitely wins out with overall IQ in a best to best scenario. However FSR1.0 seems to get pretty close and produce a very good image without necessarily introducing ghosting in movement that DLSS is notorious for or any of the other weird artifacts issues that most people gloss over.

DLSS is definitely the superior overall tech, however it is also more difficult/time consuming to implement as a developer vs FSR, and more importantly FSR is cross platform with consoles as well and doesn't require dedicated hardware to run which means it can be used on almost all GPUs, which is a pretty big advantage.

Where FSR seems to really fall down is lower resolution inputs, to be clear DLSS loses quality here too as it is only natural that lower input data in will produce a lower quality output, but DLSS seems to be far better than FSR here when it comes to making a good image from a lower quality input, really great work from Nvidia here.

As for articles and opinions, there seem to be tons out there but almost everyone agrees that FSR produces good results and they are happy to have it as a feature. The only mostly negative review came from DF who seemed to go against the consensus of the rest of the tech community, which makes it worse that their methodology was so flawed, but I don't want to keep harping on about that or going into ever further detail.

But articles and videos are all well and good, we can actually use it right now if you own any of the supported games, and if not anyone can download the Riftbreaker demo from steam for free and test it out with their own eyes, results by all accounts seem to be pretty good.
It's not even close, DLSS can do really impressive things with low resolution inputs.



It's a different technology.

We should wait until a game supports both so we can do some direct comparisons.
 
It's not even close, DLSS can do really impressive things with low resolution inputs.



It's a different technology.

We should wait until a game supports both so we can do some direct comparisons.

Oh I 100% agree that they do different things in a completely different way. I also agree and have mentioned above that at lower resolution inputs DLSS2.0 is far far better than FSR.

I don't think either of these things are in contention from anyone or anywhere. I also agree that DLSS2.0 has overall better IQ.

But I think FSR still produces a very good result at the highest quality setting at high resolutions like 4K, I think at 4K target with the highest quality mode that FSR will get pretty close to DLSS2.0, it won't match it definitely as DLSS will win out for overall IQ, but it will be pretty close. Close enough that most people wouldn't honestly be able to tell the difference unless 300% zooming or other things of that nature, which all in all is a pretty good result I think.

At the moment we don't have any games that support both to be able to use as a direct comparison, but it will happen eventually and then we will be able to see it for ourselves. Just to be clear, DLSS will win out in a best to best comparison, but not by as much as some people are assuming.

I've actually tried out FSR myself today on Riftbreaker at 3440x1440 ultrawide with Ultra Quality FSR. The results were really good, I was honestly impressed. What is more impressive is that FSR supports scaling at this resolution out of the box, which I don't think DLSS did? I could be wrong on that but my understanding is that non standard aspect ratios/resolutions have currently spotty support with DLSS?

Anyway, all in all it looks like adoption for FSR is going to pick up quickly and the great thing is that it supports pretty much all hardware so everyone can benefit from it while also benefitting from games that support DLSS if you have a compatible Nvidia GPU, so it seems like a win-win scenario to me. By all accounts, my own first hand one included the results of Ultra Quality FSR at high resolutions are quite good.
 
Well, I don't agree with this poster about the tech.
Based on what exactly?

Whatever his opinions of DF is doesn't really matter here.
It's more an observation than an 'opinion'. And many people are seeing it. More importantly, as posted previously by IrishWhiskey, DF actually made a critical mistake once again. And, you cannot deny that in the three-way comparison, the trunk and moss looks sharper on FSR than both TAA and 'simple', yet he argues that FSR doesn't improve texture quality. That alone should already raise some red flags that something is wrong. There is either malice or incompetence, and, it looks to be incompetence. And at this point, if you still defend DF, you either have not done your homework, or you're THAT biased that you refuse to accept reality. If it's the latter, just know that I won't be wasting my time trying to explain what is actually going on.

The big elephant in the room is can this FSR best or even match DLSS 2.0. That answer is clearly a "no".
Whether something is the "best" or not, depends on the criteria that you use, presume or require. If you need a car to drive to work in an urban city for 15 minutes, and you have to pick a Bugatti or a Toyota, the Toyota is still the better choice. Even if the Bugatti is faster in every way.

The same applies here. You have preemptively put the framework that FSR is only viable if under the condition of 400% zoom and static image analysis, it turns out to be superior to DLSS 2.0. But that is simply biased, and well, it's a perspective based on ego, arrogance and prick-waving.
The real questions for every upscaling tech are the same;
Can you tell the difference in image quality during normal gameplay? If not, it's good. If you can, does that difference justify the performance?
And obviously, NO ONE could actually tell the difference between 4K native and Ultra Quality FSR under normal circumstances. Multiple reviewers have confirmed that they had to double check whether FSR was actually on, in case of 4K with UQ-FSR. The majority could not tell the difference between 4k Native and Q-FSR either. It's at balanced that they started to notice the difference becoming larger. Obviously at lower resolutions it's less efficient.
If that is the result, what does it matter, if when you zoom in on static images, DLSS looks slightly better? There is a reason the majority of people, including the ones that have 1000 series nVidia cards, are happy with this tech.

