AMD FSR ,Extended Life for your NVidia 10 series GPU

HUB's Tim: "...no mention of AI/temporal, although I would be surprised if there wasn't a temporal element that used information for multiple frames." Time will tell.
That particular image has FSR at Quality Mode rather than Ultra Quality though, if that matters. Edit: ah, this was already brought up, my bad.





So this means our bois are going to keep selling used 10 series cards at high prices? Shhh, keep this a secret.
 
Last edited:
HUB's Tim: "...no mention of AI/temporal, although I would be surprised if there wasn't a temporal element that used information for multiple frames." Time will tell.
That particular image has FSR at Quality Mode rather than Ultra Quality though, if that matters. Edit: ah, this was already brought up, my bad.





So this means our bois are going to keep selling used 10 series cards at high prices? Shhh, keep this a secret.
The RTX cards have dedicated Tensor cores for DLSS. Without additional silicon on older cards, it is nearly impossible to implement real-time machine learning without significant processing on existing cores. This simply seems to be a "one size fits all" algorithmic image processing. If they cannot add accurate information to the image when upscaling which DLSS 2.0 does, it is unlikely it will match native rendering, especially when you have different types of scenes across different types of games.
 
GPUs are anything but my forte. You think it won't match DLSS 1? Many (hopefuls maybe) are expecting it to be close to DLSS 1.5 or just good enough.

Granted there's no hardware acceleration and AMD aren't claiming it to be anything beyond spatial upscaling just yet, some think there could be ML in it, but I don't have anything to back this so won't put this forward. Probably the hardware part will come with RDNA 3, unless there's already hardware that can partially support these computations in RDNA/2 (read somewhere, not sure).
Strictly speaking dedicated hardware isn't needed really needed to leverage temporal data, is it? There was something about TSR achieving this on reddit - link

If it indeed falls short of DLSS 1 then it'd be a disaster almost, given the hype they created. If it's decent enough, then wider adoption is a good thing for the market and being open source it will be improved upon. Maybe the consoles will also benefit from this since they are all AMD.
It's unlikely that Nvidia is going to support FSR right away so all those 10 series cards aren't going to get this now. But if FSR works out then NV will be forced to introduce something like DLSS Compatible™ down the road.
 
Bumping for tomorrow.
The games currently said to support FSR aren't that great but more interested in seeing how it performs. And Nvidia just announced DLSS support for a bunch of games, including Doom and RDR2
 

@princeoo7 , we ****in need your feedbacks on this as well as other amd card owners
I don't have any of those cames. will download the riftebreaker and try it out.
But in terms of downscaling the resolution which I can do on sapphire Trixx, it works well. not that much of difference in quality loss for those who don't get into the monitor itself.
ok here we go, download the zip file and check the images by your self. each image is of around 14.7mb so can't upload here. so decided to make a zip file and upload here. All settings where on like RT, RT Ambein, etc.

You can feel the difference definately with low version. but ultra Quality and Quality settings are free performance upgrade :P rest you own jugdment.

total file size is 70 mb or so.

 
Last edited:
I wished I had godfall :P would have really made a difference but that game on keyboard is a fun killer for me. controller mangta ;)
 
FSR looks very good in these games at 1440p & 4k in ultra quality & quality mode. In other modes or at 1080p, the differences are visible but it's something which can be easily solved by increasing your viewing distance by a couple of feet. And this is absolutely fine considering such performance boosts are usually only required above 1080p so they are actually serving a real use case here. The most impressive aspect FSR is that it is driver & hardware independent & is actually part of the game so it will be supported on all hardware irrespective of the brand. What remains to be seen is the game support in the future as well as if this has scope for improvement with future iterations. I am guessing since this is not based upon any AI or machine learning algorithm, the scope of improvement down the line might be limited. But the fact that FSR is supported across both brands & is supported on console may result in it having excellent adoption across games.

Question is where does this leave DLSS? In my opinion at 4k & 1440p quality & ultra quality mode both DLSS & FSR are going to look very comparable with DLSS having an obvious advantage, but I feel that would be mostly academic & practically nobody will actually care about that. At 1080p & lower quality options however, DLSS will continue to have an advantage due to machine learning. Basically, FSR gets limited by the input resolution when it is not able to resolve details which are lost at these low resolutions. It seems to be good at reconstructing details which are visible but obviously fails to bring out details which never showed up at the low input resolutions. This issue I think will never be resolved in the current implementation of FSR. DLSS however is able to resolve those details relatively well due to the machine learned data that it already has.