Then there is the little fact that FSR does not somehow lock out other techniques. There is nothing stopping any developer from using TAA then upscaling with FSR. They are not mutually exclusive. There is very little wrong with this tech. But yet, once again, it's AMD's... So... Yeah, let's nitpick it and trash on it as much as possible. Am I right?

Nvidia put a lot of work into their hardware supported DLSS solution with ML.
Why should we as the end user care how much work companies put in? Isn't it the result that matters...? And I'm talking practical results, not prick-waving results.

AMD is behind on that and RT. It's just the way it is at the moment.
Why do you need to bring up RT in this discussion? This is pretty much more prick-waving at this point, which somehow you always seem to do when the subject involves nVidia and AMD.
And it's not as if AMD is 'behind'. They purposefully chose another approach in order to have as wide a market as possible, which is a smart move considering the position they are in. You should see all the people on older nVidia cards praising this tech, to for example bring additional life to their GTX 1070 cards.

Watching the article seems to portray that AMD's solution has a LOT of drawbacks.
Alright. Show us how many more drawbacks FSR has compared to TAA.

The video comparisons don't lie and we can all see the implementation in use. It's just not good enough.
In case of DF, the video comparison DOES lie. Go read the reddit post linked by IrishWhiskey. TAAU disabled depth of field. Depth of field purposefully adds blur to certain parts of the image, as I'm sure you know. And the fact that exactly in that scene, the trunk and moss in the FSR image looks superior should tell you something.
Additionally, Alex confirmed that he actually turned ON sharpening for the 'simple' upscaler and TAAU, giving them another edge over FSR.

There is so much wrong with the DF video it's not even funny. There are enough people doing their own testing, posting their own lossless screenshots and comparisons, to show that DF's 'review' has a lot of faults. And anyone that still believes that DF is the only good one and everyone else is wrong, I'm sorry, but you're simply not credible anymore, because you have proved you are incapable of objectively determining what is correct information and what isn't.
 
It's more an observation than an 'opinion'. And many people are seeing it. More importantly, as posted previously by IrishWhiskey, DF actually made a critical mistake once again. And, you cannot deny that in the three-way comparison, the trunk and moss looks sharper on FSR than both TAA and 'simple', yet he argues that FSR doesn't improve texture quality. That alone should already raise some red flags that something is wrong. There is either malice or incompetence, and, it looks to be incompetence. And at this point, if you still defend DF, you either have not done your homework, or you're THAT biased that you refuse to accept reality. If it's the latter, just know that I won't be wasting my time trying to explain what is actually going on.


Whether something is the "best" or not, depends on the criteria that you use, presume or require. If you need a car to drive to work in an urban city for 15 minutes, and you have to pick a Bugatti or a Toyota, the Toyota is still the better choice. Even if the Bugatti is faster in every way.

The same applies here. You have preemptively put the framework that FSR is only viable if under the condition of 400% zoom and static image analysis, it turns out to be superior to DLSS 2.0. But that is simply biased, and well, it's a perspective based on ego, arrogance and prick-waving.
The real questions for every upscaling tech are the same;
Can you tell the difference in image quality during normal gameplay? If not, it's good. If you can, does that difference justify the performance?
And obviously, NO ONE could actually tell the difference between 4K native and Ultra Quality FSR under normal circumstances. Multiple reviewers have confirmed that they had to double check whether FSR was actually on, in case of 4K with UQ-FSR. The majority could not tell the difference between 4k Native and Q-FSR either. It's at balanced that they started to notice the difference becoming larger. Obviously at lower resolutions it's less efficient.
If that is the result, what does it matter, if when you zoom in on static images, DLSS looks slightly better? There is a reason the majority of people, including the ones that have 1000 series nVidia cards, are happy with this tech.

Then there is the little fact that FSR does not somehow lock out other techniques. There is nothing stopping any developer from using TAA then upscaling with FSR. They are not mutually exclusive. There is very little wrong with this tech. But yet, once again, it's AMD's... So... Yeah, let's nitpick it and trash on it as much as possible. Am I right?

Why should we as the end user care how much work companies put in? Isn't it the result that matters...? And I'm talking practical results, not prick-waving results.


Why do you need to bring up RT in this discussion? This is pretty much more prick-waving at this point, which somehow you always seem to do when the subject involves nVidia and AMD.
And it's not as if AMD is 'behind'. They purposefully chose another approach in order to have as wide a market as possible, which is a smart move considering the position they are in. You should see all the people on older nVidia cards praising this tech, to for example bring additional life to their GTX 1070 cards.

Alright. Show us how many more drawbacks FSR has compared to TAA.


In case of DF, the video comparison DOES lie. Go read the reddit post linked by IrishWhiskey. TAAU disabled depth of field. Depth of field purposefully adds blur to certain parts of the image, as I'm sure you know. And the fact that exactly in that scene, the trunk and moss in the FSR image looks superior should tell you something.
Additionally, Alex confirmed that he actually turned ON sharpening for the 'simple' upscaler and TAAU, giving them another edge over FSR.

There is so much wrong with the DF video it's not even funny. There are enough people doing their own testing, posting their own lossless screenshots and comparisons, to show that DF's 'review' has a lot of faults. And anyone that still believes that DF is the only good one and everyone else is wrong, I'm sorry, but you're simply not credible anymore, because you have proved you are incapable of objectively determining what is correct information and what isn't.
Ok. Let's just skip the detailed line by line rebuttals and do this. I have not read the algorithm that FSR implements and I'm certain that if I read docs on it, I can determine without seeing anything which solution is better. For the record, I'm a sucker for complete and utter native resolution everything. 4k backbuffers, 4k textures or higher, 4k depth/stencil buffers, and 4k post-processing buffers. When it comes to sampling theory, I'm used to doing as little as possible to the original signal unless putting a low-pass filter kernel on it will remove unwanted noise in the rendered frame. So I'm only biased in what solution will give the most accurate and cleanest results. Tricks are well and dandy for realtime but ultimately they suck at being the perfect solution when it concerns rendering.

Having said that, I took DF's video with a non-skeptical mindset and trusted their results shown. If people are now saying that the difference between the two techniques is minuscule, then I want proof. I'm used to seeing correct renders for many years and I could definitely notice the differences in rendering without zooming in on pixels especially if I'm allowed to see the game loop run in realtime. But I realize that I'm one of the odd men out on this one so it's a moot point on whether I specifically can notice as opposed to the general masses.

Let's start over and you make your claim. I'll properly investigate it with no bias whatsoever. Deal?

On the RT front, I'm sorry but AMD chose a most inferior solution by using the SMUs to do the work instead of dedicated hardware. That's not going to get them very far in the next 10yrs or so. That's why I say they are behind. They simply cannot compete with the money and resources that Nvidia has. It's not being a fanboy, it's just being realistic. What's my proof? Wait for their next generations of hardware and let's see if they stick to the SMUs as a means for accelerating RT as opposed to making their own dedicated custom hardware.
 
Having said that, I took DF's video with a non-skeptical mindset and trusted their results shown. If people are now saying that the difference between the two techniques is minuscule, then I want proof.

I'm assuming you are referring to differences between FSR and TAAU that were brought up in the DF review? If you are talking about FSR vs DLSS2.0 then nobody has done a direct comparison as no game currently supports both. However people have discussed general "quality" of the images in a generic non apples to apples comparison and mentioned that FSR gets pretty close. Of course they are likely simply saying "close to native" much as DLSS aims to be as close to the native target resolution as possible.

Generally I don't have any kind of problem with DF but in this instance they were showing their ass quite a bit and had a lot of problems with their methodology. I assume a big chunk of that was time constraints because AMD messed up by not giving reviewers enough time between pre-release and release to test.

Anyway I'll give you bullet points for the issues people had with the DF review and I think you would agree that the issues were certainly not ideal.

  • Their 4K native vs FSR comparisons they had CAS sharpening turned on for Godfall.

  • Their TAAU and "simple upscale" comparisons had CAS sharpening enabled for Godfall.

  • Neither of the above was mentioned in the review itself which is a little misleading.

  • DoF was turned on in the simple upscale and FSR sections for Kingshunt and then later Godfall but turned off in the TAAU section due to a bug with TAAU being modded into these non supporting titles. This was the biggest and most glaring issue of all because it makes the TAAU image look much cleaner/sharper than the others which is the main reason Alex and people came away thinking TAAU was far superior. When turned off the FSR image (which was at the lowest quality setting for FSR btw) looks way sharper than before. How Alex did not notice this given his supposed eye for graphics rendering and small details is beyond me but a fairly glaring oversight. Surely you can agree that this was a pretty huge oversight that would affect the final conclusion and comparison pretty heavily?

  • Alex said "the performance is essentially equal" when comparing FSR to TAAU. So that means the FPS gained when using them was the same right? Wrong, Alex locked the games to 60FPS and then tried to look at GPU utilization % (which is never a good metric for anything really) to try to learn the "cost" of each technique and framed that as "performance". If you can't see what is wrong with that from a methodology point of view then I don't know what to tell you. For the record, in other people performing these tests FSR has much higher FPS than TAAU.

  • For the FSR vs TAAU comparison, Alex only showed performance mode on FSR. This is fine from an academic point of view and to show how both techniques perform with the same input resolution but AMD themselves does not recommend using this mode unless extra performance is absolutely necessary and as such in a realistic scenario nobody would use it with those settings. So they should have also included a best to best scenario of FSR Ultra Quality vs TAAU best case scenario to round out the review and give people a proper understanding of the tech. In addition for this hypothetical comparison they should have also shown an FPS chart to show which was producing a higher FPS count.

  • DF, famous for FPS charts while reviewing a tech designed to lose as little IQ as possible vs native while gaining significant frame rates for some reason shied away from showing pretty much any FPS charts for FSR on vs off or FSR vs TAAU. You would imagine that a channel famous for this kind of thing when giving an objective review of a technology designed to boost FPS would....ya know...maybe show the FPS gains?