It remains to be seen if this DLSS advantage is actually worth investing into a locked technology or if Nvidia is going to be pushed into innovating even more due to FSR. All in all it is a big win for all of us & hopefully will allow developers to spend graphics resources in improving the visual quality rather than increasing resolution.

Edit:
DFs review is up, this will be interesting.
 
Last edited:
Oh boy, DFs Alex Battaglia has made me rethink of my observation mentioned above. Do watch that video as the video zooms in on some comparison shots to make the differences more obvious, specially for those of us who are comparing this on a 1080p screens. Basically the image downgrade at 4k ultra quality isn't negligible & I think would be visible if we were actually running the footage on a 4k 32inch or greater monitor. At any lower resolution or other quality modes, the image degradation is significant. There are some shimmering artifacts in motions as well but overall so far I was still convinced that these issues would in reality be of very little consequence specially while actually playing the game at high frame rates.

But than he compares the 4k FSR performance footage, basically internal 1080 to 4k, to a simple 1080p to 4k upscaling as well as 1080p to 4k TAA super sampling provided by unreal engine & the TAA upscaled image actually has some significant improvement over FSR in multiple areas. So FSR is an improvement over basic up sampling but not much better or in some areas worse than an existing TAA up sampler. This has me worried it a bit right now specially considering that UE5 TSR is already a major improvement over TAA up sampling. Basically, FSR seems to be doing good reconstruction of edges, but doesn't seem to do much to improve the details that are getting filled inside these edges. Let's see how it fares in future titles where a combination of TAA & FSR might be utilized for an overall better picture quality or if the UE5 TSR will end up becoming a better overall solution with plugin support.
 
Last edited:
now do you feel the title was catchy and correct as well
Catchy, sure. Correct? Not entirely. Pretty sure there are way more people owning non-10 series cards than 10 series owners.
And moreover, FSR doesn't work on just 10 series cards or older GPUs, but basically all GPUs - even those not specifically optimised for FSR since it's implemented at the game level, not driver level. So the title, although not incorrect, is still quite the misrepresentation just for the sake of catchiness. I thought I was on a forum, not browsing for click bait articles.
 
Last edited:
Overall, it's good for the industry. Free performance when needed.


Edit: Looks like Ultra at 1440p is good enough. Maybe you guys check in that demo game.
 
Last edited:
@DanteErodov Did you check Kitguru's review? I read they did a better job than DF at TAAU vs FSR and found FSR to be better (?) I didn't watch their video yet.
Not sure if there's any truth to this, but DF tends to be biased towards NV and not the best source for AMD content some say.
 
@DanteErodov Did you check Kitguru's review? I read they did a better job than DF at TAAU vs FSR and found FSR to be better (?) I didn't watch their video yet.
Not sure if there's any truth to this, but DF tends to be biased towards NV and not the best source for AMD content some say.
I will check that out! While I think DF might be a bit biased as they only focused on the negatives, I feel other reviewers are also a bit less critical compared to how they reviewed DLSS 1.0. This could be due to DLSS being a proprietary tech & FSR being open or maybe DLSS was first to the market so got a more critical reception. The worrying aspect for me is not the image quality downgrade itself, it is minor & can be looked past, but that it might be similar or slightly better than TAA. With TSR around the corner & already available for testing in UE5, I wonder how FSR will fare against that. I mean, AMD should be more worried about these techs rather than DLSS if it wants mass adoption.

Anyways, I'll try out the The Riftbreaker demo as you suggested & will see for myself what all the fuss is about. Will also try to post screenshots of comparison if I end up with a different observation.
 
Can someone dumb this down for me? ELI5 please? I have a GTX 1060 3GB and play mainly just CSGO. Will I see fps improvements?
 
Overall, it's good for the industry. Free performance when needed.


Edit: Looks like Ultra at 1440p is good enough. Maybe you guys check in that demo game.
for me Ultra quality and Quality setting worked well. below that was no fun but fps boost was significant. tested this on my 32 " 2k 75hz monitor. Riftebreaker didn't allowed me to select me 27" 2k 144hz monitor as primary so what was an bummer. also I set very close to my 32" monitor so my saying is based on that only. the extract distance between me and my 3 and 1/2 iphone XR in height. don't have a ruler sitting around for quick measurement :P
 
Back
Top