  • Regarding TAAU, the fact that they felt the need to essentially mod games that don't support this tech to enable the feature while ignoring the fact that it might mess up the rendering or post processing pipeline (which it did with the DoF) and then on top of that to fail to notice it seems like a strange decision/large oversight.
So with all that in mind, do I think DF are biased or fanboys or shills or whatever? No, I think they simply love technology and so love newer advancements in techniques or technology. As such I think Alex looked at FSR before launch and thought "Oh it is only spatial? Then without ML or temporal data this is 'old' tech so I don't like it as much as other more advanced solutions".

Alex's comments on twitter and in their podcast seem to confirm that he made his mind up about FSR before he reviewed it, and it looks like in his review, in conjunction with some time constraints, worked his way backwards to prove his conclusion. Kind of like a film reviewer deciding a film is bad before watching it and then their review shockingly says it was bad, I mean it could just be they made a good prediction but it doesn't exactly scream objectivity.

As for proof of the above blunders and FSR vs TAAU performing much better than DF shows, please see this linked reddit post where they go over methodology and have multiple images.

 
  • Star
Reactions: Ascend
Ok. Let's just skip the detailed line by line rebuttals and do this. I have not read the algorithm that FSR implements and I'm certain that if I read docs on it, I can determine without seeing anything which solution is better. For the record, I'm a sucker for complete and utter native resolution everything. 4k backbuffers, 4k textures or higher, 4k depth/stencil buffers, and 4k post-processing buffers. When it comes to sampling theory, I'm used to doing as little as possible to the original signal unless putting a low-pass filter kernel on it will remove unwanted noise in the rendered frame. So I'm only biased in what solution will give the most accurate and cleanest results. Tricks are well and dandy for realtime but ultimately they suck at being the perfect solution when it concerns rendering.
I generally agree. But then again, when it was DLSS, everyone forgot about native, and even claims of it being better than native were being thrown around everywhere. Not by you though, at least, not as far as I have seen.

Having said that, I took DF's video with a non-skeptical mindset and trusted their results shown. If people are now saying that the difference between the two techniques is minuscule, then I want proof. I'm used to seeing correct renders for many years and I could definitely notice the differences in rendering without zooming in on pixels especially if I'm allowed to see the game loop run in realtime. But I realize that I'm one of the odd men out on this one so it's a moot point on whether I specifically can notice as opposed to the general masses.
Then I have to ask; Did you notice that the trunk and moss were sharper on the FSR image compared to TAAU and simple upscaling? Did that not ring any bells nor raise any flags?
FYI, Hardware Unboxed concluded FSR upscaling looks better than photoshop upscaling. But I will let you do your own investigation on that. Just know that DF's ' review' really is not representative and the worst out there. I'm guessing Hardware Unboxed raises a few eyebrows among certain crowds... But... Anyone is free to replicate their results, including nVidia users. So yeah. It's up to the ones that disagree to show the different conclusions with evidence, rather than talk. Like it has been done with DF.

Let's start over and you make your claim. I'll properly investigate it with no bias whatsoever. Deal?
I can live with that.

On the RT front, I'm sorry but AMD chose a most inferior solution by using the SMUs to do the work instead of dedicated hardware. That's not going to get them very far in the next 10yrs or so. That's why I say they are behind. They simply cannot compete with the money and resources that Nvidia has. It's not being a fanboy, it's just being realistic. What's my proof? Wait for their next generations of hardware and let's see if they stick to the SMUs as a means for accelerating RT as opposed to making their own dedicated custom hardware.
You are not wrong. At least, not entirely. 10 years is a bold statement, and the SMUs are doing part of the work, not all the work. But other than that, this is all true. I will not delve further into RT in this thread, because I consider that derailing.
 
  • Like
Reactions: VFX_Veteran
To add to my previous example, here are the 3440x1440 Kingshunt images taken by the redditor who discovered the DoF bug.

Bare in mind that this resolution is even below 4K target so with a 4K target FSR would look a little better again as it would be using a higher base resolution.

In both of these images DoF has been disabled. Both are 3440x1440 21:9

FSR Ultra Quality - 158FPS (same average)
tYfMja1.jpg


TAAU 77% Render Scale - 159FPS (same average)
SPJs8Xg.jpg


As you can see the FSR image is noticeably sharper than the TAAU image, including the wall and tree textures. The TAAU image is much blurrier, especially if you look at the tree foliage.

In addition due to TAAU being TAA based it suffers from standard TAA issues in movement such as ghosting which FSR does not.

So how could this TAAU be superior to FSR as many people, DF included are claiming when the above clearly shows the opposite in a best to best comparison? And this is not even the absolute best case scenario for FSR as these images are below 4K resolution, but they are high enough above 1440p that FSR Ultra works well here.

Lets look at a zoom comparison of the images:

FSR Character:
BjE6ly6.png


TAAU Character:
gShiKha.png


FSR TREE:
eIasMl8.png

TAAU TREE:
xmX2l4V.png

FSR WALL:
0qMWzpK.png

TAAU WALL:
wgrcAPd.png
 
Last edited:
Then I have to ask; Did you notice that the trunk and moss were sharper on the FSR image compared to TAAU and simple upscaling? Did that not ring any bells nor raise any flags?
Do you happen to know what timeframe in the video you are speaking about the tree trunk?

UPDATE: Found the timeline.

I agree with you guys about the textures BUT the transparency textures on the veil is another matter. Are you saying that DOF was enabled on the AMD and not on the TAAU comparison? If they are both disabled in the DF video, then FSR is indeed missing information where transparency masks are involved in motion.
 
Last edited:
To add to my previous example, here are the 3440x1440 Kingshunt images taken by the redditor who discovered the DoF bug.

Bare in mind that this resolution is even below 4K target so with a 4K target FSR would look a little better again as it would be using a higher base resolution.

In both of these images DoF has been disabled. Both are 3440x1440 21:9

FSR Ultra Quality - 158FPS (same average)
tYfMja1.jpg


TAAU 77% Render Scale - 159FPS (same average)
SPJs8Xg.jpg


As you can see the FSR image is noticeably sharper than the TAAU image, including the wall and tree textures. The TAAU image is much blurrier, especially if you look at the tree foliage.

In addition due to TAAU being TAA based it suffers from standard TAA issues in movement such as ghosting which FSR does not.

So how could this TAAU be superior to FSR as many people, DF included are claiming when the above clearly shows the opposite in a best to best comparison? And this is not even the absolute best case scenario for FSR as these images are below 4K resolution, but they are high enough above 1440p that FSR Ultra works well here.

Lets look at a zoom comparison of the images:

FSR Character:
BjE6ly6.png


TAAU Character:
gShiKha.png


FSR TREE:
eIasMl8.png

TAAU TREE:
xmX2l4V.png

FSR WALL:
0qMWzpK.png

TAAU WALL:
wgrcAPd.png
I concur that the filter kernel is too aggressive with TAAU with these images.

I'm going to watch the video at this frame to see what they are comparing.

For the record, I never liked TAA because it introduces a high aggressive kernel that removes detail from the normal maps (which are extremely accurate in height values). With that kind of precision, just a small low-pass filter will remove a lot of the values at the upper end of the curve. I run my games with no AA at all. I can even tell that DLSS 2.0 average/performance modes blurs the image noticeably than no AA. DLSS 2.0 in quality mode though is sufficient enough to run with little change to the rendered image.
 
Last edited:
I concur that the filter kernel is too aggressive with TAAU.

I'm going to watch the video at this frame to see what they are comparing.

For the record, I never liked TAA because it introduces a high aggressive kernel that removes detail from the normal maps (which are extremely accurate in height values). With that kind of precision, just a small low-pass filter will remove a lot of the values at the upper end of the curve. I run my games with no AA at all. I can even tell that DLSS 2.0 average/performance modes blurs the image noticeably than no AA. DLSS 2.0 in quality mode though is sufficient enough to run with little change to the rendered image.
Just to clarify, the images I included above are not from DF's review. These images are from the redditor who discovered that DF's comparison had DoF enabled. He figured out how to disable it and then ran his own tests at Ultrawide 3440x1440.

Other differences between these shots and DF's review:
1. DF used sharpening as far as I know on native and TAAU (Or maybe this was only in Godfall?)
2. DoF was enabled on native and FSR but not on TAAU due to a bug. (This is the biggest difference)
3. Target resolution in DF's video was 4K, target res in the screenshots I posted was 3440x1440
4. DF is using FSR performance mode which is the lowest quality setting to upscale from 1080p->4k target. The screenshots I showed are using FSR Ultra Quality which is the best IQ mode.
5. DF's review had TAAU at a lower resolution scale to reach a 1080p->4K target. I'm assuming 50% render scale to get a 1080p->4K image? The screenshots I showed are using 77% render scale for TAAU.

If you watch DFs review or look at their article and see the comparisons for Kingshunt you will see much worse results for FSR than my screenshots above show, this was one of the major issues people had with their review and we can see with independent testing that FSR is in fact superior to Unreal's TAAU at least in the hacked in version forced in this game.

I haven't seen anyone test the same with Godfall yet but I'm sure if either game is on gamepass people can test them at home.
 
Last edited:
No idea what's going on, or how the AMD thing is open source and works with Geforce cards? But with my amateur eye looking at @IrishWhiskey post about the FSR looks better than the TAAU ¯\_(ツ)_/¯
 
  • Cheers
Reactions: IrishWhiskey
Do you happen to know what timeframe in the video you are speaking about the tree trunk?

UPDATE: Found the timeline.

I agree with you guys about the textures BUT the transparency textures on the veil is another matter. Are you saying that DOF was enabled on the AMD and not on the TAAU comparison? If they are both disabled in the DF video, then FSR is indeed missing information where transparency masks are involved in motion.
Yes, that is exactly what was happening due to a bug in the render pipeline when they forced TAAU into these games. While they are UE4 titles, they are not designed with the TAAU feature/plug in in mind. DoF was forced off in TAAU due to this bug in the render pipeline caused by forcing this TAAU mod into these games.

To me this completely invalidates the DF comparison. They also doubled down after the Kingshunt DoF issues were pointed out to them and did the same thing with Godfall. It had the same DoF issue and it again had to be pointed out to them on twitter that DoF was still active again on FSR but not TAAU. Again how could they miss this shit twice?

Not to mention the fact that on top of everything else with sharpening being turned on for TAAU and Native and the DoF issues, they also went out of their way to show FSR in its worst case scenario at performance mode which everyone agrees is kind of shit and not something people would really want to use in the real world.

That would have been fine if they also showed Ultra Quality vs TAAU but they purposely ignored how people would actually use it and dodged that comparison, knowing that FSR gives much better results at the highest quality setting/higher input resolution. They then used this in conjunction with the other nonsense to claim that TAAU was superior, which my screenshots prove it is clearly not.

They also didn't mention anything about the ghosting issues caused by TAAU which are not present with FSR, you would think that kind of thing would be relevant for an IQ comparison of technologies such as these.
 
Last edited:
So FSR happens after the AA phase, meaning a game is free to use whatever AA solution it wants including TAA.

FSR has been tested on multiple tech channels vs simple render scale and render scale+sharpening and FSR wins out handily.

I don't really watch much of DF content but by all accounts they are normally quite good, knowledgeable and thorough but in this case as the evidence mounts it has become clear they are showing their ass with their FSR take.

I'm assuming it was likely because their video was rushed, but for a channel that is all about pixel counting, fine details and graphics techniques to not notice that DoF was turned on on FSR and off with TAAU is pretty damning, especially given that it had to be pointed out by random redditors. The DoF change is the main difference between their comparison of TAAU and FSR which makes FSR appear more blurry and which was the main evidence used by them to claim TAA/TAAU based solutions were superior.

So how did this DoF bug/mistake happen? Well both Godfall and Kingshunt are UE4 games, TAAU was added in to the engine and can be forced via command line arguments/config changes. So the games render pipelines were not designed to take into account TAAU, hence there are bugs with DoF where TAAU diasbles it, showing a much cleaner image compared to FSR or the "simple native upscale".

In addition the DoF effect seems to be kind of bugged anyway as the character models in focus shouldn't be getting affected by the effect but anyway, not really the point I'm making.

Once DoF is disabled you can see that the FSR image looks way better than it did before and that FSR and TAAU are both quite close with each having different pros/cons in the image and how they handle certain elements of it.

Speaking of TAAU for a second (which is Unreal Engines new temporal upscaling method/plugin) it suffers in motion from artifacts and ghosting, such as any TAA based solution will as such FSR doesn't introduce these artifacts and would (in theory) look better in motion compared to TAAU. Granted I'm not an expert on either tech, just explaining my understanding.

I'm concentrating a lot on the DF review here but that is because this is where most of the misconceptions and negativity regarding FSR come from in the tech community at the moment. So for now lets continue talking about DF's take.

They show 3 side by side comparisons, the first one is some undefined "simple upscale". Nobody knows what this means or how it has been achieved. Do they mean they took a still image and applied some upscale/sharpening filters in something like photoshop/after effects or a similar program? If so then it is not really an apt comparison as things like FSR/TAAU/DLSS all work in real time in a game and whatever this is would not.

So if it is not post processing on a single still image....are they using some kind of render scale? Are they using sharpening? Nobody knows as they have been completely opaque about it.

Just to be clear, I'm not angry about this or having a go at you or anything, just stating some inconsistencies and issues I have with DF's methodology in this case.

So back to FSR vs TAAU, they both seem to be great options so I'm very happy we are getting more options and not locked in to one hardware vendor or another. However in DF's comparison (even if we remove the terrible DoF effect which is 95% of what makes the FSR image look worse) they show the worst case scenario for FSR which is a 1080p->4K upscale using the performance mode.

This is fine from an academic standpoint vs TAAU 1080p->4K, however they mention the performance (FPS) are essentially the same but don't show any frame graphs or metrics, which seems pretty odd given that is there whole schtik, I'm assuming it was likely due to lack of time rather than any kind of malice but it does seem odd that they seemed very reluctant to show any FPS charts/metrics with any of the FSR modes (not specifically with relation to vs TAAU).

So in other words we don't actually know if the performance in FPS is close/the same with those two modes compared, and some other people running independent tests to recreate this are getting different results so I'm not sure what the deal was with this from DF, but again I put it down to lack of time most likely.

EDIT: Actually upon further inspection it looks like Alex from DF locked the games to 60FPS during his FSR vs TAAU test, then looked at GPU utilization % and used that as a "performance" comparison to say that "perfomance is essentially identical", rather than....you know actual FPS performance. Worse than I thought...

So back to resolutions, we know FSR works best with higher input data/resolutions and we know nobody recommends using performance mode, so in a realistic scenario they should have compared best to best (Ultra Quality FSR vs best case TAAU) to get a fuller picture as well as the lower res upscale comparison. I'm happy they did the lower resolution input comparison as it does show some of the weaknesses of FSR, but they should have also balanced that with higher resolution comparisons to show how people would actually use both in games and for a more well rounded overview.

I mean if graphics techniques and comparisons are your whole deal and purpose of your channel then it presents some pretty huge oversights and problems with methodology all of things I've mentioned above. Again I'm assuming it is likely due to the time constraints for most of it, but some of these are pretty inexcusable oversights, especially when you have the reach/audience of DF and people take their word on tech as gospel.

Anyway, sorry if this seemed like a sperg out, I was just reading a ton about the issues with DF's video today and lots of people have pointed out those issues, which makes sense as to why almost the entire rest of the tech reviewer community seems to have an opposite take to DF here. I don't generally have an issue with DF and I don't think they are fanboys or paid shills or whatever, but they had a bad take here and for the sake of trying get an objective view on these new technologies they should be called out on where they might have messed up to improve in future.
I can't get in the weeds on this. I just don't have the education, outside of enjoying dipping into the technical part of things and learning at my own pace.

But I will say after years of following them, this is the one journalistic outfit I would trust not to get it wrong, to account for everything, and to promptly apologize if they did. If it's their word versus the tech-tubers, which I don't think much of, I pick their word. We'll see if that makes me wrong or not, but they've earned that sheer level of faith in them.
 
I can't get in the weeds on this. I just don't have the education, outside of enjoying dipping into the technical part of things and learning at my own pace.

But I will say after years of following them, this is the one journalistic outfit I would trust not to get it wrong, to account for everything, and to promptly apologize if they did. If it's their word versus the tech-tubers, which I don't think much of, I pick their word. We'll see if that makes me wrong or not, but they've earned that sheer level of faith in them.
I've seen Leadbetter apologize. I've seen a few others do it also. I've never seen Alex do it.
 
I can't get in the weeds on this. I just don't have the education, outside of enjoying dipping into the technical part of things and learning at my own pace.

But I will say after years of following them, this is the one journalistic outfit I would trust not to get it wrong, to account for everything, and to promptly apologize if they did. If it's their word versus the tech-tubers, which I don't think much of, I pick their word. We'll see if that makes me wrong or not, but they've earned that sheer level of faith in them.
That's a fair enough take, afterall they are by all accounts normally pretty good with this stuff. As I said I don't have any axe to grind against DF in general.

My only issue is with their flawed methodology for this particular review and the conclusion drawn from a basis of that flawed methodology. When almost the rest of the entire tech press disagrees you know something is up or should at least raise an eyebrow.

Worse still is that their audience/reach is so wide that there are tons of people that take DF's findings, reviews or opinions as gospel without the time, energy or knowledge to be able to investigate independently or be able to recognize flaws in the methodology.

Regarding owning up to mistakes, to be fair to Alex he did own up to the Depth of Field issue when it was pointed out to him on reddit and updated his article. However he didn't change his final conclusion or even really acknowledge the huge difference that the effect has on final image clarity/quality.



He swapped from Kingshunt to Godfall for his comparison to try to rectify the DoF issue but again he didn't notice that DoF was in fact still enabled on his Godfall comparisons and he had to be called out again. Everybody makes mistakes, we are all only human afterall but for a channel whose claim to fame is knowledge of graphical techniques and having an eye for minute detail, with 400% zooms etc... it seems kind of inexcusable to miss such an obvious effect that was clearly noticeable and able to be detected without any zooms and was successfully noticed in both cases by random Reddit/Twitter users. I would perhaps reconsider the blind faith a little just based on that in and of itself.

Anyway, mistakes aside it sounds like Alex made up his mind about FSR before actually reviewing it based on his comments prior to their review and so in a lot of people's opinion it seems like he did his best to work backwards to prove that conclusion. While that is kind of a weird thing to do if people's suspicions are correct then that would have affected all of the methodology they used during their review.

I just think he doesn't like it because it is "old" in a tech sense and not as cutting edge as using temporal data or AI/machine learning. The DF boys do seem to love new technology and new graphical effects/solutions so I think that maybe coloured their coverage in this case to some degree.

I mean anyone is free to disagree and/or form their own opinions of FSR in general seeing as there are tons of reviews and data out there from the tech press and just everyday people making their own comparisons. Similarly if people still want to agree with DF's conclusions overall about this then that is fine too. However people should at least acknowledge that their review had some pretty large flaws what with the depth of field effect for example.
 
Last edited:
  • Brain
  • 100%
Reactions: Kadayi and Ascend
Anyone know if there's a site that lists games which have been updated to support FSR? Keen to give my humble old RX580 a shot at this new-fangled tech.
 
  • Like
Reactions: IrishWhiskey
Here's DF discussing the tech, implementation, use cases and future of FSR. I found it to be very interesting.


Thanks for posting this, was an interesting discussion.

They seemed a lot fairer to FSR here than in their review and the bald guy...Rich I think? Seemed to be the most objective about it.

You could see that a lot of people's suspicions were correct that due to FSR being "old" tech in their view that they have a much lower opinion of it compared to TAA based solutions/DLSS. It kind of makes sense as they love technology and want things to be at the cutting edge when it comes to techniques etc... so new > old would be a simple way to look at their leaning, which is fine for a tech channel.

I did like the complete dodge they did when asked if FSR was superior to DLSS1.0, which basically confirms what everyone else is saying, that yes it is in fact superior to it. Granted we don't have an apples to apples comparison where both are implemented in the same game, but seemed weird that they were so cagey on giving a direct answer.

It was a little eye rolling when they mentioned lack of inner edge details such as textures etc.. when compared to TAAU while then using footage as "proof" where Depth of Field was turned off for TAAU but not for FSR/Native. They also didn't mention that sharpening/CAS was turned on for both their "simple upscale" and the TAAU image.

Alex also had a pretty funny moment when responding to people complaining about him comparing lower resolution inputs of FSR vs TAAU. He literally admits that as you go up in input resolution that the two techniques start to become very hard to distinguish, in other words FSR improves quite substantially with higher input resolution/quality presets. He then mentioned that this somehow made TAAU better for "reasons", in other words it was "newer" tech and regardless of what your eyes show you it simply has to be better.

I don't necessarily disagree with his sub point about distance fences and things of that nature and the fact that more data in to work with in general is better than less, however in their review and in this discussion they didn't really go into a proper comparison of pros/cons for both techniques such as edges and closer objects to view looking better with FSR while further objects in the distance and micro detail structures like wire fences etc.. would look better with TAAU etc.. anyway not going to keep harping on that stuff.

I did like the point they made about Series S for example not being a great use case for FSR due to FSR working much better at higher resolutions rather than lower input resolutions. That is a very good point and should not be overlooked. Of course if the game was not using a TAA based solution or a bad one and was just upsampling from a lower resolution then FSR would probably be a slight improvement there but not by much as it does not perform well at lower input resolutions.

I also liked that they mentioned the elephant in the room when it comes to upsampling techniques and Ray Tracing effects, namely that if you are upscaling from lets say 1080p or 1440p to 4K for example, you are only getting the RT calculations/quality of a lower resolution so reflections etc.. will look noticeably worse/lower res/blurry when compared to native 4K rendering for example. And they even reluctantly admit that this is a problem for DLSS too, which was nice to see someone acknowledge this as I don't believe any major outlets have brought this up regarding DLSS/other upsampling techniques in the past.

Overall though it was an interesting discussion and it will be interesting to see where AMD takes things with further revisions of FSR. I've heard various random rumours such as a better version with temporal components will release, to a version that makes use of AI to improve the image will release close to potentially RDNA3 which could have AI cores/silicon to further accellerate it and various in between things. Hopefully they continue to improve it and adoption continues to grow.

Looks like a game people actually want to play will have FSR soon in Resident Evil 8 so that will be nice to see and have people test, especially with the RT effects. Far Cry 6 should also have it probably at launch which should also be a cool showcase. FSR will likely live or die based on how quickly it can be adopted and integrated into game engines, so far that appears to be happening pretty quickly but it will take some time to see if it has any real staying power. Will be interesting to see what the support landscape looks like in 6 months to a year to properly gauge how it is going.

On the engine front good news seems to be that Unity are one of the supported developers so there will probably be a simple plug in that indies that use to plug and play with FSR. AMD seems to have a very close relationship with Epic for Unreal Engine 5, which is a complete reversal of where they were in the past with UE4 heavily favouring Nvidia, but oddly Epic Games were absent from the list of developers currently supporting FSR. Would be nice if AMD/Epic could get a plug in working for Unreal Engine 5 to make use of FSR, but then again maybe with Epic pushing their new TSR that it would make other solutions redundant? If AMD are smart they will leverage their current close relationship to get engine support in the form of an easy plugin.

Then again maybe FSR in its current form is so easy to implement that having a plug in would be redundant/require more work than native support? Either way will be interesting to see how this all develops. The more options we get and the wider the adoption of open standards the better in my book.
 
  • 100%
Reactions: Ascend
"Implementing DLSS was quite complex to integrate into Unity for a small studio like us; it required tweaking the engine and creating an external plugin to bridge Unity and DLSS. It was complicated, but in the end, it gave amazing results. FSR, on the other hand, was very easy to implement, it only took me a few hours, requiring only simple data. However, it did require the source data to be as sharp as possible since FSR is not a deep learning method and cannot reconstruct loss details. We needed to replace the built-in Unity TAA with the AMD Cauldron TAA (opensource) that we ported to Unity as it is a lot better at preserving details."

"Quality-wise, when there is a lot of pixel info available for upscaling to 4K, both technologies give amazing results, and I have a hard time seeing differences between them. I'd even say that I slightly prefer the FSR for 4K resolution since it doesn't introduce any artifacts/minor blurriness that DLSS can sometimes introduce. For lower resolutions like upscaling to 1080p or 720p, I think DLSS gives a better result since it can reconstruct parts of missing details due to the nature of the technique."

 
  • Brain
Reactions: regawdless